Professor Mark Gilthorpe

Background Compositional data comprise the parts of some whole, for which all parts sum to that whole. They are prevalent in many epidemiological contexts. Although many of the challenges associated with analysing compositional data have been discussed previously, we do so within a formal causal framework by utilizing directed acyclic graphs (DAGs). Methods We depict compositional data using DAGs and identify two distinct effect estimands in the generic case: (i) the total effect, and (ii) the relative effect. We consider each in the context of three specific example scenarios involving compositional data: (1) the relationship between the economically active population and area-level gross domestic product; (2) the relationship between fat consumption and body weight; and (3) the relationship between time spent sedentary and body weight. For each, we consider the distinct interpretation of each effect, and the resulting implications for related analyses. Results For scenarios (1) and (2), both the total and relative effects may be identifiable and causally meaningful, depending upon the specific question of interest. For scenario (3), only the relative effect is identifiable. In all scenarios, the relative effect represents a joint effect, and thus requires careful interpretation. Conclusions DAGs are useful for considering causal effects for compositional data. In all analyses involving compositional data, researchers should explicitly consider and declare which causal effect is sought and how it should be interpreted.

Machine learning methods, combined with large electronic health databases, could enable a personalised approach to medicine through improved diagnosis and prediction of individual responses to therapies. If successful, this strategy would represent a revolution in clinical research and practice. However, although the vision of individually tailored medicine is alluring, there is a need to distinguish genuine potential from hype. We argue that the goal of personalised medical care faces serious challenges, many of which cannot be addressed through algorithmic complexity, and call for collaboration between traditional methodologists and experts in medical machine learning to avoid extensive research waste.

Prediction and causal explanation are fundamentally distinct tasks of data analysis. In health applications, this difference can be understood in terms of the difference between prognosis (prediction) and prevention/treatment (causal explanation). Nevertheless, these two concepts are often conflated in practice. We use the framework of generalized linear models (GLMs) to illustrate that predictive and causal queries require distinct processes for their application and subsequent interpretation of results. In particular, we identify five primary ways in which GLMs for prediction differ from GLMs for causal inference: (i) the covariates that should be considered for inclusion in (and possibly exclusion from) the model; (ii) how a suitable set of covariates to include in the model is determined; (iii) which covariates are ultimately selected and what functional form (i.e. parameterization) they take; (iv) how the model is evaluated; and (v) how the model is interpreted. We outline some of the potential consequences of failing to acknowledge and respect these differences, and additionally consider the implications for machine learning (ML) methods. We then conclude with three recommendations that we hope will help ensure that both prediction and causal modelling are used appropriately and to greatest effect in health research.

We commend Nickerson and Brown on their insightful exposition of the mathematical algebra behind Simpson’s paradox, suppression and Lord’s paradox; we also acknowledge there can be differences in how Lord’s paradox is approached analytically, compared to Simpson’s paradox and suppression, though not in every example of Lord’s paradox. Furthermore, Simpson’s paradox, suppression and Lord’s paradox ask the same contextual questions, seeking to understand if statistical adjustment is valid and meaningful, identifying which analytical option is correct. In our exposition of this, we focus on the perspective of context, which must invoke causal thinking. From a causal thinking perspective, Simpson’s paradox, suppression and Lord’s paradox present very similar analytical challenges.

Background: Despite early childhood weight gain being a key indicator of obesity risk, we do not have a good understanding of the different patterns that exist. Objectives: To identify and characterise distinct groups of children displaying similar early-life weight trajectories. Methods: A growth mixture model captured heterogeneity in weight trajectories between 0 and 60 months in 1390 children in the Avon Longitudinal Study of Parents and Children. Differences between the classes in characteristics and body size/composition at 9 years were investigated. Results: The best model had five classes. The “Normal” (45%) and “Normal after initial catch-down” (24%) classes were close to the 50th centile of a growth standard between 24 and 60 months. The “High-decreasing” (21%) and “Stable-high” (7%) classes peaked at the ~91st centile at 12-18 months, but while the former declined to the ~75th centile and comprised constitutionally big children, the latter did not. The “Rapidly increasing” (3%) class gained weight from below the 50th centile at 4 months to above the 91st centile at 60 months. By 9 years, their mean body mass index (BMI) placed them at the 98th centile. This class was characterised by the highest maternal BMI; highest parity; highest levels of gestational hypertension and diabetes; and the lowest socio-economic position. At 9 years, the “Rapidly increasing” class was estimated to have 68.2% (95% confidence interval [CI] 48.3, 88.1) more fat mass than the “Normal” class, but only 14.0% (95% CI 9.1, 18.9) more lean mass. Conclusions: Criteria used in growth monitoring practice are unlikely to consistently distinguish between the different patterns of weight gain reported here.

The present study aimed to compare the predictive acuity of latent class regression (LCR) modelling with: standard generalised linear modelling (GLM); and GLMs that include the membership of subgroups/classes (identified through prior latent class analysis; LCA) as alternative or additional candidate predictors. Using real world demographic and clinical data from 1,802 heart failure patients enrolled in the UK-HEART2 cohort, the study found that univariable GLMs using LCA-generated subgroup/class membership as the sole candidate predictor of survival were inferior to standard multivariable GLMs using the same four covariates as those used in the LCA. The inclusion of the LCA subgroup/class membership together with these four covariates as candidate predictors in a multivariable GLM showed no improvement in predictive acuity. In contrast, LCR modelling resulted in a 18–22% improvement in predictive acuity and provided a range of alternative models from which it would be possible to balance predictive acuity against entropy to select models that were optimally suited to improve the efficient allocation of clinical resources to address the differential risk of the outcome (in this instance, survival). These findings provide proof-of-principle that LCR modelling can improve the predictive acuity of GLMs and enhance the clinical utility of their predictions. These improvements warrant further attention and exploration, including the use of alternative techniques (including machine learning algorithms) that are also capable of generating latent class structure while determining outcome predictions, particularly for use with large and routinely collected clinical datasets, and with binary, count and continuous variables.

Four modelling approaches are commonly used to adjust for overall energy intake when seeking to estimate the causal effect of an individual dietary component on an outcome; (1) the ‘standard model’ adjusts for total energy intake, (2) the ‘energy partition model’ adjusts for remaining energy intake, (3) the ‘nutrient density model’ examines the exposure as a proportion of total energy, and (4) the ‘residual model’ indirectly adjusts for total energy by using the residual from regressing the exposure nutrient on total energy intake. Unfortunately, it remains underappreciated that each approach evaluates a different causal effect estimand and only partially accounts for confounding by common causes of dietary intake and composition. Semi-parametric directed acyclic graphs and Monte Carlo simulations were used to identify the estimand implied by each approach and the correct interpretation of the model results. The performance of each model for estimating the corresponding target estimand was explored both in the absence and presence of confounding that acts through diet. An alternative approach based on the energy partition model that simultaneously adjusts for all competing dietary components, termed the ‘all-components model’, was also explored and compared with the four traditional approaches. This model involves using the weighted coefficients of different dietary components to estimate any desired causal effect estimand. The ‘standard model’ and the mathematically identical ‘residual model’ both estimate the average relative causal effect (i.e. a ‘substitution’ effect) but provide biased estimates even in the absence of any confounding. The ‘energy partition model’, that adjusts for remaining energy intake, estimates the total causal effect (i.e. an ‘additive’ effect) but only provides unbiased estimates in the absence of confounding or when all individual nutrients have equal effects on the outcome. The ‘nutrient density model’ does not target a causally meaningful estimand but can provide extremely biased estimates of the average relative causal effect of the exposure rescaled as a percentage of total energy intake. Accurate estimates of both the total and average relative causal effects were obtained with the ‘all-components model’. Only the ‘all-components model’ produces unbiased estimates of different causal effects. Lack of awareness of the estimand differences and accuracy of the different modelling approaches may explain some of the apparent heterogeneity among existing nutritional studies. Serious questions may be raised regarding the validity of meta-analyses where different strategies returning different estimands have been inappropriately pooled.

Dietary guidelines often recommend substituting certain nutrients or foods with healthier alternatives, based on the available evidence from nutritional epidemiology. The effects of food substitutions can be examined by conducting isocaloric dietary interventions, but experimental studies are often not practical or sufficiently generalisable. Therefore, nutritional epidemiology is highly reliant on observational data, in which food substitutions can be explored using mathematical modelling. The two modelling approaches commonly used for estimating substitution effects are known as (1) the ‘leave-one-out’ model, in which total energy intake and all dietary components are included as covariates, excluding the nutrient(s) that the exposure should be substituted with; and (2) the energy partition model, in which all dietary components are included as covariates, without further adjustment for total energy intake. It remains underappreciated that these approaches do not perform equally well for estimating substitution effects, and that there is limited evidence on whether they produce unbiased estimates. Semi-parametric directed acyclic graphs and Monte Carlo data simulations were used to explore the performance of the two approaches for estimating the following estimands: 1) the average relative causal effect (i.e. the joint effect of increasing intake of the exposure and decreasing the intake of all other nutrients, while keeping total energy intake constant), 2) the relative effect of increasing the exposure nutrient and decreasing the intake of one other nutrient, and 3) the relative effect of increasing the exposure nutrient and decreasing the intake of a combination of other nutrients. The approaches were explored both in the absence and presence of confounding that acts through diet. The ‘leave-one-out’ model produced a biased estimate of the average relative causal effect even in the absence of any confounding. It robustly estimated substituting the exposure with another specific nutrient regardless of whether confounding was present but produced biased estimates of substituting the exposure for a combination of other nutrients even in the absence of confounding. The energy partition model robustly estimated all three estimands of interest, producing unbiased estimates regardless of whether confounding was present or not. Only the energy partition model produces unbiased estimates of different substitution effects in the context of nutritional epidemiology. It performs equally well even in the presence of confounding that acts through diet. Substitution analyses using the ‘leave-one-out’ approach might not be robust and any existing studies using this model might suffer from bias.

Lifecourse trajectories of clinical or anthropological attributes are useful for identifying how our early-life experiences influence later-life morbidity and mortality. Researchers often use growth mixture models (GMMs) to estimate such phenomena. It is common to place constrains on the random part of the GMM to improve parsimony or to aid convergence, but this can lead to an autoregressive structure that distorts the nature of the mixtures and subsequent model interpretation. This is especially true if changes in the outcome within individuals are gradual compared with the magnitude of differences between individuals. This is not widely appreciated, nor is its impact well understood. Using repeat measures of body mass index (BMI) for 1528 US adolescents, we estimated GMMs that required variance-covariance constraints to attain convergence. We contrasted constrained models with and without an autocorrelation structure to assess the impact this had on the ideal number of latent classes, their size and composition. We also contrasted model options using simulations. When the GMM variance-covariance structure was constrained, a within-class autocorrelation structure emerged. When not modelled explicitly, this led to poorer model fit and models that differed substantially in the ideal number of latent classes, as well as class size and composition. Failure to carefully consider the random structure of data within a GMM framework may lead to erroneous model inferences, especially for outcomes with greater within-person than between-person homogeneity, such as BMI. It is crucial to reflect on the underlying data generation processes when building such models.

The analysis of time to treatment data and the evaluation of subsequent effects on health outcomes can be complex due to the nature of the data and the relationships amongst the variables. This study proposes an alternative method of analyzing such data using latent class analysis (LCA). The association between time to begin adjuvant chemotherapy after breast cancer surgery and survival was investigated using both "traditional" regression analysis and LCA. Women with breast cancer undergoing surgery and subsequent adjuvant chemotherapy in two English regions between January 01, 1998 and December 31, 2004 were identified from a linked cancer registry-Hospital Episode Statistics dataset (n = 10,366). Patient, tumor, and treatment information were extracted. A Cox proportional hazards model was used to analyze 5-year survival using regression analysis and LCA. Using "traditional" regression analysis, women beginning chemotherapy >10 weeks after surgery had worse survival in region 1 (HR = 1.49, 95% CI 1.13-1.95 compared to <3 weeks) but not region 2. LCA split the women into three groups representing short, medium, and long waits. The median time to begin chemotherapy in the "long" wait group was 70 (region 1) and 57 (region 2) days. In this group, increased time to begin chemotherapy was associated with worse survival (region 1 HR = 1.15, 95% CI 1.11-1.18; region 2 HR = 1.08, 95% CI 1.03-1.13 per week increase). LCA identified a group of 13-15% of women for whom a longer time to begin chemotherapy had an adverse effect on survival. This methodology provides an excellent framework in which to examine complex associations between the delivery of patient care and patient outcomes.

Objective: Acute heart failure syndrome (AHFS) is a major cause of hospitalisation and imparts a substantial burden on patients and healthcare systems. Tools to define risk of AHFS hospitalisation are lacking. Methods: A prospective cohort study (n=628) of patients with stable chronic heart failure (CHF) secondary to left ventricular systolic dysfunction was used to derive an AHFS prediction model which was then assessed in a prospectively recruited validation cohort (n=462). Results: Within the derivation cohort, 44 (7%) patients were hospitalised as a result of AHFS during 1 year of follow-up. Predictors of AHFS hospitalisation included furosemide equivalent dose, the presence of type 2 diabetes mellitus, AHFS hospitalisation within the previous year and pulmonary congestion on chest radiograph, all assessed at baseline. A multivariable model containing these four variables exhibited good calibration (Hosmer-Lemeshow p=0.38) and discrimination (C-statistic 0.77; 95% CI 0.71 to 0.84). Using a 2.5% risk cut-off for predicted AHFS, the model defined 38.5% of patients as low risk, with negative predictive value of 99.1%; this low risk cohort exhibited <1% excess all-cause mortality per annum when compared with contemporaneous actuarial data. Within the validation cohort, an identically applied model derived comparable performance parameters (C-statistic 0.81 (95% CI 0.74 to 0.87), Hosmer-Lemeshow p=0.15, negative predictive value 100%). Conclusions: A prospectively derived and validated model using simply obtained clinical data can identify patients with CHF at low risk of hospitalisation due to AHFS in the year following assessment. This may guide the design of future strategies allocating resources to the management of CHF.

Objective: Data-driven approaches to dietary patterns are under-utilized; latent class analyses (LCA) are particularly rare. The present study used an LCA to identify subgroups of people with similar dietary patterns, explore changes in dietary patterns over a 10-year period and relate these dynamics to sociodemographic factors and health outcomes. Design: The 1998 baseline and 2008 follow-up of the Cork and Kerry Diabetes and Heart Disease Study. Diets were assessed with a standard FFQ. LCA, under the assumption of conditional independence, was used to identify mutually exclusive subgroups with different dietary patterns, based on food group consumption. Setting: Republic of Ireland. Subjects: Men and women aged 50-69 years at baseline (n 923) and at 10-year follow-up (n 320). Results: Three dietary classes emerged: Western, Healthy and Low-Energy. Significant differences in demographic, lifestyle and health outcomes were associated with class membership. Between baseline and follow-up most people remained 'stable' in their dietary class. Most of those who changed class moved to the Healthy class. Higher education was associated with transition to a healthy diet; lower education was associated with stability in an unhealthy pattern. Transition to a healthy diet was associated with higher CVD risk factors at baseline: respondents were significantly more likely to be smokers, centrally obese and to have hypertension (non-significant). Conclusions: LCA is useful for exploring dietary patterns transitions. Understanding the predictors of longitudinal stability/transitions in dietary patterns will help target public health initiatives by identifying subgroups most/least likely to change and most/least likely to sustain a change.

Background Each year numerous studies evaluate longitudinal data within a lifecourse context with later-life health status (e.g. systolic blood pressure, SBP) analysed with respect to repeated measures of early-life experiences (e.g. body mass index, BMI), using standard multiple linear regression inappropriately. We examine these problems, via simulation, to give clear guidance on what happens if basic theory of causation (as informed by directed acyclic graphs; DAGs) is ignored. In our previous work on birth weight (BW) we have shown that for ages 0–17 years, our simulation results are consistent with the Avon Longitudinal Study of Parents and Children (ALSPAC) study with maximal attenuation when later life mediators are included. This study extends this analysis to age 70 years. Methods We simulated a lifecourse dataset comprising BW, repeated measures of BMI up to 70 years, and a single z-score SBP measure at age 70, using a multivariate normal distribution designed to represent real data, informed for ages 0–17 by the ALSPAC study. We conducted a series of multivariable regression analyses taking BW and BMI at age 70 years as exposures, similar to those frequently seen in lifecourse research. Results When we extended the maximum age to 70 years, the analyses with BW as the exposure showed that adjusting for any mediator results in attenuation away from the true association. This was most severe when later life measures were included. When BMI at 70 years was the exposure, optimal confounder adjustment was achieved when later life measures of BMI were included in the model. Further adjustment (e.g. for early adulthood BMI) yielded little additional effect, and need not be considered in the interests of parsimony. However, here the attenuation observed was positive, whereas the attenuation observed was negative with our previous analyses of a younger cohort. This disparity is due to much greater collinearity between BMI in later life than between BMI in early life (i.e. BMI 60 years and 70 years versus BMI 17 years vs. 12 years).

We recently reported that differential gene expression and DNA methylation profiles in blood leukocytes of apparently healthy smokers predicts with remarkable efficiency diseases and conditions known to be causally associated with smoking, suggesting that blood-based omic profiling of human populations may be useful for linking environmental exposures to potential health effects. Here we report on the sex-specific effects of tobacco smoking on transcriptomic and epigenetic features derived from genome-wide profiling in white blood cells, identifying 26 expression probes and 92 CpG sites, almost all of which are affected only in female smokers. Strikingly, these features relate to numerous genes with a key role in the pathogenesis of cardiovascular disease, especially thrombin signaling, including the thrombin receptors on platelets F2R (coagulation factor II (thrombin) receptor; PAR1) and GP5 (glycoprotein 5), as well as HMOX1 (haem oxygenase 1) and BCL2L1 (BCL2-like 1) which are involved in protection against oxidative stress and apoptosis, respectively. These results are in concordance with epidemiological evidence of higher female susceptibility to tobacco-induced cardiovascular disease and underline the potential of blood-based omic profiling in hazard and risk assessment.

Background Dose escalation of evidence-based chronic heart failure pharmacotherapy in real-life does not approach that achieved in clinical trials; it is unclear whether this impacts upon mortality. We aimed to quantify the association between temporal changes in β-adrenoceptor antagonist (β-blocker) and ACE inhibitor (ACEI) dose and mortality in patients with chronic heart failure.Methods Prospective observational study of 408 stable chronic heart failure patients with left ventricular systolic dysfunction, managed in a multidisciplinary outpatient clinic, with repeat visit for clinical assessment (mean 354 days after recruitment). The association of between- and within-patient temporal differences in dose of heart failure pharmacotherapies, to all-cause mortality was studied after accounting for collinearity and confounding (including within- and between-patient temporal differences in clinical status).Results During a mean follow-up period of 1060 days, 97 patients (21.6%) died. Between patient analyses revealed increasing dose of ACEI and β-blocker to be associated with reduced mortality, whilst increasing diuretic dose was associated with rising mortality, even after adjustment for confounders. Within patient analyses revealed that upward titration of β-blocker (but not ACEI) was associated with major reductions in mortality, even after accounting for confounders. Temporal changes in diuretic dose, haemodynamic status and renal function were not significantly associated with mortality.Conclusions Sustained between-individual differences in ACEI, β-blocker and diuretic dose are associated with mortality risk after accounting for likely confounders. However, within individuals only escalation of β-blocker dose is associated with improved prognosis.

Background: Obesity is a growing problem worldwide and can often result in a variety of negative health outcomes. In this study we aim to apply partial least squares (PLS) methodology to estimate the separate effects of age, period and cohort on the trends in obesity as measured by body mass index (BMI). Methods. Using PLS we will obtain gender specific linear effects of age, period and cohort on obesity. We also explore and model nonlinear relationships of BMI with age, period and cohort. We analysed the results from 7,796 men and 10,220 women collected through the SLAN (Surveys of Lifestyle, attitudes and Nutrition) in Ireland in the years 1998, 2002 and 2007. Results: PLS analysis revealed a positive period effect over the years. Additionally, men born later tended to have lower BMI (-0.026 kg·m-2 yr-1, 95% CI: -0.030 to -0.024) and older men had in general higher BMI (0.029 kg·m -2 yr-1, 95% CI: 0.026 to 0.033). Similarly for women, those born later had lower BMI (-0.025 kg·m-2 yr-1, 95% CI: -0.029 to -0.022) and older women in general had higher BMI (0.029 kg·m-2 yr-1, 95% CI: 0.025 to 0.033). Nonlinear analyses revealed that BMI has a substantial curvilinear relationship with age, though less so with birth cohort. Conclusion: We notice a generally positive age and period effect but a slightly negative cohort effect. Knowing this, we have a better understanding of the different risk groups which allows for effective public intervention measures to be designed and targeted for these specific population subgroups.

In the last decades, the prevalence of obesity has increased in the Taiwanese population. This has the potential to impact on the risks of cardiovascular diseases and diabetes. This study investigated trends in the changes in several indices of obesity in the last decade, and the relationship between blood pressure (BP) and these obesity indices available in Mei-Jaw Corporation health-screening data from 1996/1998 to 2006. Three cross-sectional surveys among healthy individuals ages 20-59 years, in which 14,362 subjects examined in year 1996, 17,368 in 1998, and 28,524 in 2006, were included in the analysis. Body weight and height data were available from 1996, whereas %body fat, waist circumference and waist-hip ratio (Whratio) were only available from 1998 onwards. We found that the association between systolic BP and body weight, body mass index, %body fat, Whratio and waist became stronger for both men and women in 2006 than 1996 after adjustment for age, education level, alcohol intake, smoking and betel nut chewing. In contrast, non-obese people seemed to have lower diastolic BP in 2006 than in 1996. This trend is consistent irrespective of the index of obesity used. Among healthy individuals, the average values for the obesity indices increased in men but remained similar in women. However, in both men and women, the relationship between obesity and BP has changed. Further research is required to investigate the impact of these intriguing changes in the associations on the risk of cardiovascular diseases in the Taiwanese population.

BACKGROUND: Using routinely collected patient data we explore the utility of multilevel latent class (MLLC) models to adjust for patient casemix and rank Trust performance. We contrast this with ranks derived from Trust standardised mortality ratios (SMRs). METHODS: Patients with colorectal cancer diagnosed between 1998 and 2004 and resident in Northern and Yorkshire regions were identified from the cancer registry database (n = 24,640). Patient age, sex, stage-at-diagnosis (Dukes), and Trust of diagnosis/treatment were extracted. Socioeconomic background was derived using the Townsend Index. Outcome was survival at 3 years after diagnosis. MLLC-modelled and SMR-generated Trust ranks were compared. RESULTS: Patients were assigned to two classes of similar size: one with reasonable prognosis (63.0% died within 3 years), and one with better prognosis (39.3% died within 3 years). In patient class one, all patients diagnosed at stage B or C died within 3 years; in patient class two, all patients diagnosed at stage A, B or C survived. Trusts were assigned two classes with 51.3% and 53.2% of patients respectively dying within 3 years. Differences in the ranked Trust performance between the MLLC model and SMRs were all within estimated 95% CIs. CONCLUSIONS: A novel approach to casemix adjustment is illustrated, ranking Trust performance whilst facilitating the evaluation of factors associated with the patient journey (e.g. treatments) and factors associated with the processes of healthcare delivery (e.g. delays). Further research can demonstrate the value of modelling patient pathways and evaluating healthcare processes across provider institutions.

Due to a problem of identification, how to estimate the distinct effects of age, time period and cohort has been a controversial issue in the analysis of trends in health outcomes in epidemiology. In this study, we propose a novel approach, partial least squares (PLS) analysis, to separate the effects of age, period, and cohort. Our example for illustration is taken from the Glasgow Alumni cohort. A total of 15,322 students (11,755 men and 3,567 women) received medical screening at the Glasgow University between 1948 and 1968. The aim is to investigate the secular trends in blood pressure over 1925 and 1950 while taking into account the year of examination and age at examination. We excluded students born before 1925 or aged over 25 years at examination and those with missing values in confounders from the analyses, resulting in 12,546 and 12,516 students for analysis of systolic and diastolic blood pressure, respectively. PLS analysis shows that both systolic and diastolic blood pressure increased with students' age, and students born later had on average lower blood pressure (SBP: −0.17 mmHg/per year [95% confidence intervals: −0.19 to −0.15] for men and −0.25 [−0.28 to −0.22] for women; DBP: −0.14 [−0.15 to −0.13] for men; −0.09 [−0.11 to −0.07] for women). PLS also shows a decreasing trend in blood pressure over the examination period. As identification is not a problem for PLS, it provides a flexible modelling strategy for age-period-cohort analysis. More emphasis is then required to clarify the substantive and conceptual issues surrounding the definitions and interpretations of age, period and cohort effects.

BACKGROUND: We investigate whether the changing environment caused by rapid economic growth yielded differential effects for successive Taiwanese generations on 8 components of metabolic syndrome (MetS): body mass index (BMI), systolic blood pressure (SBP), diastolic blood pressure (DBP), fasting plasma glucose (FPG), triglycerides (TG), high-density lipoprotein (HDL), Low-density lipoproteins (LDL) and uric acid (UA). METHODS: To assess the impact of age, birth year and year of examination on MetS components, we used partial least squares regression to analyze data collected by Mei-Jaw clinics in Taiwan in years 1996 and 2006. Confounders, such as the number of years in formal education, alcohol intake, smoking history status, and betel-nut chewing were adjusted for. RESULTS: As the age of individuals increased, the values of components generally increased except for UA. Men born after 1970 had lower FPG, lower BMI, lower DBP, lower TG, Lower LDL and greater HDL; women born after 1970 had lower BMI, lower DBP, lower TG, Lower LDL and greater HDL and UA. There is a similar pattern between the trend in levels of metabolic syndrome components against birth year of birth and economic growth in Taiwan. CONCLUSIONS: We found cohort effects in some MetS components, suggesting associations between the changing environment and health outcomes in later life. This ecological association is worthy of further investigation.

OBJECTIVES: The aims of this study were to: (1) investigate the relationship between ethnicity and breast cancer incidence and survival using cancer registry and Hospital Episode Statistics (HES) data; and (2) assess the impact of missing data and the recording of multiple ethnicities for some patients. DESIGN: A total of 48,234 breast cancer patients diagnosed between 1997 and 2003 in two English regions were identified. Ethnicity was missing in 16% of cases. Multiple imputation (10 iterations) of missing ethnicity was undertaken using a range of predictor variables. Multiple ethnicities for a single patient were recorded in 4% of cases. Three methods of assigning ethnicity were used: 'most popular' code, 'last recorded' code, and proportions calculated using all recorded episodes for each patient. Age-standardised incidence rate ratios (IRR) and 5-year survival were calculated before and after imputation for the three methods of assigning ethnicity. RESULTS: Breast cancer incidence was lower in the South Asian group (IRR=0.59, 95% confidence interval [CI] 0.51-0.69 compared to the White group). In unadjusted analyses, the South Asian group had consistently higher survival compared with the White group (hazard ratio [HR]=0.81, 95% CI 0.68-0.95). After adjustment for age and stage, there were no survival differences amongst the White, South Asian and Black groups. Survival was higher in the 'Other' ethnic group when using the 'last recorded' method to assign ethnicity (HR=0.62, 95% CI 0.45-0.85 compared with the White group). The results were similar before and after imputation, using all three methods of assigning ethnicity. CONCLUSIONS: Breast cancer incidence was lower in the South Asian group than in the White group. After adjusting for casemix there were no consistent survival differences amongst the ethnic groups. Although the impact of missing data and multiple ethnicities was minimal in this study, researchers should always consider these issues, as the results may not be generalisable to other populations and datasets.

Statistics is an indispensible tool for evaluating treatment effects in clinical research. Due to the complexities of periodontal disease progression and data collection, statistical analyses for periodontal research have been a great challenge for both clinicians and statisticians. The aim of this article is to provide an overview of several basic, but important, statistical issues related to the evaluation of treatment effects and to clarify some common statistical misconceptions. Some of these issues are general, concerning many disciplines, and some are unique to periodontal research. We first discuss several statistical concepts that have sometimes been overlooked or misunderstood by periodontal researchers. For instance, decisions about whether to use the t-test or analysis of covariance, or whether to use parametric tests such as the t-test or its non-parametric counterpart, the Mann-Whitney U-test, have perplexed many periodontal researchers. We also describe more advanced methodological issues that have sometimes been overlooked by researchers. For instance, the phenomenon of regression to the mean is a fundamental issue to be considered when evaluating treatment effects, and collinearity amongst covariates is a conundrum that must be resolved when explaining and predicting treatment effects. Quick and easy solutions to these methodological and analytical issues are not always available in the literature, and careful statistical thinking is paramount when conducting useful and meaningful research. © 2012 John Wiley & Sons A/S.

RATIONALE: Calcium entry through Orai1 channels drives vascular smooth muscle cell migration and neointimal hyperplasia. The channels are activated by the important growth factor platelet-derived growth factor (PDGF). Channel activation is suggested to depend on store depletion, which redistributes and clusters stromal interaction molecule 1 (STIM1), which then coclusters and activates Orai1. OBJECTIVE: To determine the relevance of STIM1 and Orai1 redistribution in PDGF responses. METHODS AND RESULTS: Vascular smooth muscle cells were cultured from human saphenous vein. STIM1 and Orai1 were tagged with green and red fluorescent proteins to track them in live cells. Under basal conditions, the proteins were mobile but mostly independent of each other. Inhibition of sarco-endoplasmic reticulum calcium ATPase led to store depletion and dramatic redistribution of STIM1 and Orai1 into coclusters. PDGF did not evoke redistribution, even though it caused calcium release and Orai1-mediated calcium entry in the same time period. After chemical blockade of Orai1-mediated calcium entry, however, PDGF caused redistribution. Similarly, mutagenic disruption of calcium flux through Orai1 caused PDGF to evoke redistribution, showing that calcium flux through the wild-type channels had been filling the stores. Acidification of the extracellular medium to pH 6.4 caused inhibition of Orai1-mediated calcium entry and conferred capability for PDGF to evoke complete redistribution and coclustering. CONCLUSIONS: The data suggest that PDGF has a nonclustering mechanism by which to activate Orai1 channels and maintain calcium stores replete. Redistribution and clustering become important, however, when the endoplasmic reticulum stress signal of store depletion arises, for example when acidosis inhibits Orai1 channels.

The National Vascular Database (NVD) is a prospective audit database collecting information of the quality of care and outcomes of patients admitted to acute hospitals in England, Wales, Scotland and Northern Ireland with several vascular disorders. The NVD has proved to be an important resource for clinical audit but by contrast its potential as a valuable research tool remains under exploited. We demonstrate proof-of-principle linkage of the NVD to Hospital Episode Statistics (HES) and UK Statistics Authority data. We present and validate Multiple Imputation (MI) methods to address problems with missingness in the linked dataset, focusing on a specific risk model. MI is applied to these linked data to extend the chosen risk model to long term mortality outcomes.

BACKGROUND: In order to improve the quality of care delivered to patients and to enable patient choice, public reports comparing hospital performances are routinely published. Robust systems of hospital 'report cards' on performance monitoring and evaluation are therefore crucial in medical decision-making processes. In particular, such systems should effectively account for and minimise systematic differences with regard to definitions and data quality, care and treatment quality, and 'case mix'. METHODS: Four methods for assessing hospital performance on mortality outcome measures were considered. The methods included combinations of Bayesian fixed- and random-effects models, and risk-adjusted mortality rate, and rank-based profiling techniques. The methods were empirically compared using 30-day mortality in patients admitted with acute coronary syndrome. Agreement was firstly assessed using median estimates between risk-adjusted mortality rates for a hospital and between ranks associated with a hospital's risk-adjusted mortality rates. Secondly, assessment of agreement was based on a classification of hospitals into low, normal or high performing using risk-adjusted mortality rates and ranks. RESULTS: There was poor agreement between the point estimates of risk-adjusted mortality rates, but better agreement between ranks. However, for categorised performance, the observed agreement between the methods' classification of the hospital performance ranged from 90 to 98%. In only two of the six possible pair-wise comparisons was agreement reasonable, as reflected by a Kappa statistic; it was 0.71 between the methods of identifying outliers with the fixed-effect model and 0.77 with the hierarchical model. In the remaining four pair-wise comparisons, the agreement was, at best, moderate. CONCLUSIONS: Even though the inconsistencies among the studied methods raise questions about which hospitals performed better or worse than others, it seems that the choice of the definition of outlying performance is less critical than that of the statistical approach. Therefore there is a need to find robust systems of 'regulation' or 'performance monitoring' that are meaningful to health service practitioners and providers.

BACKGROUND: The obesity-psychological distress relationship remains controversial. PURPOSE: This study aims to assess whether adolescents' psychological distress was associated with body mass index (BMI) class membership determined by latent class analysis. METHODS: Distress (anxiety, depression) and BMI were measured annually for 4 years in 1,528 adolescents. Growth mixture modeling derived latent BMI trajectory classes for models with 2-11 classes. The relationship of distress to class membership was examined in the best-fitting model using vector generalized linear regression. RESULTS: BMI trajectories were basically flat. The five-class model [normal weight (48.8%), overweight (36.7%), obese who become overweight (3.7%), obese (9.4%), and severely obese (1.3%)] was the preferred model (Bayesian information criterion = 22789.2, df = 31; ρ = 0.84). Greater distress was associated with higher baseline BMI and, therefore, class membership. CONCLUSIONS: Psychological distress is associated with higher BMI class during adolescence. To determine whether distress "leads" to greater weight gain may require studies of younger populations.

BACKGROUND: Stage of disease and socioeconomic background (SEB) are often used to 'explain' differences in breast cancer outcomes. There are challenges for all types of analysis (eg, survival analysis, logistic regression), including missing data, measurement error and the 'reversal paradox'. This study investigates the association between SEB and survival status within 5 years of breast cancer diagnosis using (1) logistic regression with and without adjustment for stage and (2) logistic latent class analysis (LCA) excluding stage as a covariate but with and without stage as a latent class predictor. METHODS: Women diagnosed with invasive breast cancer between 1998 and 2000 in one UK region were identified (n=11 781). Multilevel logistic regression was performed using standard regression and LCA. Models included SEB (2001 Townsend Index), age and stage ('missing' stage (8.0%) modelled as a separate category). The association of SEB with stage was also assessed. RESULTS: Using standard regression, there was a substantial association between SEB and death within 5 years, with and without adjustment for stage. Using LCA, patients were assigned to a large good prognosis group and a small poor prognosis group. The association between SEB and survival was substantive in both classes for the model without stage, but only in the larger class for the model with stage. Increasing deprivation was associated with more advanced stage at diagnosis. CONCLUSIONS: LCA categorises patients into prognostic groups according to patient and tumour characteristics, providing an alternative strategy to the usual statistical adjustment for stage.

OBJECTIVE: To provide an overview to the problems in study design and associated analyses of follow-up studies in dental research, particularly addressing three issues: treatment-baselineinteractions; statistical power; and nonrandomization. BACKGROUND: Our previous work has shown that many studies purport an interacion between change (from baseline) and baseline values, which is often based on inappropriate statistical analyses. A priori power calculations are essential for randomized controlled trials (RCTs), but in the pre-test/post-test RCT design it is not well known to dental researchers that the choice of statistical method affects power, and that power is affected by treatment-baseline interactions. A common (good) practice in the analysis of RCT data is to adjust for baseline outcome values using ancova, thereby increasing statistical power. However, an important requirement for ancova is there to be no interaction between the groups and baseline outcome (i.e. effective randomization); the patient-selection process should not cause differences in mean baseline values across groups. This assumption is often violated for nonrandomized (observational) studies and the use of ancova is thus problematic, potentially giving biased estimates, invoking Lord's paradox and leading to difficulties in the interpretation of results. METHODS: Baseline interaction issues can be overcome by use of statistical methods; not widely practiced in dental research: Oldham's method and multilevel modelling; the latter is preferred for its greater flexibility to deal with more than one follow-up occasion as well as additional covariates To illustrate these three key issues, hypothetical examples are considered from the fields of periodontology, orthodontics, and oral implantology. CONCLUSION: Caution needs to be exercised when considering the design and analysis of follow-up studies. ancova is generally inappropriate for nonrandomized studies and causal inferences from observational data should be avoided.

This chapter contains sections titled: Introduction Univariate analysis Generalized linear models Multivariate methods Conclusions References Introduction Univariate analysis Generalized linear models Multivariate methods Conclusions References

Analysis of change is probably the most commonly adopted study design in medical and dental research when comparing the efficacy of two or more treatment modalities. The most commonly used methods for testing the difference in treatment efficacy are the two-sample t-test and the analysis of covariance (ANCOVA). It has been suggested that ancova should be used in the analysis of change for data from randomized controlled trials (RCTs) as a result of its greater statistical power. However, it is less well known that although both methods will give rise to similar results in the analysis of change for RCTs, there are different assumptions behind these methods in terms of the relationship between baseline value and the subsequent change, and the results may therefore differ if baseline values are not balanced between groups. This article uses structural equation modelling as a conceptual framework to explain the assumptions behind these methods, and two examples are used to show when the two methods yield similar results and why, in some non-randomized studies, the two methods might give substantially different results, known as 'Lord's paradox' in the statistical literature. For the appropriate interpretation of non-randomized studies, the assumptions underlying these methods therefore need to be taken into consideration.

Background Chronic Heart Failure (CHF) is one of the leading cause of hospitalizations and deaths, more especially in old people, and this causes a substantial clinical and economic burden to the government. Using risk prediction models to accurately understand the dynamics of survival patterns amongst patients with CHF conditions would provide guidance to health care professionals in decision making on how to improve delivery of care. However, prediction models used in medical research often fail to accurately predict health outcomes due to methodological limitations. These models particularly perform poorly when predicting narrowly targeted subgroups of patients. We explore the role of latent class regression (LCR) analysis to model the survival of patients with CHF. We seek to show that using LCR improves the modelling of health outcomes as it accounts for unobserved heterogeneity that exists naturally within the patient data.Methods LCR generally involves identifying hidden latent classes within data and uses patient’s demographic characteristics and other covariates to predict class membership and separate regression models for each class. These latent classes may correspond to subgroups of patients with specific characteristics that affect their survival. The rationale is that one class will be more susceptible to deaths compared to another. The United Kingdom Heart Failure Evaluation and Assessment of Risk Trial (UK-HEART) recruited patients with signs and symptoms of CHF between July 2006 and December 2014. A total of 1802 records were available on patient characteristics as well as medications. We used some of these variables to model survival of patients within a latent class framework by estimating a single regression model for both latent classes. We increased complexity of our model by allowing each class to have a separate survival model.Results We used the area under the receiver operating characteristic (ROC) curve to assess the performance of these two class models. Overall, our novel approach performed better than the traditional one-model-fits-all approach. Our model gave an area under the curve (AUC) of 0.87 while the traditional model yielded an AUC of 0.68.

Background In health studies, proportions and percentages can often seem more informative than raw counts and therefore appear to be of more interest to analysts. However, it has long been acknowledged that their use is problematic in correlation and regression analyses where they comprise common components that are present in both the dependent and independent constituents of a model (exposure and outcome), as in the regression analysis of proportions with common denominators. We demonstrate this so-called mathematical coupling with real-world examples aided by directed acyclic graphs (DAGs) and simulations.Methods We consider three possible real-world scenarios: (1) the population size (N) of a geographical area causes both the number of people living in detached houses (X) and the number of people living in care homes (Y), within each area, but the number of detached houses (X) does not cause the number of care homes (Y) within any area, or vice versa; (2) the population size (N) of a geographical area causes both the number of people with no formal qualifications (X) and the number of people with poor self-reported health (Y), while both the population size (N) and number of people with no formal qualifications (X) are causes of the number of people with self-reported poor health (Y); and (3) within a geographical area, the area wealth (X) causes the number of elderly people (N), while both area wealth (X) and the number of elderly people (N) cause social care expenditure (Y).Results We show how historical solutions to the issue of mathematical coupling caused by a common denominator hold under the situation when the denominator is a confounder of the exposure outcome relationship; i.e. the results of the simulated examples under scenarios 1 and 2 result in expected regression coefficients. The same solution does not hold in scenario 3, when the denominator is a mediator (i.e. lies on the causal path) between the exposure and outcome.

Background It is widely acknowledged that the obesity epidemic exhibits many characteristics of a complex system – individual heterogeneity and autonomy, interdependence, adaptivity, feedback loops, threshold effects, and emergence. Altering patterns of obesity requires robust methods that are capable of accurately estimating the causal effects of potential interventions in the presence of such complexity. Two groups of methods that have been proposed are of interest: (1) statistical regression models informed by directed acyclic graphs (DAGs), underpinned by graphical model theory; and (2) individual-based simulation models (IBSMs), consisting of microsimulation models (MSMs) and agent-based models (ABMs). Both groups of Methods may be used to evaluate causal effects as counterfactual contrasts, but methodological work seeking to compare and contrast the two methods is lacking, because they have only come to prominence relatively recently and have been largely confined to separate research disciplines. We sought to examine how these methods have been used within obesity research, identify the primary philosophical and methodological differences between them, and identify any implications for future causal analyses.Methods We conducted searches of Medline, EMBASE, and Scopus to identify articles (from 1996–2016) that sought to make causal inferences relating to obesity, weight, or other closely associated health-related behaviours (e.g. exercise) by utilising either DAG-based statistical methods or IBSMs.Results Our search returned 38 relevant articles utilising DAG-based statistical methods and 45 relevant articles utilising IBSMs (31 MSMs and 14 ABMs). From these results, we identified four primary differences between the two groups of Methods: (1) their relative reliance on theory versus data; (2) the timescales upon which they operate; (3) their relative focus on fixed versus random effects; and (4) the ways in which they assess the effects of (hypothetical) interventions.

Adherence to guideline-indicated care for the treatment of non-ST-elevation myocardial infarction (NSTEMI) is associated with improved outcomes. We investigated the extent and consequences of non-adherence to guideline-indicated care across a national health system. A cohort study (ClinicalTrials.gov identifier: NCT02436187) was conducted using data from the Myocardial Ischaemia National Audit Project (n = 389,057 NSTEMI, n = 247 hospitals, England and Wales, 2003-2013). Accelerated failure time models were used to quantify the impact of non-adherence on survival according to dates of guideline publication.Over a period of 1,079,044 person-years (median 2.2 years of follow-up), 113,586 (29.2%) NSTEMI patients died. Of those eligible to receive care, 337,881 (86.9%) did not receive one or more guideline-indicated intervention; the most frequently missed were dietary advice (n = 254,869, 68.1%), smoking cessation advice (n = 245,357, 87.9%), P2Y12 inhibitors (n = 192,906, 66.3%) and coronary angiography (n = 161,853, 43.4%). Missed interventions with the strongest impact on reduced survival were coronary angiography (time ratio: 0.18, 95% confidence interval (CI): 0.17-0.18), cardiac rehabilitation (time ratio: 0.49, 95% CI: 0.48-0.50), smoking cessation advice (time ratio: 0.53, 95% CI: 0.51-0.57) and statins (time ratio: 0.56, 95% CI: 0.55-0.58). If all eligible patients in the study had received optimal care at the time of guideline publication, then 32,765 (28.9%) deaths (95% CI: 30,531-33,509) may have been prevented.The majority of patients hospitalised with NSTEMI missed at least one guideline-indicated intervention for which they were eligible. This was significantly associated with excess mortality. Greater attention to the provision of guideline-indicated care for the management of NSTEMI will reduce premature cardiovascular deaths.

Objectives: To characterize the impact of PCB exposure on DNA methylation in peripheral blood leucocytes and to evaluate the corresponding changes in relation to possible health effects, with a focus on B-cell lymphoma. Methods: We conducted an epigenome-wide association study on 611 adults free of diagnosed disease, living in Italy and Sweden, in whom we also measured plasma concentrations of 6 PCB congeners, DDE and hexachlorobenzene. Results: We identified 650 CpG sites whose methylation correlates strongly (FDR < 0.01) with plasma concentrations of at least one PCB congener. Stronger effects were observed in males and in Sweden. This epigenetic exposure profile shows extensive and highly statistically significant overlaps with published profiles associated with the risk of future B-cell chronic lymphocytic leukemia (CLL) as well as with clinical CLL (38 and 28 CpG sites, respectively). For all these sites, the methylation changes were in the same direction for increasing exposure and for higher disease risk or clinical disease status, suggesting an etiological link between exposure and CLL. Mediation analysis reinforced the suggestion of a causal link between exposure, changes in DNA methylation and disease. Disease connectivity analysis identified multiple additional diseases associated with differentially methylated genes, including melanoma for which an etiological link with PCB exposure is established, as well as developmental and neurological diseases for which there is corresponding epidemiological evidence. Differentially methylated genes include many homeobox genes, suggesting that PCBs target stem cells. Furthermore, numerous polycomb protein target genes were hypermethylated with increasing exposure, an effect known to constitute an early marker of carcinogenesis. Conclusions: This study provides mechanistic evidence in support of a link between exposure to PCBs and the etiology of CLL and underlines the utility of omic profiling in the evaluation of the potential toxicity of environmental chemicals.

Objective: To investigate the impact of premature extraction of primary teeth (PEPT) on orthodontic treatment need in a cohort of children participating in the Born in Bradford (BiB) longitudinal birth cohort. Design: Observational, cross-sectional cohort. Participants: We aim to recruit 1000 children aged 7–11 years: 500 with a history of PEPT and 500 matched non-PEPT controls. Methods: After informed consent/assent, orthodontic records will be collected, including extra and intra-oral photographs and alginate impressions for study models. Participants will also complete a measure of oral health-related quality of life (COHIP-SF 19). The records will be used to quantify space loss, identify other occlusal anomalies and assess orthodontic treatment need using the Index of Orthodontic Treatment Need. For each outcome, summary statistics will be calculated and the data for children with and without PEPT compared. The records of the children identified to be in need of orthodontic treatment will be examined by an expert orthodontic panel to judge if this treatment should be undertaken at the time of the records or delayed until the early permanent dentition. Collecting robust records in the mixed dentition provides the clinical basis to link each stage of the causal chain and enable the impact of PEPT on orthodontic need to be characterised. This study is the first to provide the foundations for future longitudinal data collection allowing the long-term impact of PEPT to be studied.

Prediction and causal explanation are fundamentally distinct tasks of data analysis. In health applications, this difference can be understood in terms of the difference between prognosis (prediction) and prevention/treatment (causal explanation). Nevertheless, these two concepts are often conflated in practice. We use the framework of generalised linear models (GLMs) to illustrate that predictive and causal queries require distinct processes for their application and subsequent interpretation of results. In particular, we identify five primary ways in which GLMs for prediction differ from GLMs for causal inference: (1) The covariates that should be considered for inclusion in (and possibly exclusion from) the model; (2) How a suitable set of covariates to include in the model is determined; (3) Which covariates are ultimately selected, and what functional form (i.e. parameterisation) they take; (4) How the model is evaluated; and (5) How the model is interpreted. We outline some of the potential consequences of failing to acknowledge and respect these differences, and additionally consider the implications for machine learning (ML) methods. We then conclude with three recommendations which we hope will help ensure that both prediction and causal modelling are used appropriately and to greatest effect in health research.

Although the studies highlighted in Kinlen and Peto’s letter describe situations they take to be “national in scope”, none of these adopted the ‘region-wide’ analysis we recommend. Rather, these studies have focussed on rural areas with small populations experiencing extreme levels of inward-migration that had been selected from larger regions/nation states. To definitively avoid bias, our study points to the need for comparisons of areas with varying levels of inward migration, either by comparing all areas within an entire region/nation state or random subsets thereof.

Most prediction models that are used in medical research fail to accurately predict health outcomes due to methodological limitations. Using routinely collected patient data, we explore the use of a Cox proportional hazard (PH) model within a latent class framework to model survival of patients with chronic heart failure (CHF). We identify subgroups of patients based on their risk with the aid of available covariates. We allow each subgroup to have its own risk model.We choose an optimum number of classes based on the reported Bayesian information criteria (BIC). We assess the discriminative ability of the chosen model using an area under the receiver operating characteristic curve (AUC) for all the cross-validated and bootstrapped samples.We conduct a simulation study to compare the predictive performance of our models. Our proposed latent class model outperforms the standard one class Cox PH model.

Studies of change are a cornerstone of research in the health sciences. Robust analyses of change are however extremely challenging, especially in observational data. In simple exposure-outcome scenarios, one common approach is to create and analyse an outcome ‘change-score’ by subtracting the baseline outcome from follow-up outcome. Tens-of-thousands of articles can be found that have adopted this approach. Unfortunately, this approach fails to capture the (desired) modifiable component of the outcome variable that occurred after baseline. On the contrary, it retains sign-reversed information from the baseline outcome that can create extremely-misleading associations. Using directed acyclic graphs (DAGs) and illustrative simulations, this study explains why outcome change-scores do not capture the true causal quantity of interest and demonstrates the extent of disagreement between robust analyses and change-score analyses in various circumstances. DAGs with deterministic nodes are used to explain why change-scores do not capture the (desired) modifiable component of the outcome that occurs after baseline. The implications are then illustrated in simulated data, by analysing outcome change-scores with respect to a baseline exposure under several causal scenarios. Data were simulated using DAGitty R 0.2–2 to match three broad scenarios, with the baseline outcome as 1) competing exposure, 2) confounder, and 3) mediator for the total causal effect of the exposure on the follow-up outcome. Means, standard deviations, and distributions were informed by data from the US National Health and Nutrition Examination Survey for 2009–2014. The association between the baseline exposure and outcome change-score was estimated by linear regression; and the coefficients compared to the known truth and coefficients obtained from robust analyses. Naïve regression analyses of the outcome change-score (insulin) with respect to the baseline exposure (waist circumference) produced biased causal inferences in all scenarios except where the exposure and outcome were uncorrelated at baseline (as in a randomised experiment). When the baseline outcome (insulin) confounded the effect of the baseline exposure (waist circumference) on the follow-up outcome, the naïve regression estimate remained confounded. When the baseline outcome (insulin) mediated the effect of the baseline exposure (waist circumference) on the follow-up outcome, the naïve regression estimate had the opposite sign to the total causal effect. Analyses ofchange-scores should be avoided in observational health research, as they can produce extremely misleading coefficients. Previous observational studies that have naively analysed and interpreted change-score variables should be viewed with extreme caution and any recommendations revisited.

Compositional data (CD) comprise the parts of some whole, for which all parts sum to that whole; the whole may vary across individual units of analysis or remain fixed. Such data are common in many contexts, where interest often lies in understanding the effect of a particular part in relation to a subsequent outcome. Many of the inherent challenges associated with analysing CD have been discussed previously, though not within a formal causal framework by utilising directed acyclic graphs (DAGs). We use DAGs to consider the specific issue of collider bias as it pertains to CD. We demonstrate how to depict CD using DAGs, and identify two distinct effect estimands in the generic case: (1) the ‘unbiased’ (total) effect, and (2) the ‘collider biased’ effect. We consider each effect in the context of three specific example scenarios involving CD with variable or fixed totals: (1) the relationship between the economically active population and area-level gross domestic product (GDP) (variable total); (2) the relationship between fat consumption and body weight (variable total); and (3) the relationship between time spent sedentary and body weight (fixed total). For each scenario, we consider the distinct interpretation of each effect, and the resulting implications for related analyses. In scenario (1), the ‘unbiased’ effect represents the average change in GDP that results from adding economically active individuals to the area whilst doing nothing to the population of economically inactive individuals, whereas the ‘collider biased’ effect represents the average change that results from swapping economically inactive individuals for economically active ones. In scenario (2), the ‘unbiased’ effect represents the average change in weight that results from adding fat to an individual’s diet irrespective of other macronutrient consumption, whilst the ‘collider biased’ effect represents the average change that results in swapping ‘other’ macronutrient consumption for fat consumption. In scenario (3), only the ‘collider biased’ effect is estimable and causally meaningful; it represents the average change in weight that results from swapping time spent physically active for time spent sedentary. For CD with variable totals, both effects may be estimable and causally meaningful, depending upon the specific question of interest. Researchers should be clear about which effect is being sought and estimated, since they may be radically different quantities. For CD with fixed totals, only the ‘collider biased’ effect has any meaning. Careful attention must be paid to sensibly interpreting the relative effects that characterise this type of data.

Healthcare provider performance is commonly assessed using patient outcomes, e.g. survival rates. Patient characteristics that may affect outcomes in the absence of genuine provider-level differences must therefore be balanced across providers to ensure a fair comparison. There are many methods that can accommodate this patient ‘casemix’ but none that also allow the assessment of provider-level covariate effects, i.e. the potential causes of performance differences. We aim to demonstrate the utility of multilevel latent class (MLC) modelling to identify causal provider-level covariate effects after accommodating patient differences. We simulated data for patients and providers, based on a previously utilised real-world dataset of patients diagnosed with colorectal cancer. Age at diagnosis, sex and socioeconomic status were included at the patient level, and we explored a continuous outcome. We included both binary and continuous effects at the provider level, to reflect organisational features such as surgeon speciality or available beds, although these were analysed separately to demonstrate proof-of-principle. We simulated unique sets of 100 datasets using a range of coefficient effect values and error variances. Interest lies in the ability of the MLC model to recover these simulated provider-level coefficient effects. Models contained one patient-level latent class and up to five provider-level latent classes. For the binary provider-level covariate, median recovered values were almost identical to simulated effects throughout, e.g. for the simulated coefficient value 0.500 at 33% error variance, the median recovered value was 0.499 (95% CI 0.489–0.509) across all models. For the continuous provider-level covariate, median recovered values improved as the number of provider-level latent classes were increased, e.g. for the simulated coefficient value 0.200 at 33% error variance, the median recovered value was 0.153 (95% CI 0.113–0.184) for two provider-level classes and 0.191 (95% CI 0.168–0.210) for five provider-level classes. The MLC modelling approach achieved successful recovery of simulated coefficient values, within credible intervals for at least three provider-level latent classes. Very small simulated coefficient values were not recovered as well as higher values, which may be due to the variability introduced during simulation dominating the coefficient effect. There is also some attenuation of effect seen for the continuous provider-level covariate. We have demonstrated the utility of this approach to separate modelling for prediction (to accommodate patient casemix) and for causal inference (to explore provider-level effects) across a data hierarchy. There is much scope to extend the assessment of upper-level causal effects by consideration of a multivariable DAG.

The practice of routinely weighing pregnant women to monitor their ‘weight gain’ is controversial. In the United States, the National Academy of Medicine (NAM) advises regular weighing and recommends ‘optimum’ gain targets according to pre-pregnancy body mass index (BMI). In the United Kingdom, the National Institute for Health and Care Excellence (NICE) advises against routinely checking women’s weight as pregnancy progresses. This quite radical difference hinges on the believed causal effect of ‘gestational weight gain’ (GWG) on adverse pregnancy outcomes, such as macrosomia (birthweight≥4 kg). However, estimating this is very difficult because some association is expected between GWG and birthweight, by definition, because the total maternal weight ‘gain’ implicitly includes the offspring’s weight. This study sought to highlight this problem and explore the size of this ‘tautological association’ in simulated data. Data were simulated using DAGitty R 0.2–2 to reflect three causal scenarios: 1) Birthweight caused by maternal height alone, 2) Birthweight caused by maternal height and maternal pre-pregnancy weight 3) Birthweight caused by maternal height, maternal pre-pregnancy weight, and maternal net end-of-pregnancy weight (i.e. ‘gain’). GWG was constructed from [maternal net end-of-pregnancy weight + birthweight]-[maternal pre-pregnancy weight]. The odds ratios (ORs) for macrosomia by GWG were estimated by logistic regression, with and without conditioning on maternal pre-pregnancy BMI, constructed from [maternal pre-pregnancy weight]/[maternal height]2. Simulation parameters were informed by full and partial correlations observed in the Danish National Birth Cohort. Large associations were observed between GWG and macrosomia in all three scenarios, even though weight ‘gain’ only caused birthweight in the third scenario. The crude OR (95% credible interval) of macrosomia for GWG ‘above’ NAM guidelines compared with ‘recommended’ GWG were 1.26 (1.17–1.36), 1.34 (1.24–1.45) and 1.52 (1.41–1.65) respectively for scenarios 1 (birthweight caused by height only), 2 (height and pre-pregnancy weight), and 3 (height, pre-pregnancy weight, and end-of-pregnancy weight). Adjustment for pre-pregnancy BMI only modestly changed these associations, with ORs of 1.27 (1.18–1.37), 1.28 (1.19–1.39), and 1.42 (1.32–1.54) respectively. The apparent causal effect of maternal net weight ‘gain’ on birthweight (and hence macrosomia) is difficult to identify because the total maternal weight gain observed includes that of the offspring. A tautological association is therefore observed even when maternal weight has no causal effect on birthweight. Existing evidence regarding the ‘effect’ of GWG on birthweight should therefore be viewed with caution and should not be used to inform guidelines on ‘ideal’ gains in weight.

Background: Studies investigating the population-mixing hypothesis in childhood leukemia principally use two analytical approaches: (i) non-random selection of areas according to specific characteristics, followed by comparisons of their incidence of childhood leukemia with that expected based on the national average; and (ii) regression analyses of region-wide data to identify characteristics associated with the incidence of childhood leukemia. These approaches have generated contradictory results. We compare these approaches using observed and simulated data. Methods: We generated 10,000 simulated regions using the correlation structure and distributions from a United Kingdom dataset. We simulated cases using a Poisson distribution with the incidence rate set to the national average assuming the null hypothesis that only population size drives the number of cases. Selection of areas within each simulated region was based on characteristics considered responsible for elevated infection rates (population density and inward-migration) and/or elevated leukemia rates. We calculated effect estimates for 10,000 simulations and compared results to corresponding observed data analyses. Results: When the selection of areas for analysis is based on apparent clusters of childhood leukemia biased assessments occur; the estimated 5-year incidence of childhood leukemia ranged between 0 and 8 per 10,000 children in contrast to the simulated 2 cases per 10,000 children, similar to the observed data. Performing analyses on region-wide data avoids these biases. Conclusions: Studies using non-random selection to investigate the association between childhood leukemia and population mixing are likely to have generated biased findings. Future studies can avoid such bias using a region-wide analytical strategy.

Advanced Modelling Strategies: Challenges and pitfalls in robust causal inference with observational data summarises the lecture notes prepared for a four-day workshop sponsored by the Society for Social Medicine and hosted by the Leeds Institute for Data Analytics (LIDA) at the University of Leeds on 17th-20th July 2017.

Purpose To examine relationships between functional outcomes after pediatric acquired brain injury (ABI) and measures of rehabilitation dose. Methods An observational study of children receiving residential neurorehabilitation after severe ABI. Results Basic total rehabilitation dose shows a paradoxical inverse relationship to global outcome. This is due to confounding by both initial injury severity and length of stay, and variation in treatment content for a given total rehabilitation dose. Content-aware rehabilitation dose measures show robust positive correlations between fractions of rehabilitation treatment received and plausibly related aspects of outcome: specifically, between rates of recovery of gross motor function and the fraction of rehabilitation effort directed to active practice and motor learning. This relationship was robust to adjustment for therapists’ expectations of recovery. Conclusion Content-aware measures of rehabilitation dose are robustly causally related to pertinent aspects of outcome. These findings are step toward a goal of comparative effectiveness research in pediatric neurorehabilitation.

Background: B-cell lymphomas are a diverse group of hematological neoplasms with differential etiology and clinical trajectories. Increased insights in the etiology and the discovery of prediagnostic markers have the potential to improve the clinical course of these neoplasms. Methods: We investigated in a prospective study global gene expression in peripheral blood mononuclear cells of 263 incident B-cell lymphoma cases, diagnosed between 1 and 17 years after blood sample collection, and 439 controls, nested within two European cohorts. Results: Our analyses identified only transcriptomic markers for specific lymphoma subtypes; few markers of multiple myeloma (N = 3), and 745 differentially expressed genes in relation to future risk of chronic lymphocytic leukemia (CLL). The strongest of these associations were consistently found in both cohorts and were related to (B-) cell signaling networks and immune system regulation pathways. CLL markers exhibited very high predictive abilities of disease onset even in cases diagnosed more than 10 years after blood collection. Conclusions: This is the first investigation on blood cell global gene expression and future risk of B-cell lymphomas. We mainly identified genes in relation to future risk of CLL that are involved in biological pathways, which appear to be mechanistically involved in CLL pathogenesis. Many but not all of the top hits we identified have been reported previously in studies based on tumor tissues, therefore suggesting that a mixture of preclinical and early disease markers can be detected several years before CLL clinical diagnosis.

Background During the first wave of the COVID-19 pandemic, the United Kingdom experienced one of the highest per-capita death tolls worldwide. It is debated whether this may partly be explained by the relatively late initiation of voluntary social distancing and mandatory lockdown measures. In this study, we used simulations to estimate the number of cases and deaths that would have occurred in England by 1 June 2020 if these interventions had been implemented one or two weeks earlier, and the impact on the required duration of lockdown. Methods Using official reported data on the number of Pillar 1 lab-confirmed cases of COVID-19 and associated deaths occurring in England from 3 March to 1 June, we modelled: the natural (i.e. observed) growth of cases, and the counterfactual (i.e. hypothetical) growth of cases that would have occurred had measures been implemented one or two weeks earlier. Under each counterfactual condition, we estimated the expected number of deaths and the time required to reach the incidence observed under natural growth on 1 June. Results Introducing measures one week earlier would have reduced by 74% the number of confirmed COVID-19 cases in England by 1 June, resulting in approximately 21,000 fewer hospital deaths and 34,000 fewer total deaths; the required time spent in full lockdown could also have been halved, from 69 to 35 days. Acting two weeks earlier would have reduced cases by 93%, resulting in between 26,000 and 43,000 fewer deaths. Conclusions Our modelling supports the claim that the relatively late introduction of social distancing and lockdown measures likely increased the scale, severity, and duration of the first wave of COVID-19 in England. Our results highlight the importance of acting swiftly to minimise the spread of an infectious disease when case numbers are increasing exponentially.

Objective: Different body mass index (BMI) trajectories that result in obesity may have diverse health consequences, yet this heterogeneity is poorly understood. We aimed to identify distinct classes of individuals who share similar BMI trajectories and examine associations with cardiometabolic health. Approach and Results: Using data on 3549 participants in ALSPAC (Avon Longitudinal Study of Parents and Children), a growth mixture model was developed to capture heterogeneity in BMI trajectories between 7.5 and 24.5 years. Differences between identified classes in height growth curves, body composition trajectories, early-life characteristics, and a panel of cardiometabolic health measures at 24.5 years were investigated. The best mixture model had 6 classes. There were 2 normal-weight classes: normal weight (nonlinear; 35% of sample) and normal weight (linear; 21%). Two classes resulted in young-adulthood overweight: normal weight increasing to overweight (18%) and normal weight or overweight (16%). Two classes resulted in young-adulthood obesity: normal weight increasing to obesity (6%) and overweight or obesity (4%). The normal-weight-increasing-to-overweight class had more unfavorable levels of trunk fat, blood pressure, insulin, HDL (high-density lipoprotein) cholesterol, left ventricular mass, and E/e′ ratio compared with the always-normal-weight-or-overweight class, despite the average BMI trajectories for both classes converging at ≈26 kg/m2 at 24.5 years. Similarly, the normal-weight-increasing-to-obesity class had a worse cardiometabolic profile than the always-overweight-or-obese class. Conclusions: Individuals with high and stable BMI across childhood may have lower cardiometabolic disease risk than individuals who do not become overweight or obese until late adolescence.

Throughout the COVID-19 pandemic, many countries have relied on population-wide ‘lockdown’ measures to cut chains of transmission and prevent health services from being overwhelmed. Because of their substantial social and economic costs, these policies have typically only been implemented when it becomes clear that less stringent measures are insufficient to control virus spread. Nevertheless, the consequences of delaying lockdowns can be even more costly. In this study, we sought to quantify the extent to which delaying implementation of lockdown measures increases total case numbers and ultimately prolongs the length of lockdown required, by considering the first wave of cases across Europe. Data pertaining to COVID-19 cases and containment/closure policies across Europe were obtained from online repositories maintained by Johns Hopkins and Oxford Universities, respectively. We identified key policy dates in each country, including the first date for which any restrictions were recommended or required, the date of lockdown, and the date lockdown was eased. We estimated the parameters governing the growth of COVID-19 during the first wave within each country, using linear splines fit to incident ~ cumulative cases. To allow for standardised comparisons across countries, we considered three distinct periods of growth: initial uncontrolled growth, growth under recommended restrictions, and growth under lockdown. Using stochastic simulations, we then estimated: the natural (i.e. observed) growth of cases, and the counterfactual (i.e. hypothetical) growth of cases that would have been observed in each country had lockdown measures been implemented 1, 3, 5, or 7 days earlier. Under each condition, we estimated the percentage change in total cases, and the percentage reduction in the required length of lockdown. 39 European countries had both cases and policy data available, of which 33 entered lockdown and had estimable growth parameters. When lockdown was implemented 3 days earlier, total first wave cases were reduced by 16.93% (Q1-Q3: 10.40–19.49) and the median required length of lockdown was reduced by 6.52% (Q1-Q3: 1.92, 10.53). When lockdown was implemented 7 days earlier, total first wave cases were reduced by 31.46% (Q1-Q3: 23.53–37.86) and the median required length of lockdown was reduced by 14.69% (Q1-Q3: 5.48, 24.40). Delaying implementation of national lockdown policies is likely to result in substantial case increases that ultimately prolong the length of lockdown required. Our results highlight the importance of acting swiftly to minimise the spread of COVID-19 when case numbers are increasing exponentially.

Understanding the factors that predict/cause individuals’ withdrawal, or dropout, from weight loss can provide useful insight into adaptations that could ensure that such programmes have greater impact. If one event follows another, conclusions are drawn that the first event caused the second. However, these associations may be observed due to chance, confounding, or selection bias. Although a lot of research has been conducted to identify factors related to attrition and adherence in weight management/loss programmes, their findings do not have a concrete (causal) interpretation beyond recognising that some predictors are often favoured over others from an initial pool of candidate predictors. Dalle Grave et al. (2015) recruited 634 patients seeking obesity treatment at Italian medical centres. They performed logistic regression to assess the association between demographic, personality characteristics, eating disorder features, psychological well-being, and attrition. This study aims to illustrate the key issues through directed acyclic graph (DAG) informed re-analysis of the Dalle Grave et al. (2015)’s data to explore if and by how much conclusions might vary between common prediction approaches and a causal inference approach. According to Dalle Grave et al. (2015), personality traits, which were assessed through the Temperament and Character Inventory (TCI), are less relevant in predicting attrition. In contrast, causal inference analysis suggests that temperament scores (harm avoidance (Probability=0.33; CI=0.29, 0.37), novelty seeking (Probability=0.34; CI=0.30, 0.38), persistence (Probability=0.30; CI=0.26, 0.34), and reward dependence (Probability=0.30; CI=0.26, 0.33)) and character scores (self-transcendence (Probability=0.34; CI=0.30, 0.39), cooperativeness (Probability=0.32; CI=0.27, 0.36), self-directedness (Probability=0.32; CI=0.27, 0.37)) are causally associated with higher probability of drop-out. Additionally, Dalle Grave et al. (2015) considered body uneasiness scores to be irrelevant in predicting drop-out. Whereas, causal inference analysis indicated that higher body uneasiness scores are causally associated with the highest probability of drop-out (Probability=0.39; CI=0.34, 0.44). New insights into factors that predict/cause drop-out from weight-loss programmes can be gained through causal inference-informed analysis. On the basis of this re-analysis, factors previously identified as irrelevant or excluded with respect to a traditional prediction perspective appear to be important from a causal perspective. Dalle Grave et al. (2015)’s analysis can be considered a case of the ‘table 2’ fallacy, where mutually adjusted coefficients in a prediction model are (inappropriately) inferred to have an equivalent interpretation. Different causal models must be generated, based on a DAG, to derive ‘correct’ (causal) inferences.

In this review, we present some ethical imperatives observed in this pandemic from a data ethics perspective. Our exposition connects recurrent ethical problems in the discipline, such as, privacy, surveillance, transparency, accountability, and trust, to broader societal concerns about equality, discrimination, and justice. We acknowledge data ethic’s role as significant to develop technological, inclusive, and pluralist societies.

During the COVID-19 pandemic we have seen various disastrous approaches regarding the use an implementation of measures and studies that performed on past and current health data. Accordingly, in this study, we criticize the lack of conceptual engineering to integrate ethical principles and values into the design and application of data-driven endeavours, with a particular examination at health data. We argue how we cannot strive for a robust ethical assessment without a critically causal framework Firstly, we analyse the translational gap and conceptual conflation of the terms: ‘bias and fairness’ and ‘transparency and explainability’, highlighting the misleading definitions and uses given to these concepts at a technical and ethical level. The main distinctions presented clarify the moral expectations given to these concepts and criticise the insufficient development of a conceptual analysis that targets them. We suggest that a fundamental part of a solution to reduce this translational gap implies embracing and applying a causal framework. Thus, we show why using causal models and, most importantly, a causal narrative cannot only help to prevent unethical effects, but it can also influence the efficiency of prediction models and their outcomes. Efficiency, in this case, transforms into an ethically laden concept that demands a causal narrative to align with ethical principles. Finally, we go through examples of COVID-19 decision-making that could have benefitted from a causal approach, highlighting the negative consequences of the NHS electronic health records platform and an OpenSAFELY publication in Nature that substantially suffers from the Table 2 Fallacy. This analysis puts into discussion an interdisciplinary approach to increase critical ethical awareness about fairness. Providing robust and reliable frameworks to analyse and present data, especially in sensitive times like a world pandemic, requires trustworthy practices. Integrating ethics into data-driven solutions cannot be limited by the bias-aware fairness formalisations or the naïve applications of transparency and explainability. When it comes to the real-world application of models, their effects can harm individuals in society. Non-causal approaches tend to dissipate elements of agency and responsibility, which are fundamental to the development of what we can call ‘good science’.

Where performance comparison of healthcare providers is of interest, characteristics of both patients and the health condition of interest must be balanced across providers for a fair comparison. This is unlikely to be feasible within observational data, as patient population characteristics may vary geographically and patient care may vary by characteristics of the health condition. We simulated data for patients and providers, based on a previously utilized real-world dataset, and separately considered both binary and continuous covariate-effects at the upper level. Multilevel latent class (MLC) modelling is proposed to partition a prediction focus at the patient level (accommodating casemix) and a causal inference focus at the provider level. The MLC model recovered a range of simulated Trust-level effects. Median recovered values were almost identical to simulated values for the binary Trust-level covariate, and we observed successful recovery of the continuous Trust-level covariate with at least 3 latent Trust classes. Credible intervals widen as the error variance increases. The MLC approach successfully partitioned modelling for prediction and for causal inference, addressing the potential conflict between these two distinct analytical strategies. This improves upon strategies which only adjust for differential selection. Patient-level variation and measurement uncertainty are accommodated within the latent classes.

It has been acknowledged that the use of ratio variables is problematic in regression analysis where ratios comprise common components and are present in both the dependent and independent constituents of the model, e.g. regression analyses of ratio variables with common population denominators. However, such ratio variables are ubiquitous in health research and their resultant mathematical coupling (MC) has not been investigated extensively in relation to studies in health geography, where common population denominators are frequently encountered. It is common, for instance, that area level measures for health outcomes are considered in relation to area levels of mortality and/or indicators of social deprivation, where the common denominator is the area population. Our study seeks to illustrate this issue and examines the implications of this form of MC from a causal inference perspective. We examine the impact of MC amongst ratio variables in regression analyses using simulated data based on the correlation structure and distribution of variables derived from the UK census. Specifically, we consider the proportion of limiting long term illness (LLTI) in relation to mortality rates and the Townsend Material Deprivation Score (constructed from percentages of the population of an area that are experiencing pre-defined properties). Simulations are constructed under the null hypothesis, i.e. there is no impact of an area measure of deprivation on the relationship between mortality and proportion of limiting long term illness. A causal framework is introduced utilising directed acyclic graphs (DAGs) to assess variable relationships and to suggest analytical strategies that mitigate any problems arising due to MC. We show that artefactual relationships arise in the regression analyses of composite proportions due to MC: area measures of deprivation appear to influence the relationship between LLTI and mortality when no such influence is present in the simulated data. A DAG aids comprehension of this issue from a causal inference perspective and, depending upon the exact nature of the MC present, the DAG can also point to alternative analytical strategies that are discussed. Mathematical coupling of ratio variables has been recognised and reported on in the past, yet its problems remain pervasive. By setting the problem within a causal framework, we provide a means by which the issue might be more readily identified. Furthermore, using DAGs can help direct alternative analytical strategies to remove bias due to MC from future health research.

Empirical researchers working with observational data have been slow to adopt modern statistical methods for causal inference, which remain poorly recognised among applied quantitative researchers. First introduced in 2010, DAGitty is a free web application (and R package) that enables empirical researchers to draw directed acyclic graphs (DAGs) and identify minimally-sufficient adjustment sets without explicit knowledge of graphical model theory. This review examines empirical research articles that have used DAGitty as an aid for analysing observational data. Articles citing ‘DAGitty’ published before 1 July 2016 were identified through searching Web of Science, Medline, Scopus, PubMed, and Google Scholar. Original articles describing the analysis of observational data were identified by inspecting the published manuscripts. Information on the use and presentation of DAGs and adjustment sets were extracted into a standardised table. Bibliographic details (including journal discipline) were obtained from Thompson-Reuter’s Journal Citations Reports. 124 original articles describing the analysis of observational data were identified from 151 unique articles citing DAGitty. Two (2%) were published in 2012, seven (6%) in 2013, 23 (19%) in 2014, 46 (37%) in 2015, and 46 (37%) in the first half of 2016. The first authors came from 18 countries, most commonly the USA (n=36, 29%), Germany (n=19, 15%), Australia (n=14, 11%), Sweden (n=12, 10%), the UK (n=10, 8%), and Denmark (n=6, 5%). The host journals represented 43 academic disciplines, most commonly ‘Public, environmental, and occupational health’ (n=29, 23%), ‘environmental studies’ (n=13,10%), ‘multidisciplinary sciences’ (n=11, 9%), ‘oncology’ (n=10, 8%), ‘nutrition and dietetics’ (n=9, 7%), and ‘immunology’ (n=8, 6%). 29 (23%) articles included a DAG in the manuscript, 41 (33%) in supplementary material, while 53 (44%) contained no DAG. DAGs varied greatly in scope from three-variable overviews to graphs with 30+variables. Very few DAGs were saturated, whether completely or in order of transit. At the extreme, some researchers omitted all arcs except those that were explicitly evidenced. Adjustment sets were often modified beyond minimally-sufficient set(s) by adding: competing exposures (for ‘improve precision‘), mediators (to ‘improve face validity‘), and interaction terms; or by removing variables using stepwise (p-value) methods or criteria for ‘minimum change‘. Use of DAGitty in empirical research is increasing exponentially. There is however huge variation in practice, with many choosing to blend DAG-based methods with more traditional/accepted approaches to model specification. Guidelines for ‘best practice’ should be developed and included in teaching material and/or journal guidelines.

Background: Controlling for maturational status and timing is crucial in lifecourse epidemiology. One popular non-invasive measure of maturity is the age at peak height velocity (PHV). There are several ways to estimate age at PHV, but it is unclear which of these to use in practice. Aim: To find the optimal approach for estimating age at PHV. Subjects and methods: Methods included the Preece & Baines non-linear growth model, multi-level models with fractional polynomials, SuperImposition by Translation And Rotation (SITAR) and functional data analysis. These were compared through a simulation study and using data from a large cohort of adolescent boys from the Christ’s Hospital School. Results: The SITAR model gave close to unbiased estimates of age at PHV, but convergence issues arose when measurement error was large. Preece & Baines achieved close to unbiased estimates, but shares similarity with the data generation model for our simulation study and was also computationally inefficient, taking 24 hours to fit the data from Christ’s Hospital School. Functional data analysis consistently converged, but had higher mean bias than SITAR. Almost all methods demonstrated strong correlations (r > 0.9) between true and estimated age at PHV. Conclusions: Both SITAR or the PBGM are useful models for adolescent growth and provide unbiased estimates of age at peak height velocity. Care should be taken as substantial bias and variance can occur with large measurement error.

Recent prospective studies have shown that dysregulation of the immune system may precede the development of B‐cell lymphomas (BCL) in immunocompetent individuals. However, to date, the studies were restricted to a few immune markers, which were considered separately. Using a nested case–control study within two European prospective cohorts, we measured plasma levels of 28 immune markers in samples collected a median of 6 years before diagnosis (range 2.01–15.97) in 268 incident cases of BCL (including multiple myeloma [MM]) and matched controls. Linear mixed models and partial least square analyses were used to analyze the association between levels of immune marker and the incidence of BCL and its main histological subtypes and to investigate potential biomarkers predictive of the time to diagnosis. Linear mixed model analyses identified associations linking lower levels of fibroblast growth factor‐2 (FGF‐2 p = 7.2 × 10−4) and transforming growth factor alpha (TGF‐α, p = 6.5 × 10−5) and BCL incidence. Analyses stratified by histological subtypes identified inverse associations for MM subtype including FGF‐2 (p = 7.8 × 10−7), TGF‐α (p = 4.08 × 10−5), fractalkine (p = 1.12 × 10−3), monocyte chemotactic protein‐3 (p = 1.36 × 10−4), macrophage inflammatory protein 1‐alpha (p = 4.6 × 10−4) and vascular endothelial growth factor (p = 4.23 × 10−5). Our results also provided marginal support for already reported associations between chemokines and diffuse large BCL (DLBCL) and cytokines and chronic lymphocytic leukemia (CLL). Case‐only analyses showed that Granulocyte‐macrophage colony stimulating factor levels were consistently higher closer to diagnosis, which provides further evidence of its role in tumor progression. In conclusion, our study suggests a role of growth‐factors in the incidence of MM and of chemokine and cytokine regulation in DLBCL and CLL.

Many studies in dental research are based on repeated measurements of several continuous variables. Statistical analyses of such data require advanced methods to explore the complexity of information within the data. Currently, the most frequently adopted approach is to undertake multiple univariate tests. Occasionally, more advanced and sophisticated statistical methodologies, such as multilevel modelling and generalized estimating equation, have been used. In the last decade, a novel statistical methodology known as latent growth curve modelling has been developed in the social sciences. Latent growth curve modelling can be considered a special application of structural equation modelling and is generally conducted using structural equation modelling software. Recent development of statistical theory shows that latent growth curve modelling is equivalent to multilevel modelling, and both approaches yield identical results. However, in some study designs latent growth curve modelling can provide a more flexible framework of statistical modelling than multilevel modelling and generalized estimating equation for longitudinal data. The aim of this article was to present a non‐technical introduction to latent growth curve modelling for dental researchers. The emphasis was on conceptual understanding, rather than mathematical rigor, so path diagrams were used for visual presentations of various statistical models. When properly applied, latent growth curve modelling has great potential to give new directions for future longitudinal dental research.

BACKGROUND: Some clinical outcome variables in periodontal research are mathematically coupled, and it is not feasible to include all the mathematically coupled variables in an ordinary least squares (OLS) regression analysis. The simplest solution to this problem is to drop at least one of the mathematically coupled variables. However, this solution is not satisfactory when the mathematically coupled variables have distinctive clinical implications. MATERIAL AND METHODS: Partial least squares (PLS) methods were used to analyse data from a study on guided tissue regeneration. Relationships between characteristics of baseline lesions and treatment outcomes after 1 year were analysed using PLS, and the results were compared with those from OLS regression. RESULTS: PLS analysis suggested that there were multiple dimensions in the characteristics of baseline lesion: vertical dimension was positively associated with probing pocket depth (PPD) reduction and clinical attachment level (CAL) gain, whilst horizontal dimension was negatively associated with the outcome. Baseline gingival recession had a negative association with PPD reduction but a small positive one with CAL gain. CONCLUSION: PLS analysis provides new insights into the relationships between baseline characteristics of infrabony defects and periodontal treatment outcomes. The hypothesis of multiple dimensions in baseline lesions needs to be validated by further analysis of different datasets.

Count data may possess an 'excess' of zeros relative to standard distributions. Zero-inflated Poisson (ZiP) or binomial (ZiB) and generic mixture models have been proposed to deal with such data. We consider biomedical count data with an excess number of zeros and seek to address the following: (i) do zero-inflated models need covariates in the distribution part to predict class membership; (ii) what model-fit criteria have clinical relevance to predicted counts; (iii) can very different model parameterizations have near-identical fit; and (iv) how could model selection and hence model interpretation be aided by considering data generation processes? We show that covariates in the distribution part of zero-inflated models are needed to predict class membership. A range of model-fit criteria should be considered, as consensus is rarely achieved, and considering predicted outcomes may be just as valuable as likelihood-based criteria. Zero-inflated and generic mixture models may be indistinguishable according to both likelihood-based model-fit criteria and predicted outcomes, in which case model differentiation, hence, model selection and interpretation, might be guided by the consideration of a priori data generation processes. Zero-inflated models reflect whether or not there are (or have been) risk differences in disease onset and disease progression, while generic mixture models identify sub-types of individuals with similar risks of disease onset and progression. One or both modelling strategies may be used, though a priori knowledge or clinical impression of data generation might help to distinguish between two or more parameterizations that exhibit similar fit and yield near-identical predicted counts.

BACKGROUND: Recent studies have found that postnatal catch-up growth might have a stronger impact than birth size on health in later life. Because growth is a continuing process, the challenge is to tease out the impact of body size at different critical phases. Ordinary least squares regression cannot differentiate the effects of birth size, growth, and current body size simultaneously, because growth is generally defined as the difference between birth size and current size, giving rise to perfect collinearity. METHODS: This paper aims to describe and justify the use of a novel approach, partial least squares, to estimate life course effects of body size z-scores on later blood pressure, using longitudinal data from a cohort of 960 Filipino boys. Body weight z-scores and changes in z-score weight were measured from birth to age 19 years, and systolic and diastolic blood pressures (BPs) were measured at age 19. RESULTS: In general, birth size had a very modest association with systolic BP. The early changes in z-scores between birth and the age of 2 years were positively associated with the outcome. Growth after age 8 had a larger effect than early growth, but the confidence intervals are generally large. For diastolic BP, the association was similar for early and later growth. Current body size had the strongest relationships with both outcomes. CONCLUSION: By creating weighted composites of perfectly collinear variables as components, partial least squares estimates the life course effects of body size on later health according to the correlations between body size and health outcomes.

BACKGROUND: General practitioners' remuneration is now linked directly to the scores attained in the Quality and Outcomes Framework (QOF). The success of this approach depends in part on designing a robust and clinically meaningful set of indicators. The aim of this study was to assess the extent to which measures of health observed in practice populations are correlated with their QOF scores, after accounting for the established associations between health outcomes and socio-demographics. METHODS: QOF data for the period April 2004 to March 2005 were obtained for all general practices in two English Primary Care Trusts. These data were linked to data for emergency hospital admissions (for asthma, cancer, chronic obstructive pulmonary disease, coronary hear disease, diabetes, stroke and all other conditions) and all cause mortality for the period September 2004 to August 2005. Multilevel logistic regression models explored the association between health outcomes (hospital admission and death) and practice QOF scores (clinical, additional services and organisational domains), age, sex and socio-economic deprivation. RESULTS: Higher clinical domain scores were generally associated with lower admission rates and this was significant for cancer and other conditions in PCT 2. Higher scores in the additional services domain were associated with higher admission rates, significantly so for asthma, CHD, stroke and other conditions in PCT 1 and cancer in PCT 2. Little association was observed between the organisational domain scores and admissions. The relationship between the QOF variables and mortality was less clear. Being female was associated with fewer admissions for cancer and CHD and lower mortality rates. Increasing age was mainly associated with an increased number of events. Increasing deprivation was associated with higher admission rates for all conditions and with higher mortality rates. CONCLUSION: The associations between QOF scores and emergency admissions and mortality were small and inconsistent, whilst the impact of socio-economic deprivation on the outcomes was much stronger. These results have implications for the use of target-based remuneration of general practitioners and emphasise the need to tackle inequalities and improve the health of disadvantaged groups and the population as a whole.

A number of investigations have looked at psychological changes occurring in association with orthognathic treatment. However, most of these studies have used a presurgery questionnaire as the baseline measurement. There is little data relating to the true baseline, i.e. that prior to any active treatment. Until this aspect is investigated, it is not possible to assume that pre-surgery is an acceptable baseline. This questionnaire based study aimed to assess changes in six psychological outcome measures between T1 (prior to any active treatment) and T2 (following pre-surgical orthodontics/prior to surgery). The outcome variables were: state anxiety, trait anxiety, depression, self-esteem, body image, and facial body image. Sixty-two patients (39 females and 23 males) completed both questionnaires. The results showed that intervention, in the form of orthodontic treatment, had a minimal effect on the chosen psychometric outcome variables. There was a significant reduction in satisfaction with body image amongst patients who initially reported mild to moderate dental/facial problems, whilst a moderate increase in satisfaction occurred in those patients reporting severe conditions initially. Also of note were significant increases in state anxiety amongst older patients whilst trait anxiety showed greater increases in females than males.

Objective: To describe the prevalence of dental pain and other oral health problems among a young Syrian adult population and to identify socio-demographic factors associated with these experiences. Subjects: An industrial sample of 400 men and women aged 18-34 years from Damascus, Syria. Methods: Study participants were interviewed about their experience of oral health problems in the previous year (1998). Socio-demographic information was collected. Results: 93% (369) of the interviews were completed. The prevalence of oral health problems was high, with 96% (353) of respondents claiming that they experienced one or more problems in the previous year. Two thirds of participants (65%, 239) claimed they had dental pain in the previous year. Analysis revealed that dental pain experience was significantly associated with age and gender. Analysis considering all factors revealed that the odds of experiencing dental pain were higher amongst the younger age group (18-24 compared to 25-34 year-olds). Conclusion: The prevalence of oral health problems was high among the population studied. Socio-demographic variations in experience of dental pain were apparent, with young men of lower education having the greatest odds of dental pain experience in the previous year.

Genes encoding resistance to mercury and to antibiotics are often carried on the same mobile genetic element and so it is possible that mercury-containing dental materials may select for bacteria resistant to mercury and to antibiotics. The main aim of this study was to determine whether the prevalence of Hg-resistant oral bacteria was greater in children with mercury amalgam fillings than in those without. A secondary aim was to determine whether the Hg-resistant isolates were also antibiotic resistant. Bacteria in dental plaque and saliva from 41 children with amalgam fillings and 42 children without such fillings were screened for mercury resistance by cultivation on a HgCl2-containing medium. Surviving organisms were identified and their susceptibility to mercury and to several antibiotics was determined. Seventy-eight per cent and 74% of children in the amalgam group and amalgam-free group, respectively, harboured Hg-resistant bacteria; this difference was not statistically significant. Nor was there any significant difference between the groups in terms of the proportions of Hg-resistant bacteria in the oral microflora of the children. Of Hg-resistant bacteria, 88% and 92% from the amalgam group and the amalgam-free group, respectively, were streptococci; 41% and 33% were resistant to at least one antibiotic, most frequently tetracycline. The results of this study show that there was no significant difference between children with amalgam fillings and those without such fillings with regard to the prevalence, or the proportion, of Hg-resistant bacteria in their oral microflora. The study also found that Hg-resistant bacteria were common in children regardless of whether or not they had amalgam fillings and that many of these organisms were also resistant to antibiotics.

Previous studies using correlation or regression analysis have showed that treatment effects measured by change in clinical parameters are often associated with baseline values of the same parameters. These studies, however, have a methodological weakness. Correlation/regression between baseline measures and the derived change variable invalidates the statistical procedures of testing the null hypothesis: that the coefficient of correlation/regression is zero. This is due to the phenomenon of mathematical coupling. To investigate the impact that this has on the observed correlation/regression coefficient when in reality this is zero, we used random simulations of hypothetical data to model the treatment of periodontal pockets. Results showed a strong probability of obtaining statistically significant correlation/regression coefficients. To separate this artificial effect of mathematical coupling from the true underlying biological relationship, one must apply analytical strategies to re-evaluate previous evidence within the periodontal literature.

Whilst associations between inequalities in healthcare utilisation and socio-economic deprivation are well established in the UK it is argued that deprivation indices, such as the Townsend index, remain insensitive to rural/urban differences. This study examines how Townsend and its components differ in their association with healthcare utilisation across the rural/urban spectrum of a large health region. Our research was carried out in the West Midlands National Health Service region (population 5.3 million), comprising of a similar geographical population diversity to that of the United Kingdom (UK) using Hospital Episode Statistics (1994/5-1998/9) and 1991 census socio-demographic data. Retrospective multilevel multivariate models compare three ward-level healthcare utilisation measures (standardised episode-, admission-, and bed-rates) in relation to the Townsend index of material deprivation, its components, and four rural/urban characteristics (population density, population potential, electoral ward area and perimeter size). The associations between outcomes and Townsend were generally not attenuated by the rural/urban characteristics. The constituent component of car-ownership was similarly unperturbed, whereas population potential significantly perturbed the home-ownership model and overcrowding was significantly perturbed by all four rural/urban characteristics considered. A deprivation index may encapsulate different meanings to that of its components when used to assess variations in healthcare utilisation. Constituent components may yield considerable perturbation in relation to healthcare utilisation across the rural/urban spectrum, whilst the composite measure does not. In particular, and contrary to anecdotal opinion, car-ownership and unemployment (as recorded in the 1991 UK census) exhibited a stable relationship across different rural/urban areas with respect to healthcare utilisation. © 2003 Elsevier Science Ltd. All rights reserved.

PURPOSE: This study investigated the changes that might occur in passive tactile sensibility during a period of 3 months following Implant placement in a group of edentulous subjects treated with dental implants. The effect of changing the velocity of force application on passive tactile sensibility was also investigated. MATERIALS AND METHODS: Five edentulous subjects who had been treated (as a part of an immediate loading study) with 2 or more Nobel Biocare dental implants in the anterior mandible were studied. Pushing forces were applied directly and perpendicular to the long axes of the abutments until the subjects felt the first sensation of pressure, using a computer-controlled, custom-made device. The force was measured with an integral transducer. The applied force had a ramped staircase pattern, which was used at 2 different tip velocities. The measurements were taken on 4 occasions: 1, 2, 4, and 12 weeks after fitting the abutments. RESULTS: Statistical analysis, using multilevel modeling, demonstrated that there was a significant decrease In the tactile threshold over successive weeks following implant placement. It also demonstrated that high velocity exhibited a higher threshold than low velocity. DISCUSSION AND CONCLUSION: It could be concluded that there was a significant increase In passive tactile sensibility during the healing phase following implant placement.

OBJECTIVES: Previous periodontal literature has shown that there is a strong relationship between treatment effects, such as guided tissue regeneration (GTR), and baseline disease severity. However, relating change to baseline values using correlation or regression is methodologically flawed due to mathematical coupling, where the statistical procedure of testing the null hypothesis-that the coefficient of correlation or slope of regression is equal to zero-becomes erroneous. The aim of this study is to investigate if baseline disease severity is genuinely associated with the treatment outcome of intrabony defects using GTR after adjustment for mathematical coupling. In particular, we seek to demonstrate the potential effect that mathematical coupling has in distorting the results from the statistical analyses of trials of dental treatment, using data from the periodontal literature on GTR. The erroneous results arising from the use of simple correlation and regression techniques to analyse this association will be demonstrated, also the methodological flaw where the statistical procedure tests the null hypothesis-that the coefficient of correlation or the slope of regression is equal to zero. METHODS: Three main periodontal journals were electronically and manually searched to extract the data for the clinical outcomes of pocket probing depth (PPD) and lifetime cumulative attachment loss (LCAL) in the studies using GTR. The relationship between clinical outcomes and baseline measurements were reanalysed using Oldham's method and the variance ratio test. RESULTS: The results of these analyses were compared with those from the papers where the authors used the standard approach of correlation or regression. This shows that mathematical coupling caused spurious correlations between baseline disease severity and treatment effect. Ten out of 12 studies for PPD and nine out of 14 for LCAL initially claimed a significant positive relationship; after using either of the more appropriate statistical methods of adjustment, only three correlations in each group of studies remained significant. CONCLUSIONS: Previous evidence suggesting an association between baseline disease severity and treatment effect for GTR is challenged and therefore needs to be critically reviewed. All future clinical research should avoid using mathematically coupled data in correlation or regression analysis. In seeking to examine the bivariate association between baseline and subsequent change, Oldham's method is recommended.

The Index of Orthodontic Treatment Need (IOTN) has been used in dental epidemiology and to prioritize orthodontic treatment. The aim of this paper was to use the aesthetic component (AC) of the IOTN to measure agreement between normative and perceived orthodontic need amongst school children. Three hundred and seventy‐eight children aged 11–14 years, enrolled in London UK state schools participated in this survey. The study focused on three ethnic groups: white, black and South Asian. Townsend deprivation scores suggested that the children were from areas of high socio‐economic deprivation. Logistic regression analysis was carried out for agreement between normative and perceived need at each threshold value. Perceived need for braces, ethnic background, social class and hours of television viewing were significant variables. Black pupils were significantly less likely to concur on normative and perceived need scores, tending to perceive less need for treatment than did the dentist. Subjects from lower social classes were significantly more likely to concur on normative and perceived need scores. In conclusion, the study showed that using the IOTN AC at various points along the scale, different influences play a significant role in agreement/disagreement between normative and perceived needs, indicating that patient–clinician agreement regards orthodontic treatment is sensitive to several cultural factors.

Summary.

Objectives.  To determine which variables were best related to the overall morbidity of a child undergoing dental general anaesthetic (GA) and then to use these variables to determine those factors that might influence the extent and severity of morbidity experienced by healthy children following dental GA.

Sample and methods.  Data were collected on anxiety, pain and morbidity, GA procedure and dental procedure from 121 children attending a day stay GA unit for dental treatment. Patients were interviewed preoperatively, postoperatively before discharge then four further times over the next 148 h. Data were analysed using multivariate regression.

Results.  Thirty‐one per cent of subjects had restorative work, 60% had at least one tooth extracted, 54% had a surgical procedure. Use of local analgesia reduced postoperative pain whilst an increase in the number of surgical procedures increased it. Increase in anaesthetic time was related to increased odds of feeling sleepy and nauseous, females were more likely to complain of sleepiness or weakness. Feelings of dizziness were increased if the patient was given local analgesia during the procedure.

Conclusions.  Pain following dental GA was the most prevalent and long lasting symptom of postoperative morbidity in this study. Reductions in operating time and improvement in pain control have the potential to reduce reported morbidity following dental GA.

STATEMENT OF PROBLEM: Edentulous patients treated with implant-supported prostheses have shown increased passive tactile sensibility compared with those using conventional complete dentures. This is thought to be due to the close mechanical coupling between the implant and bone via the osseointegrated interface, yet the phenomenon has received little attention. PURPOSE: The purpose of this study was to measure passive tactile sensibility in a group of edentulous subjects treated with dental implants, and to relate the measured sensibility to a range of factors thought to be of possible relevance, namely, patient age, gender, time since implant placement, implant length, and implant separation. MATERIAL AND METHODS: Twenty edentulous subjects successfully treated with 2 or more Nobel Biocare dental implants in the anterior mandible were studied. The inclusion criteria were : (1) age of less than 50 years, (2) a period of at least 12 months since implant placement, (3) implant length of at least 10 mm and of standard diameter (excluding narrow and wide platform designs), and (4) implant separation of at least 18 mm. Using a computer-controlled custom-made device, pushing forces (2.1, 2.4, 2.7, and 3.0 N/s) were applied directly and perpendicular to the long axes of the implant abutments until the subjects felt the first sensation of pressure. The magnitude of these forces was measured with an integral transducer. The applied force had a ramped staircase pattern, and force application rates were varied between 2.1 and 3.0 N/s. Multilevel modeling was used to analyze the collected data (alpha=.05). RESULTS: The threshold values of passive tactile sensibility ranged between 3.1 and 15.7 N (mean 10.9; SD 3.9). Analysis failed to show any significant association between passive tactile sensibility and the variables studied. CONCLUSION: Within the limitations of this study, which included a small sample size, no relationship was found between passive tactile sensibility associated with long-standing implants and any of the variables studied (age, gender, time since implant placement, implant length, and implant separation).

Objective. For over a century, statisticians have highlighted concerns about the inappropriate use of ratio variables in correlation and regression analysis. However, little attention has been paid to these concerns in medical and dental research. The use of ratio variables in correlation and regression analysis can give rise to spurious results due to inappropriate model specification and mathematical coupling, leading to serious misinterpretation of data and consequently to incorrect study conclusions. Methods. Data were reanalysed from two recently published articles: one on the efficacy of guided tissue regeneration on root coverage; the other a randomised controlled trial comparing three surgical approaches in the treatment of periodontal infrabony defects. The reanalysis was performed to examine whether the assumptions behind the correlation/regression analyses have been seriously violated in these two studies, and to see if the interpretation of results is tenable. Results. Use of ratio variables seriously violated the assumptions underpinning the statistical methods utilised in these two studies, and consequently the conclusions were substantially misleading. Recommendations made in these studies were not tenable. Conclusions. The reanalyses illustrate how the inappropriate use of ratio variables remains prevalent in dental research, leading to incorrect interpretation of the evidence. This emphasises the need for collaboration between clinicians and statisticians to avoid the risk of yielding erroneous conclusions from flawed statistical analyses. © 2003 Elsevier Ltd. All rights reserved.

Background: Statistical analyses of periodontal data that average site measurements to subject mean values are unable to explore the site‐specific nature of periodontal diseases. Multilevel modeling (MLM) overcomes this, taking hierarchical structure into account. MLM was used to investigate longitudinal relationships between the outcomes of lifetime cumulative attachment loss (LCAL) and probing depth (PD) in relation to potential risk factors for periodontal disease progression.

Methods: One hundred males (mean age 17 years) received a comprehensive periodontal examination at baseline and at 12 and 30 months. The resulting data were analyzed in two stages. In stage one (reported here), the absolute levels of disease were analyzed in relation to potential risk factors; in stage two (reported in a second paper), changes in disease patterns over time were analyzed in relation to the same risk factors. Each approach yielded substantially different insights.

Results: For absolute levels of disease, subject‐level risk factors (covariates) had limited prediction for LCAL/PD throughout the 30‐ month observation period. Tooth position demonstrated a near linear relationship for both outcomes, with disease increasing from anterior to posterior teeth. Sites with subgingival calculus and bleeding on probing demonstrated more LCAL and PD, and supragingival calculus had an apparently protective effect. Covariates had more “explanatory power” for the variation in PD than for the variation in LCAL, suggesting that LCAL and PD might be generally associated with a different profile of covariates.

Conclusion: This study provides, for a relatively young cohort, considerable insights into the factors associated with early‐life periodontal disease and its progression at all levels of the natural hierarchy of sites within teeth within subjects. J Periodontol 2004;75:127‐136.

Background: The aim of this study was to investigate the longitudinal relationships between the outcome measurements of changes in lifetime cumulative attachment loss (cLCAL) and changes in probing depth (cPD) in relation to potential risk factors or other risk markers for periodontal disease progression from a cohort of 100 young males. In order to account for the hierarchical data structure, and to explore explicitly the site, tooth, and subject levels simultaneously, multilevel modeling was undertaken.

Methods: The analyses were undertaken in two parts. Within a previous article, the absolute levels of disease were analyzed in relation to potential risk factors; within this article, changes in disease are analyzed in relation to these factors. Each analytical approach yielded substantively different insights.

Results: Subject‐level risk factors had limited predictive value for cLCAL/cPD throughout the 30‐month observation period. Tooth position demonstrated a near linear relationship for both outcomes, with disease increasing from anterior to posterior teeth. Supragingival plaque had no significant effect on cLCAL/ cPD, while subgingival calculus and bleeding on probing were negatively associated with cLCAL/cPD. In contrast to the outcomes LCAL/PD, supragingival calculus had no significant protective effect on cLCAL/cPD. There was no significant influence of smoking in this cohort.

Conclusions: This study provides, for a relatively young cohort, considerable insights into the factors associated with longitudinal patterns of early‐life periodontal disease at all levels of the natural hierarchy of sites within teeth within subjects. Furthermore, it is demonstrated how multilevel modeling can provide considerable insight into some of the inconsistencies and controversies found in the previous periodontal literature. J Periodontol 2004;75:137‐145.

Background: Previous studies have shown that the use of index teeth may underestimate the prevalence of chronic periodontitis in adults. However, there is little information on the effect of using index teeth to estimate the prevalence of early periodontitis in younger adults and the effect this may have on planning treatment needs and health care resources. The aim of this study was to compare full mouth examination with partial examination using index teeth in a group of young British males.

Methods: One hundred subjects aged between 16 and 20 years (mean 17 years) on entry to the study were examined at baseline, 12 months later, and 30 months later. Lifetime cumulative attachment loss (LCAL) ≥1 mm was measured on the mesio‐buccal, disto‐buccal, mesio‐lingual, and disto‐lingual surfaces of all teeth, excluding third molars. All data were entered into a database. The indices used to express LCAL were prevalence, defined as the percentage of subjects with LCAL ≥1 mm, 2 mm, or 3 mm, and extent, defined as the percentage of sites with LCAL ≥1 mm, 2 mm, or 3 mm. Two sets of index teeth were chosen to compare with full mouth recordings, Ramfjord index teeth and the Periodontal Index for Treatment (PIT) teeth.

Results: The prevalence of LCAL ≥1 mm was similar (approaching 100%) for the full mouth and both partial mouth recordings. However, as LCAL increased from a minimum of 1 to 3 mm, partial mouth recording resulted in an underestimation of the prevalence of disease. LCAL ≥2 mm was underestimated by up to 22% and LCAL ≥3 mm by up to 36%. The extent of LCAL was less affected by partial mouth recording, in that the percentage of sites with no sign of early attachment loss was underestimated by up to 11%. However, the percentage of sites with LCAL ≥1 mm and 2 mm were overestimated by 11% and, 7% respectively.

Conclusions: These data indicate that the use of index teeth in epidemiological studies which include young adults may result in an underestimation of the prevalence of early periodontitis and an overestimation of the extent. J Periodontol 2001;72:140‐145.

The four models proposed for exploring the foetal origins of adult disease (FOAD) hypothesis use the product term between size at birth and current size to determine the relative importance of pre- and post-natal growth on disease in later life. This is a common approach for testing the interaction between an exposure (in this instance size at birth) and an effect modifier (in this instance current size)-incorporating the product term obtained by multiplying the exposure and effect modifier variables within a statistical regression model. This study examines the mathematical basis for this approach and uses computer simulations to demonstrate two potential statistical flaws that might generate misleading findings. The first of these is that the expected value of the partial regression coefficient for the product term (between exposure and effect modifier) will be zero when the outcome, exposure and effect modifier are all continuously distributed and follow a multivariate normal distribution. This is because testing the product interaction term amounts to testing for multivariate normality among the three variables, irrespective of the pair-wise correlations amongst them. The second flaw is that it is possible to generate a statistically significant interaction between exposure and effect modifier, even when none exists, simply by categorising either or both of these variables. These flaws pose a serious challenge to the four models approach proposed for exploring the FOAD hypothesis. The interaction between exposure and effect modifier variables should be interpreted with caution both here and elsewhere in epidemiological analyses. © 2007 Springer Science+Business Media B.V.

Background: Numerous studies have reported inverse associations between birth weight and a range of diseases in later life. These have led to the development of the 'foetal origins of adult disease hypothesis'. However, many such studies have only been able to demonstrate a statistically significant association between birth weight and disease in later life by adjusting for current size. This has been interpreted as evidence that the impact of low birth weight on subsequent disease is somehow dependent on subsequent weight gain, and has led to a broadening of the hypothesis into the 'developmental origins of health and disease'. Unfortunately, much of the epidemiological evidence used for both of these interpretations is prone to a statistical artefact known as the 'reversal paradox'. The aim of this paper is to illustrate why, using vector geometry. © 2006 Tu et al; licensee BioMed Central Ltd.

The statistical validity of the negative associations observed between birthweight and disease in later life has recently been questioned, because these associations might be due, in part, to inappropriate adjustment for current body size, creating a statistical artefact known as the 'reversal paradox'. The aim of this study was to explore the effect of adjusting for more than one measure of current body size on the association between birthweight and disease in later life using simulations and meta-analyses of empirical studies. The simulations examined the relation between birthweight and adult systolic blood pressure before and after adjusting for one, two or three measures of current body size by including current weight and subsequently adding body mass index and height in successive analytical models. Meta-analyses were then performed to compare the effect sizes observed among empirical studies reporting associations between birthweight and blood pressure before and after adjusting for one or two measures of current body size. The meta-analyses confirmed the results of the simulations - both showed that associations between birthweight and blood pressure tend to become increasingly negative following adjustment for current body size, and that this effect is enhanced after adjusting for additional measures of current body size.

The 1995 Calman–Hine plan outlined radical reform of the UK's cancer services with the aim of improving outcomes and reducing inequalities in NHS cancer care. Its main recommendation was to concentrate care into the hands of site-specialist, multi-disciplinary teams. This study aimed to determine if the implementation of Calman–Hine cancer teams was associated with improved processes and outcomes of care for colorectal cancer patients. The design included longitudinal survey of 13 colorectal cancer teams in Yorkshire and retrospective study of population-based data collected by the Northern and Yorkshire Cancer Registry and Information Service. The population was all colorectal cancer patients diagnosed and treated in Yorkshire between 1995 and 2000. The main outcome measures were: variations in the use of anterior resection and preoperative radiotherapy in rectal cancer, chemotherapy in Dukes stage C and D patients, and five-year survival. Using multilevel models, these outcomes were assessed in relation to measures of the extent of Calman–Hine implementation throughout the study period, namely: (i) each team's degree of adherence to the Manual of Cancer Service Standards (which outlines the specification of the ‘ideal’ colorectal cancer team) and (ii) the extent of site specialisation of each team's surgeons. Variation was observed in the extent to which the colorectal cancer teams in Yorkshire had conformed to the Calman–Hine recommendations. An increase in surgical site specialisation was associated with increased use of preoperative radiotherapy (OR=1.43, 95% CI=1.04–1.98, P<0.04) and anterior resection (OR=1.43, 95% CI=1.16–1.76, P<0.01) in rectal cancer patients. Increases in adherence to the Manual of Cancer Service Standards was associated with improved five-year survival after adjustment for the casemix factors of age, stage of disease, socioeconomic status and year of diagnosis, especially for colon cancer (HR=0.97, 95% CI=0.94–0.99 P<0.01). There was a similar trend of improved survival in relation to increased surgical site specialisation for rectal cancer, although the effect was not statistically significant (HR=0.93, 95% CI=0.84–1.03, P=0.15). In conclusion, the extent of implementation of the Calman–Hine report has been variable and its recommendations are associated with improvements in processes and outcomes of care for colorectal cancer patients.

Recent studies suggest that measurement error in food frequency questionnaires includes a person-specific component correlated with that of other self-reported dietary assessments. Use of biomarkers has been recommended to adequately calibrate dietary assessment tools for unbiased estimation of associations between diet and disease. Data on biomarkers of intake are often collected only in small subsamples, because collection of biomarker data can be expensive and inconvenient for participants. In this paper, the authors propose a novel approach using itemized household grocery till receipts to calibrate dietary assessment. Till receipts are not self-recorded and the data obtained from them are not subject to person-specific bias, but the data need to be supported by self-completed diaries for foods eaten away from home. Till receipts may also prove cheaper to collect in larger samples. The authors discuss the many methodological challenges of using household-level data and discuss how till receipts might be used in practice, with or without the use of biomarkers.

BACKGROUND: The inverse association between birth weight and blood pressure may partly be the result of inappropriate adjustment for adult body size, but it remains unclear whether adjustment for adult height elicits this effect. AIM: The study investigated the impact of adjustment for adult height on the relationship between birth weight and blood pressure. METHODS: A systematic search of Medline from 1996 to 2006 was conducted using the terms 'birth weight', 'blood pressure' and 'hypertension', and any papers containing linear regression analyses of blood pressure on birth weight for populations with an average age of 25+ were eligible for inclusion in comparative meta-analyses. RESULTS: None of the 30 studies identified had published regression coefficients for blood pressure on birth weight before and after adjustment for adult height, and only two studies were found to adjust for adult height at all. Data from these studies were obtained, and it was found that adjustment for height made the association between birth weight and systolic blood pressure (SBP) more negative in one study but less negative in the other. When compared with meta-analyses of comparable models, it was found that both studies were substantially different from the combined estimate of the relationship between birth weight and SBP. CONCLUSIONS: Both the differences between the two selected studies and their differences from the combined estimates obtained by meta-analysis are likely to be due to differences in the age of the participants. The relationship between birth weight and SBP tended to become more strongly inverse in studies with older participants. Additionally, the correlations between height and SBP were found to change from positive to negative with increasing age, which explained the differential impact of adjustment for height in the two selected studies. It therefore appears that adjustment for height may have little effect for older participants, but more so for younger participants.

The relation between initial disease status and subsequent change following treatment has attracted great interest in clinical research. However, statisticians have repeatedly warned against correlating/regressing change with baseline due to two methodological concerns known as mathematical coupling and regression to the mean. Oldham's method and Blomqvist's formula are the two most often adopted methods to rectify these problems. The aims of this article are to review briefly the proposed solutions in the statistical and psychological literature, and to clarify the popular misconception that Blomqvist's formula is superior to Oldham's method. We argue that this misconception is due to a failure to recognize that the heterogeneity of individual responses to treatment is a source of regression to the mean in the analysis of the relation between change and initial value. Furthermore, we demonstrate how each method actually answers different research questions, and how confusion arises when this is not always understood.Copyright © 2006 John Wiley & Sons, Ltd.

In a large population-based series of invasive breast cancer patients, we investigated socioeconomic background (SEB) in relation to (a) stage at diagnosis; (b) treatment pattern; and (c) 5-year survival. Women diagnosed during 1998–2000 and resident in the Northern and Yorkshire regions of England were identified from the cancer registry database (N=12 768). Logistic regression and Cox proportional hazards analyses were used to estimate associations between SEB (defined using the Townsend Index for area of residence) and tumour stage, treatment pattern, and survival. Living in a more deprived area was associated with increased likelihood of being diagnosed with stage III or IV disease (age-adjusted odds ratio (OR) 1.13; 95% confidence interval (CI) 1.08–1.18 per quartile increase in Townsend score), and, after adjustment for age and stage, reduced odds of having surgery (OR 0.85; 95% CI 0.80–0.91), and receiving radiotherapy (OR 0.91; 95% CI 0.88–0.94). Amongst patients receiving surgery, those living in more deprived areas had decreased odds of having breast conserving surgery (age plus stage-adjusted OR 0.92; 95% CI 0.89–0.95). Living in a more deprived area was also associated with increased mortality (age- plus stage-adjusted hazard ratio 1.08; 95% CI 1.05–1.11). These effects may operate through several pathways, such as later presentation leading to advanced disease.

Abstract

Background/Aim: Tobacco use reduces the effect of non‐surgical periodontal therapy. Host‐modulation with low‐dose doxycycline (LDD) might favour repair and promote an improved treatment response. The aim of this study was to investigate the effect of LDD in smokers on non‐surgical periodontal therapy.

Material and Methods: This was a parallel arm, randomized, identical placebo‐controlled trial with masking of examiner, care‐giver, participant and statistician and 6 months of follow‐up. Patients received non‐surgical therapy and 3 months of test or control drug. Statistical analysis used both conventional methods and multilevel modelling.

Results: Eighteen control and 16 test patients completed the study. The velocity of change was statistically greater for the test group for clinical attachment level −0.19 mm/month (95% CI=−0.34, 0.04; p=0.012) and probing depth 0.30 mm/month (95% CI=−0.42, −0.17; p<0.001). However, no differences were observed for absolute change in clinical or biochemical markers at 6 months.

Conclusions: This study does not provide evidence of a benefit of using LDD as an adjunct to non‐surgical periodontal therapy in smokers.

Background: In many laboratory-based high throughput microarray experiments, there are very few replicates of gene expression levels. Thus, estimates of gene variances are inaccurate. Visual inspection of graphical summaries of these data usually reveals that heteroscedasticity is present, and the standard approach to address this is to take a log2 transformation. In such circumstances, it is then common to assume that gene variability is constant when an analysis of these data is undertaken. However, this is perhaps too stringent an assumption. More careful inspection reveals that the simple log2 transformation does not remove the problem of heteroscedasticity. An alternative strategy is to assume independent gene-specific variances; although again this is problematic as variance estimates based on few replications are highly unstable. More meaningful and reliable comparisons of gene expression might be achieved, for different conditions or different tissue samples, where the test statistics are based on accurate estimates of gene variability; a crucial step in the identification of differentially expressed genes. Results: We propose a Bayesian mixture model, which classifies genes according to similarity in their variance. The result is that genes in the same latent class share the similar variance, estimated from a larger number of replicates than purely those per gene, i.e. the total of all replicates of all genes in the same latent class. An example dataset, consisting of 9216 genes with four replicates per condition, resulted in four latent classes based on their similarity of the variance. Conclusion: The mixture variance model provides a realistic and flexible estimate for the variance of gene expression data under limited replicates. We believe that in using the latent class variances, estimated from a larger number of genes in each derived latent group, the p-values obtained are more robust than either using a constant gene or gene-specific variance estimate. © 2007 Manda et al; licensee BioMed Central Ltd.

Abstract

Inferring haplotypes from genotype data is commonly undertaken in population genetic association studies. Within such studies the importance of accounting for uncertainty in the inference of haplotypes is well recognised. We investigate the effectiveness of correcting for uncertainty using simple methods based on the output provided by the PHASE haplotype inference methodology. In case‐control analyses investigating non‐Hodgkin lymphoma and haplotypes associated with immune regulation we find little effect of making adjustment for uncertainty in inferred haplotypes. Using simulation we introduce a higher degree of haplotype uncertainty than was present in our study data. The simulation represents two genetic loci, physically close on a chromosome, forming haplotypes. Considering a range of allele frequencies, degrees of linkage between the loci, and frequency of missing genotype data, we detail the characteristics of genetic regions which may be susceptible to the influence of haplotype uncertainty. Within our evaluation we find that bias is avoided by considering haplotype probabilities or using multiple imputation, provided that for each of these methods haplotypes are inferred separately for case and control populations; furthermore using multiple imputation provides the facility to incorporate haplotype uncertainty in the estimation of confidence intervals. We discuss the implications of our findings within the context of the complexity of haplotype inference for larger marker rich regions as would typically be encountered in genetic analyses. Genet. Epidemiol. 2007. © 2007 Wiley‐Liss, Inc.

Abstract

In linear mixed models the influence of covariates is restricted to a strictly parametric form. With the rise of semi‐ and non‐parametric regression also the mixed model has been expanded to allow for additive predictors. The common approach uses the representation of additive models as mixed models. An alternative approach that is proposed in the present paper is likelihood based boosting. Boosting originates in the machine learning community where it has been proposed as a technique to improve classification procedures by combining estimates with reweighted observations. Likelihood based boosting is a general method which may be seen as an extension of L2 boost. In additive mixed models the advantage of boosting techniques in the form of componentwise boosting is that it is suitable for high dimensional settings where many explanatory variables are present. It allows to fit additive models for many covariates with implicit selection of relevant variables and automatic selection of smoothing parameters. Moreover, boosting techniques may be used to incorporate the subject‐specific variation of smooth influence functions by specifying ‘random slopes’ on smooth effects. This results in flexible semiparametric mixed models which are appropriate in cases where a simple random intercept is unable to capture the variation of effects across subjects. Copyright © 2006 John Wiley & Sons, Ltd.

Abstract

The relation between initial disease status and subsequent change following treatment has attracted great interest in clinical research. However, statisticians have repeatedly warned against correlating/regressing change with baseline due to two methodological concerns known as mathematical coupling and regression to the mean. Oldham's method and Blomqvist's formula are the two most often adopted methods to rectify these problems. The aims of this article are to review briefly the proposed solutions in the statistical and psychological literature, and to clarify the popular misconception that Blomqvist's formula is superior to Oldham's method. We argue that this misconception is due to a failure to recognize that the heterogeneity of individual responses to treatment is a source of regression to the mean in the analysis of the relation between change and initial value. Furthermore, we demonstrate how each method actually answers different research questions, and how confusion arises when this is not always understood. Copyright © 2006 John Wiley & Sons, Ltd.

Background. The achievement of desirable hemoglobin (Hb) levels in renal anemia treated with epoetins is often incomplete and subject to much variation of outcome values and applied dosage. The further development of clinical decision support for renal anemia requires the characterization of patient responses and an analysis of the dynamics of the dose and response variables. Methods. In this methodological paper, the extended data of a randomized controlled clinical trial comparing two epoetins were examined by functional data analysis, in order to describe and analyze the patterns of treatment response. Results. The description of the trajectory of Hb values in each patient as a mathematical function did allow the characterization of individual responses, with a wide variety of patterns being revealed. An analysis of the degree of system control in the management of the anemia was then possible through phase plotting. The analysis also allowed an expression of the dynamic characteristics of the entire experimental system, analyzed in summary group waveforms with standard statistical properties. Lastly, a quantification of the notional instability of patient responses enabled the determination of a subset of patients for whom control might be improved in a modified management system. Conclusion. Functional data analysis provided a basis for further characterization and experimental study of the control of renal anemia.

Objective: To use data from the Glasgow Alumni Cohort to investigate whether oral health in young adulthood is independently associated with later life cardiovascular disease (CVD) and cancer mortality.

Methods and results: Of the original cohort (n = 15 322), 12 631 subjects were traced through the National Health Service Central Register. Of these, 9569 men and 2654 women were 30 years or younger at baseline. During up to 57 years of follow-up, 1432 deaths occurred among subjects with complete data, including 509 deaths from CVD and 549 from cancer. After adjusting for potential confounders, no substantial association was found between the number of missing teeth (as a continuous variable) and all-cause mortality (hazard ratio (HR) for each extra missing tooth  = 1.01; 95% confidence interval (CI) 1.00 to 1.02), CVD mortality (HR = 1.01; 95% CI 0.99 to 1.03) or cancer mortality (HR = 1.00; 95% CI 0.98 to 1.02). When the number of missing teeth was treated as a categorical variable, there was evidence that students with nine or more missing teeth at baseline had an increased risk of CVD (HR = 1.35; 95% CI 1.03 to 1.77) compared with those with fewer than five missing teeth. When the number of missing teeth was transformed using fractional polynomials, there seemed to be a non-linear relation between missing teeth and CVD mortality.

Conclusions: Although some evidence was found to support the relation between tooth loss and CVD mortality, causal mechanisms underlying this association remain uncertain.

OBJECTIVE: To examine the changes over a decade in caries experience amongst children aged 4-5 years living in a deprived multiethnic community in the United Kingdom. DESIGN: Cross-sectional surveys. SETTING: Schools and nurseries in the Old Trafford area, Manchester, England, 1989, 1990, 1991 and 1998. MAIN OUTCOME MEASURES: Mean dmft, oral cleanliness and proportion of children with rampant caries. RESULTS: The unadjusted Odds Ratio for caries free children examined in 1998 compared with children examined prior to 1998, was only significant amongst the white group. White children examined in 1998 were over three times more likely to be caries free than white children examined previously. South Asian children whose mothers were non English speaking examined in 1998 were almost twice as likely to have good/fair oral cleanliness than those examined prior to 1998. Moreover, South Asian children whose mothers were non-English speaking in 1998 were over three times more likely not to have rampant caries than their counterparts in the earlier years. CONCLUSION: There were significant improvements in caries and oral health amongst white children over the decade, and although less marked these were mirrored amongst South Asian children.

This questionnaire-based study investigated the psychological profile of orthognathic patients prior to starting treatment and compared the findings with a control group of non-patients. Comparison of the data used multivariate multiple regression analysis where outcome variables and independent variables were studied simultaneously. Some differences were found in the psychological profile of the orthognathic patient. They displayed higher levels of state anxiety (P < 0.001), higher numbers of individuals in their social support network (P < 0.05), and lower body image and facial body image (P < 0.001). Self-esteem was also found to be lower, but only at borderline levels of significance (P = 0.052).

The orbital parameters of Cosmos 1603 (1984-106A) have been determined at 43 epochs from over 2900 observations, of which over 80% were supplied by the U.S. Naval Research Laboratory. Hewitt camera observations were available for 25 of the determinations. Orbital elements were determined between January and December 1987, during which time the satellite was close to 14:1 resonance. The satellite experienced the interesting property of being temporarily trapped with respect to a secondary resonance parameter due to the low air-drag in 1987. This effect gave rise to a quasi-secular increase in the eccentricity and libration of the secondary resonance variable. Analysis of the inclination and eccentricity yielded six lumped harmonic coefficients of order 14. Analysis of the mean motion yielded additional pairs of lumped harmonics of orders 14, 28 and 42, the 14th-order harmonics superseding those obtained from analysis of the inclination. The derived values were used to test the Goddard Earth Models, GEM-T1 and GEM-T2, at high order. © 1990.

A pilot survey in Tower Hamlets, London, indicated that many general practitioners (GPs) might not be recognizing abuse of elderly patients through lack of training. The survey was replicated on a large scale in Birmingham, to allow further analysis. 561 Birmingham GPs were mailed questionnaires and responses from 291 were analysed, providing data from 95% of the practices. The findings were similar to those in Tower Hamlets: just under half had diagnosed elder abuse in the previous year. Regression analysis of the combined data-sets (n = 363) indicated that the strongest factor predicting GP diagnosis of abuse was knowledge of 5 or more risk situations (odds ratio 6.77, 95% confidence interval 4.19, 10.93). The findings of these surveys suggest that research-based education and training would help GPs to become better at identifying and managing elder abuse.

AIM: To two contrasting centres, to describe the provision of dental general anaesthesia (DGA) for simple non-surgical extractions in terms of the type of treatment provided, including the number of primary and permanent teeth extracted, and the characteristics of child patients attending in terms of their age group and gender. DESIGN: Retrospective analysis of hospital records. METHOD: Data were drawn from records of services over a 12-month period in 1996/97 at: a) a London dental hospital (Centre 1), and b) in the community dental services in Rochdale, Lancashire (Centre 2). Information was collated and analysed using the SPSS statistical software package. RESULTS: The majority of patients at both centres were aged less than 9 years. Almost one third (31%) of those seen at Centre 1 were below 5 years of age, but fewer of this age group were treated at Centre 2. Children aged 9 years or less had an average of 5.4 (SD = 3.0) primary teeth extracted at Centre 1 and 3.0 (SD = 2.0) at Centre 2. For permanent teeth, an average of 3.2 (SD = 1.2) and 2.7 (SD = 1.4) were extracted at Centres 1 and 2 respectively. CONCLUSIONS: Both services were used primarily for the extraction of primary teeth although the services differed in the ages of patients who used them and in the numbers of teeth extracted. Numbers of patients attending the service at Centre 1 had declined over time but numbers of teeth extracted per child had increased.

The present study provides an assessment of the importance of tobacco and alcohol consumption upon the development of oral epithelial dysplasia (OED) in a large group of European patients. Data were collected in a case-control study based upon 630 patients with OED and 643 control subjects selected from UK dental hospital patients with oral disease not caused by tobacco or alcohol. Logistic regression was used to determine the association of several independent factors on the risk of OED. No relationship emerged between patient's gender, age or ethnicity and risk of OED. The regression model demonstrated a combined effect of tobacco smoking and alcohol drinking upon the risk of OED. Non-filter cigarette smoking was a significant predictor of OED, as was alcohol consumption, and the two habits compounded one another in the overall risk of disease. When both factors combined were included in the model through interaction terms, their individual impact was only moderately reduced, illustrating the importance of both factors in their own right. However, more detailed analysis of tobacco smoking habits revealed that the increased risk of OED from smoking was largely attributable to heavy smoking (20 cigarettes per day, OR = 4.38, 95% C.I. = 2.6, 7.2) especially non-filter cigarettes (OR = 1.95, 95% C.I. = 0.9, 4.0) relative to non-smoking. Both heavy smoking and non-filtered tobacco were higher risks for OED than alcohol consumption alone. Tobacco cessation was associated with a significant decline in risk of OED, the reduction being rapid and marked. For alcohol consumption the association with OED was considerably stronger for drinkers of fortified wines and spirits (OR = 3.75, 95% C.I. = 1.40, 10.05 and OR = 1.36, 95% C.I. = 0.76, 2.45, respectively). It is concluded that, while tobacco and alcohol synergistically influence the development of OED, exclusive tobacco consumption is more likely than exclusive alcohol consumption to give rise to OED. The risk of OED may thus be significantly reduced by behavioural changes such as moderation of tobacco and alcohol use.

The study retrospectively investigated variations in the use of secondary healthcare for head injury, particularly assault. A total of 25,300 emergency head-related admission were examined over a two-year period, of which 3756 were for assault. More males were admitted during summer and holiday periods, while there were fewer female patients with head injuries and the incidence varied less. The largest number of admissions was among men aged 15-44 and most assaults occurred at weekends. Females were more likely than males to die from all head injuries (OR=1.31) and violent head injuries (OR=2.38). Women (15+) stayed longer in hospital than males. Injuries among males are primarily associated with social occasions. Females experience head injuries all the year round suggesting that these injuries are the result of domestic violence. There are important demographic differences in numbers of patients and duration of hospital care required to treat these avoidable injuries.

Failure of orthodontic bands occurs most frequently at the band-cement interface, when conventional glass ionomer cements are used. Modification of the band surface may improve clinical performance by increasing the mechanical interlock at this junction. The aim of this prospective study was to compare the retention of micro-etched and untreated first molar orthodontic bands in a randomized, half-mouth trial. Seventy-nine patients had 304 bands cemented as part of routine fixed appliance therapy. The effect of micro-etching, patient age and gender, operator, molar crossbite, treatment mechanics, and arch on band failure was investigated. Failure rates and survival times were compared for each variable assessed. Micro-etched molar bands showed a significant reduction in clinical failure rate over untreated molar bands and an increase in mean survival time (P < 0.001). Of the other variables examined, only the presence of a molar crossbite had any significant effect on band failure (P = 0.004).

The National Health Service reforms have led to a growing need to determine the efficiency and cost effectiveness of health care provision. Resource utilization and the factors that influence it form a vital part of the information procured by commissioners of health care. Currently the most comprehensive source of routinely collected data is the Hospital Episode Statistics database. This database provides the most complete coverage of all areas of hospital inpatient services; these services account for up to 70% of the health care budget. Several studies have investigated the integrity of these data, but they have been concerned only with a subset of the data or with particular hospitals. This paper derives a method of data validation that can be applied either regionally or nationally to two key fields within the database: in this instance, consultant specialty and principal diagnosis. The process uses the Mantel-Haenszel odds ratio estimate to reveal large anomalies within either specialty of diagnostic coding. Within the West Midlands, it was generally observed that diagnosis codes are coded more consistently than consultant codes. However, problems occur due to incomplete or incorrect diagnostic entries and these anomalies are irregular across provider units. In contrast, anomalies within Korner specialty codes are associated predominantly with maternity, obstetrics, geriatric and general medicine. A large number of these anomalies are well understood within the context of local practice variation. Nevertheless, many anomalies remain within a wide range of other specialties where further investigation is warranted. Where systematic error is found, provider units should be notified in order to improve data consistency and enhance quality for future data users and health care planners.

This retrospective study was designed to assess the socioeconomic status of patients who underwent inpatient oral operations from 1989 to 1994 in the area covered by the West Midlands Regional Health Authority and to compare the distribution with those people treated by other specialties. A total of 4,926,438 hospital inpatient finished consultant episodes within 56 specialties that were recorded on the Hospital Episode Statistics database, of which 61,360 (1.25%) were dental (coded as oral surgery, restorative dentistry, orthodontics, and paediatric dentistry). The main outcome measure was the socioeconomic status of patients as assessed by the Townsend score (a measure of material deprivation that covers car ownership, home ownership, overcrowding, and unemployment). There was a highly significant correlation between the use of all inpatient services and social deprivation (R2 = 0.98, P < 0.001). This observation was consistent across all specialties except oral surgery, in which the correlation was reversed (R2 = 0.69, P < 0.001), indicating that patients who avail themselves of inpatient oral surgery are from a higher socioeconomic group. The application of a deprivation index to hospital episode data will enable purchasers and providers to measure more accurately the impact of their services on groups within the community.

STUDY OBJECTIVES: Measuring socio-economic deprivation is a major challenge usually addressed through the use of composite indices. This paper aims to clarify the technical details regarding composite index construction. The distribution of some variables, for example unemployment, varies over time, and these variations must be considered when composite indices are periodically re-evaluated. The process of normalisation is examined in detail and particular attention is paid to the importance of symmetry and skewness of the composite variable distributions. DESIGN: Four different solutions of the Townsend index of socioeconomic deprivation are compared to reveal the effects that differing transformation processes have on the meaning or interpretation of the final index values. Differences in the rank order and the relative separation between values are investigated. MAIN RESULTS: Constituent variables which have been transformed to yield a more symmetric distribution provide indices that behave similarly, irrespective of the actual transformation methods adopted. Normalisation is seen to be of less importance than the removal of variable skewness. Furthermore, the degree of success of the transformation in removing skewness has a major effect in determining the variation between the individual electoral ward scores. Constituent variables undergoing no transformation produce an index that is distorted by the inherent variable skewness, and this index is not consistent between re-evaluations, either temporally or spatially. CONCLUSIONS: Effective transformation of constituent variables should always be undertaken when generating a composite index. The most important aspect is the removal of variable skewness. There is no need for the transformed variables to be normally distributed, only symmetrically distributed, before standardisation. Even where additional parameter weights are to be applied, which significantly alter the final index, appropriate transformation procedures should be adopted for the purpose of consistency over time and between different geographical areas.

OBJECTIVES: To identify and quantify any relations between occupational exposure to cadmium oxide fume and mortalities from lung cancer and from chronic non-malignant diseases of the respiratory system. METHODS: The mortality experience of 347 copper cadmium alloy workers, 624 workers employed in the vicinity of copper cadmium alloy work (vicinity workers), and 521 iron and brass foundry workers (all men) was investigated for the period 1946-92. All subjects were first employed in these types of work in the period 1922-78 and for a minimum period of one year at one of two participating factories. Two analytical approaches were used, indirect standardisation and Poisson regression. RESULTS: Compared with the general population of England and Wales, mortality from lung cancer among copper cadmium alloy workers was close to expectation (observed deaths 18, expected deaths 17.8, standardised mortality ratio (SMR) 101, 95% confidence interval (95% CI) 60 to 159). A significant excess was shown for lung cancer among vicinity workers but not among iron and brass foundry workers (vicinity workers: observed 55, expected 34.3, SMR 160, 95% CI 121 to 209, P < 0.01; iron and brass foundry workers: observed 19, expected 17.8, SMR 107, 95% CI 64 to 167). Increased SMRs for non-malignant diseases of the respiratory system were shown for each of the three groups (alloy workers: observed 54, expected 23.5, SMR 230, 95% CI 172 to 300, P < 0.001; vicinity workers: observed 71, expected 43.0, SMR 165, 95% CI 129 to 208, P < 0.001; iron and brass foundry workers: observed 34, expected 17.1, SMR 199, 95% CI 137 to 278, P < 0.01). Work histories of the copper cadmium alloy workers were combined with independent assessments of cadmium exposures over time to develop individual estimates of cumulative exposure to cadmium; this being a time dependent variable. Poisson regression was used to investigate risks of lung cancer and risks of chronic non-malignant diseases of the respiratory system in relation to three levels of cumulative cadmium exposure (< 1600, 1600-4799, > or = 4800 micrograms.m-3.y). After adjustment for age, year of starting alloy work, factory, and time from starting alloy work, there was a significant positive trend (P < 0.01) between cumulative exposure to cadmium and risks of mortality from chronic non-malignant diseases of the respiratory system. Relative to a risk of unity for the lowest exposure category, risks were 4.54 (95% CI 1.96 to 10.51) for the middle exposure category and 4.74 (95% CI 1.81 to 12.43) for the highest exposure category. There was a non-significant negative trend between cumulative cadmium exposure and risks of mortality from lung cancer. Relative to a risk of unity for the lowest exposure category, risks were 0.85 (95% CI 0.27 to 2.68) for the middle exposure category and 0.81 (95% CI 0.18 to 3.73) for the highest exposure category. Similar findings were obtained when adjustment was made for age only. CONCLUSIONS: The findings are consistent with the hypothesis that exposure to cadmium oxide fume increases risks of mortality from chronic non-malignant diseases of the respiratory system. The findings do not support the hypothesis that exposure to cadmium oxide fume increases risks of mortality for lung cancer.

PURPOSE OF THE STUDY: This study set out to examine the prevalence of betel-quid chewing with or without the inclusion of tobacco and to provide baseline information on the experience, behaviour and perceptions of risk of first generation Bangladeshi adults. POPULATION STUDIED: A total of 127 households, which formed the sample base for this study, were selected by a systematic sampling procedure. Each household was visited by two interviewers and all adults resident in the home were asked to participate in the study. METHOD: A pretested questionnaire which explored the use, attitudes, knowledge of the health risk, and behaviour towards betel-quid chewing and the use of tobacco was used. FINDINGS: A total of 92% of males and 96% of females chewed betel-quid on a daily basis with 39% and 82% respectively including tobacco within their quid. There was a general acceptance of the use of tobacco, that is, males on the whole, smoking and females chewing tobacco. The perception of health risk, with regard to tobacco chewing, was low. CONCLUSION: The general acceptance of tobacco use and low perceived health risk in those engaged in tobacco chewing is a major concern that health professionals involved in health education need to address. General dental practitioners should be aware there is a high level of use in this community.

This study outlines the development and application of a methodology that allows direct comparison of the elements that make up in-patient healthcare across the West Midlands Regional Health Authority. The methodology produces a model which monitors the mechanism of resource utilization within the Region and the factors that influence it. The model shows how the uptake of healthcare varies by treatment specialty, between each provider unit, and across all purchasing authorities in the Region. The methodology accounts in all instances for demography and where appropriate for socio-economic deprivation. The novelty of this study is the generation of provider unit specialty costs that reflect both recorded activity and expenditure in the locality. The results of the investigation are made available to the purchasers to assess the levels of the services they are buying.

This study outlines the development and application of a methodology that allows direct comparison of the elements that make up in-patient healthcare across the West Midlands Regional Health Authority. The methodology produces a model which monitors the mechanism of resource utilization within the Region and the factors that influence it. The model shows how the uptake of healthcare varies by treatment specialty, between each provider unit, and across all purchasing authorities in the Region. The methodology accounts in all instances for demography and where appropriate for socio-economic deprivation. The novelty of this study is the generation of provider unit specialty costs that reflect both recorded activity and expenditure in the locality. The results of the investigation are made available to the purchasers to assess the levels of the services they are buying.

BACKGROUND: Good prognostic indicators for patients with squamous cell carcinoma of the lung would help to determine the most appropriate treatment for individual patients. METHODS: A silver colloid technique that shows interphase nucleolar organiser regions (AgNORs) has been applied to representative paraffin sections from 138 cases of squamous cell carcinoma of the bronchus treated by surgical resection of the primary tumour at East Birmingham Hospital in 1977. Of the 138 patients, 23 (17%) were alive 10 years after their operation. RESULTS: The mean (SD) AgNOR count per cell was significantly higher for all grades of malignancy (well differentiated 10.5 (2.6), moderately differentiated 10.7 (3.2), and poorly differentiated 12.7 (4.5)) than for normal pseudostratified columnar epithelium from non-affected areas (2.3 (0.78)). There was a trend for AgNOR counts to be higher in poorly differentiated tumours, but a wide range of AgNOR counts was observed in all histological grades. AgNOR counts did not predict clinical outcome, irrespective of the stage of the disease, and did not relate to DNA ploidy or the percentage of cells in the proliferation phase of the cell cycle. Nine of 47 patients (19%) with tumours classified as DNA diploid and eight of 63 patients (13%) with DNA aneuploid tumours were alive 10 years after operation. Principal component analysis identified the clinicopathological stage of disease as the variable best related to survival. The percentage of patients surviving 10 years was 30% for stage I, 20% for stage II, 10% for stage IIIa, 9% for stage IIIb, and none for stage IV. CONCLUSION: The AgNOR technique is not of prognostic value in postoperative patients with squamous cell carcinoma of the bronchus.

Abstract

Aim: The aims of this study were to monitor the prevalence and progression of lifetime cumulative attachment loss (LCAL) in a group of young British male military recruits over a 3‐year period, and to determine the relationship between signs of LCAL and selected periodontal variables.

Methods: 100 subjects, aged 16–20 years (mean 17 years) at baseline, were examined at 0 (baseline), 12 and 30 months. LCAL, probing depth, plaque, bleeding on probing, gingival colour and supra‐ and subgingival calculus were assessed on the mesio‐buccal, disto‐buccal, mesio‐lingual and disto‐lingual surfaces of all teeth present, excluding third molars. Data were analysed cross‐sectionally at each examination.

Results: Over the period of the study, the prevalence of LCAL 1 and 2 mm ranged from 95–100%, whereas LCAL 3 mm ranged from 40–47%. The extent of LCAL 1 mm ranged from 76–86%. However, the extent of LCAL 2 mm was dramatically lower (10.5–12.7%), and LCAL 3 mm was uncommon (0.5–0.9%). Examining the number of subjects according to the number of sites affected above a threshold, showed that a small number of subjects have a large number of sites above threshold. Using Pearson’s rank correlation coefficient a significant correlation (p<0.05) was found between LCAL and the periodontal variables of gingival bleeding and supra‐ and subgingival calculus.

Conclusions: These data suggest that the onset and progression of chronic periodontitis can be seen in young adults, and in this group gingival bleeding and supra‐ and subgingival calculus are the variables most strongly associated with early periodontitis.

AIM: To explore ethnic variations in social background of successful applicants to undergraduate United Kingdom medical and dental schools. METHOD: Retrospective analyses of University and College Admissions Services data on all students to commence study in pre-clinical medicine and dentistry, during the academic years 1994/5, 1995/6 and 1996/7. Analyses were undertaken for two categories of social class, namely higher (professional and intermediate) and lower (skilled non-manual, skilled manual, partly skilled, and unskilled) social class. RESULTS: Over 15 thousand students were accepted to study medicine and dentistry during the three-year study period, of which 80% were from high social class backgrounds. More medical (80.9%) students were from high social class backgrounds than dental (73.3%) students (OR = 1.54, 95% CI = 1.39, 1.70). Social class differences were observed, with a greater proportion of higher social class students amongst the white students than amongst the minority ethnic students (OR = 1.42, 95% CI = 1.30, 1.55). This was more marked in dentistry (OR = 1.48, 95% CI = 1.22, 1.79) than in medicine (OR = 1.35, 95% CI = 1.22, 1.49). More students from higher social class backgrounds were observed in medicine than in dentistry amongst the black (OR = 1.55, 95% CI = 0.59, 4.00), Indian (OR = 2.04, 95% CI = 1.58, 2.62) and white (OR = 1.44, 95% CI = 1.26, 1.64) groups. CONCLUSIONS: Significant inter-ethnic differences are observed in the social background of students entering medicine and dentistry. Dentistry accepted a greater proportion of students from lower social class backgrounds and from black and minority ethnic groups.

In this review, we present some ethical imperatives observed in this pandemic from a data ethics perspective. Our exposition connects recurrent ethical problems in the discipline, such as, privacy, surveillance, transparency, accountability, and trust, to broader societal concerns about equality, discrimination, and justice. We acknowledge data ethic’s role as significant to develop technological, inclusive, and pluralist societies.

Summary

A typical microarray experiment attempts to ascertain which genes display differential expression in different samples. We model the data by using a two-component mixture model and develop an empirical Bayesian thresholding procedure, which was originally introduced for thresholding wavelet coefficients, as an alternative to the existing methods for determining differential expression across thousands of genes. The method is built on sound theoretical properties and has easy computer implementation in the R statistical package. Furthermore, we consider improvements to the standard empirical Bayesian procedure when replication is present, to increase the robustness and reliability of the method. We provide an introduction to microarrays for those who are unfamilar with the field and the proposed procedure is demonstrated with applications to two-channel complementary DNA microarray experiments.

INTRODUCTION: The aim of this article was to encourage good practice in the statistical analyses of orthodontic research data. Our objective was to highlight the statistical problems caused by mathematical coupling (MC) in correlation and regression analyses. These statistical problems are among the most common pitfalls in orthodontic research when exploring associations among clinical variables. This article will show why these problems arise and how they can be avoided and overcome. METHODS: Four orthodontic journals were electronically and manually searched for articles that used correlation and regression analyses. Studies that seemed to suffer from MC in their statistical analyses were identified and carefully examined. RESULTS: Several examples from our search illustrate that MC in correlation and regression analyses can potentially cause misleading results. More appropriate statistical methods are available and should be used to eliminate confusing results and improve any subsequent interpretations. Because many clinical and radiographic variables used in orthodontic research are correlated due to direct or indirect MC, interpretation of studies in the literature needs to be cautious. CONCLUSIONS: Correlation and regression analyses are useful tools in orthodontic research when their assumptions and limitations are recognized. However, greater care is required in formulating research questions and experimental designs. It is prudent to seek statistical advice when orthodontic research involves complex data analyses.

Survival analysis methods are increasingly used in dental research to measure risk of tooth eruption and caries as well as life spans of amalgam restorations. Analyses have been extended to account for lack of independence in the data, which arises from the clustering of observations within units such as tooth-surfaces, teeth and subjects. There are various analytical strategies and modelling approaches now available to us in dealing with clustered dental data. In this article, the modelling strategy of Cox's proportional hazards regression is formulated using the counting process approach, which can easily be extended to include time-variant covariates as well as nested random frailty effects. A semi-parametric Bayesian method is presented for the analysis of the proposed model. The methodology is applied to an analysis of nested clustered data on life-span of amalgam restorations in the UK Royal Air Force. These data have previously been analysed using a non-Bayesian approach. The Gibbs sampler, a Markov chain Monte Carlo method, is used to generate samples from the marginal posterior distribution of the parameters of this Bayesian model.

AIM: The increasing popularity of randomized-controlled trials (RCTs) has raised the issue of their quality. Frequently overlooked are the differences between superiority and equivalence trials. The purpose of this study was to apply specific methodological criteria to evaluate the quality of active-control trials using studies that compared guided tissue regeneration (GTR) with enamel matrix derivatives (EMD). MATERIALS AND METHODS: Seven RCTs were identified in the literature. Standard methodological criteria and seven additional criteria for trials using active-control groups were used to evaluate the quality of the seven RCTs. RESULTS: Two trials were considered as superiority trials. The remaining five provided no clear statement of their research aim. However, two claimed that EMD and GTR were equally effective, because their results failed to show a significant difference between EMD and GTR. Most trials did not meet the majority of the design criteria. CONCLUSIONS: The general lack of compliance with quality criteria might place doubt on the value of these trials and may render any conclusions questionable. It is therefore important to distinguish clearly between superiority trials and equivalence trials, and to incorporate appropriate additional criteria in the design of future RCTs with active-control groups.

BACKGROUND: The effects of measurement error in epidemiological exposures and confounders on estimated effects of exposure are well described, but the effects on estimates for gene-environment interactions has received rather less attention. In particular, the effects of confounder measurement error on gene-environment interactions are unknown. METHODS: We investigate these effects using simulated data and illustrate our results with a practical example in nutrition epidemiology. RESULTS: We show that the interaction regression coefficient is unchanged by confounder measurement error under certain conditions, but biased by exposure measurement error. We also confirm that confounder measurement error can lead to estimated effects of exposure biased either towards or away from the null, depending on the correlation structure, with associated effects on type II errors. CONCLUSIONS: Whilst measurement error in confounders does not lead to bias in interaction coefficients, it may still lead to bias in the estimated effects of exposure. There may still be cost implications for epidemiological studies that need to calibrate all error-prone covariates against a valid reference, in addition to the exposure, to reduce the effects of confounder measurement error

Some researchers have recently questioned the validity of associations between birth weight and health in later life. They argue that these associations might be due in part to inappropriate statistical adjustment for variables on the causal pathway (such as current body size), which creates an artifactual statistical effect known as the "reversal paradox." Computer simulations were conducted for three hypothetical relations between birth weight and adult blood pressure. The authors examined the effect of statistically adjusting for different correlations between current weight and birth weight and between current weight and adult blood pressure to assess their impact on associations between birth weight and blood pressure. When there was no genuine relation between birth weight and blood pressure, adjustment for current weight created an inverse association whose size depended on the magnitude of the positive correlations between current weight and birth weight and between current weight and blood pressure. When there was a genuine inverse relation between birth weight and blood pressure, the association was exaggerated following adjustment for current weight, whereas a positive relation between birth weight and blood pressure could be reversed after adjusting for current weight. Thus, researchers must consider the reversal paradox when adjusting for variables that lie within causal pathways.

Multivariable analysis is a widely used statistical methodology for investigating associations amongst clinical variables. However, the problems of collinearity and multicollinearity, which can give rise to spurious results, have in the past frequently been disregarded in dental research. This article illustrates and explains the problems which may be encountered, in the hope of increasing awareness and understanding of these issues, thereby improving the quality of the statistical analyses undertaken in dental research. Three examples from different clinical dental specialities are used to demonstrate how to diagnose the problem of collinearity/multicollinearity in multiple regression analyses and to illustrate how collinearity/multicollinearity can seriously distort the model development process. Lack of awareness of these problems can give rise to misleading results and erroneous interpretations. Multivariable analysis is a useful tool for dental research, though only if its users thoroughly understand the assumptions and limitations of these methods. It would benefit evidence-based dentistry enormously if researchers were more aware of both the complexities involved in multiple regression when using these methods and of the need for expert statistical consultation in developing study design and selecting appropriate statistical methodologies.

BACKGROUND: The subgingival microflora in patients presenting concurrently with periodontitis and diabetes mellitus (DM) are poorly understood. While traditional putative periodontal pathogens are implicated, research involving other oral organisms; e.g., Capnocytophaga spp., is lacking. These organisms produce a range of bacterial enzymes relevant to periodontal breakdown. It is inferred that periodontal bacteria acquire systemic access through the ulcerated periodontal pocket surface; conclusive evidence supporting this notion is limited. The aims of this investigation were to: 1) quantify and identify Capnocytophaga spp. present in healthy and diseased sites in periodontitis patients with and without DM, and 2) isolate periodontal pathogens from these patients' blood. METHODS: Twenty-one DM-periodontitis and 25 periodontitis patients were recruited. Subgingival plaque was collected from three healthy and three diseased sites per subject. Capnocytophaga spp. and total (facultative and obligate) anaerobic counts from each site were estimated. Capnocytophaga spp. were identified using 16S rRNA polymerase chain reaction (PCR) restriction fragment length polymorphism (RFLP). Statistical analyses were performed using multilevel modeling. Blood samples were subjected to HbA(1c) estimation and bacterial culture. RESULTS: A total of 848 Capnocytophaga spp. were isolated and identified. Significantly higher numbers of Capnocytophaga spp. (P <0.001) and anaerobes (P <0.001) were present in diseased sites in DM-periodontitis subjects compared to healthy sites in non-DM-periodontitis and DM-periodontitis subjects. C. ochracea (and variant) and C. granulosa were the most prevalent species. Blood samples were negative for Capnocytophaga spp. CONCLUSIONS: Total mean counts for Capnocytophaga spp. were significantly higher in DM-periodontitis subjects versus non-DM-periodontitis (P = 0.025) and at diseased sites versus healthy sites (P <0.001). Analysis of individual species revealed that the outcome varied with site status and DM status.

Childhood acute lymphoblastic leukemia and diabetes mellitus, type 1, have common epidemiologic and etiologic features, including correlated international incidence and associations with infections. The authors examined whether the diseases' similar large-scale distributions are reflected in small geographic areas while also examining the influence of sociodemographic characteristics. Details of 299 children (0-14 years) with acute lymphoblastic leukemia and 1,551 children with diabetes diagnosed between 1986 and 1998 were extracted from two registers in Yorkshire, United Kingdom. Standardized incidence ratios across 532 electoral wards were compared using Poisson regression, confirming significant associations between population mixing and the geographic heterogeneity of both conditions. Bayesian methods analysis of spatial correlation between diseases by modeling a bivariate outcome based on their standardized incidence ratios was applied; spatial and heterogeneity components were included within a hierarchical random effects model. A positive correlation between diseases of 0.33 (95% credible interval: -0.20, 0.74) was observed, and this was reduced after control for population mixing (r = 0.18), population density (r = 0.14), and deprivation (r = 0.06). The Bayesian approach showed a modest but nonsignificant joint spatial correlation between diseases, only partially suggesting that the risk of both was associated within some electoral wards. With Bayesian methodology, population mixing remained significantly associated with both diseases. The links between diabetes and acute lymphoblastic leukemia observed for large regions are weaker for small areas. More powerful replications are needed for confirmation of these findings. Copyright © 2005 by the Johns Hopkins Bloomberg School of Public Health. All rights reserved.

Dentine hypersensitivity is recognized as a pain arising from fluid movement within dentine tubules that are open to the oral environment. Blocking the tubules is considered to be the principal aim of treatment, and the accurate assessment of tubule occlusion is the primary goal of many in vitro studies. This assessment usually comprises either measuring tubule permeability or scanning electron microscope examination of the dentine surface. Several scanning microscopy studies have claimed to quantify tubule occlusion, but are descriptive, qualitative or semi-quantitative evaluations. The present study was undertaken to assess the use of digital image analysis in quantifying the effectiveness of a selected desensitizing agent from micrographs of control and treated dentine surfaces. Using a dentine disc model, an accurate methodology was sought to investigate the occluding potential of Butler Protect (J.O. Butler, Chicago, IL, USA). Subjective examination of the images indicated there was little difference after a single application, but considerable effect after 20 applications. Quantitative digital analysis of a test image, demonstrated reproducibility between two examiners when used in fully- and semi-automated mode. After a single application of Butler Protect, multilevel statistical modelling demonstrated decreases in tubule area and maximum, minimum and mean diameter measurements (P < 0.001), whereas single level analysis showed increases in area and maximum and mean diameters. Multiple application of Butler Protect demonstrated even greater decreases in all parameters (P < 0.001). This quantitative methodology was reproducible between examiners and, when combined with good controls and multilevel statistical modelling, was able to discriminate a single application of desensitizing agent.

The relationship between initial disease status and subsequent change following treatment has attracted great interest in dental research. However, medical statisticians have repeatedly warned against correlating/regressing change with baseline because of two methodological concerns known as mathematical coupling and regression to the mean. In general, mathematical coupling occurs when one variable directly or indirectly contains the whole or part of another, and the two variables are then analyzed by using correlation or regression. Consequently, the statistical procedure of testing the null hypothesis - that the coefficient of correlation or the slope of regression is zero - may become inappropriate. Regression to the mean occurs with any variable that fluctuates within an individual or a population, either owing to measurement error and/or to physiological variation. The aim of this article was to clarify the conceptual confusion around mathematical coupling and regression to the mean within the statistical literature, and to correct a popular misconception about the correct analysis of the relationship between change and initial value. As examples that use inappropriate methods to analyze the relationship between change and baseline are still found in leading dental journals, this article seeks to help oral health researchers understand these problems and explain how to overcome them.

Oral health researchers have shown great interest in the relationship between the initial status of diseases and subsequent changes following treatment. Two main approaches have been adopted to provide evidence of a positive association between baseline values and their changes following treatment. One approach is to use correlation or regression to test the relationship between baseline measurements and subsequent change (correlation/regression approach). The second approach is to categorize the lesions into subgroups, according to threshold values, and subsequently compare the treatment effects across the two (or more) subgroups (categorization approach). However, the correlation/regression approach suffers a methodological weakness known as mathematical coupling. Consequently, the statistical procedure of testing the null hypothesis becomes inappropriate. Categorization seems to avoid the problem of mathematical coupling, although it still suffers regression to the mean. We show, first, how the appropriate null hypothesis may be established to analyze the relationship between baseline values and change in the correlation approach and, second, we use computer simulations to investigate the impact of regression to the mean on the significance testing of the differences in the average treatment effects (or average baseline values) in the categorization approach. Data available from previous literature are reanalyzed by testing the appropriate null hypotheses and the results are compared to those from testing the usual (incorrect) null hypothesis. The results indicate that both the correlation and categorization approaches can give rise to misleading conclusions and that more appropriate methods, such as Oldham's method and our new approach of deriving the correct null hypothesis, should be adopted.

Observational studies suggest that physical activity can reduce the risk of mental health and substance use disorders. However, it is unclear whether this relationship is causal or explained by confounding bias (e.g., common underlying causes or reverse causality). We investigated the bidirectional causal relationship of physical activity (PA) and sedentary behaviour (SB) with ten mental health and substance use disorders, applying two-sample Mendelian Randomisation (MR). Genetic instruments for the exposures and outcomes were derived from the largest available, non-overlapping genome-wide association studies (GWAS). Summary-level data for objectively assessed PA (accelerometer-based average activity, moderate activity, and walking) and SB and self-reported moderate-to-vigorous PA were obtained from the UK Biobank. Data for mental health/substance use disorders were obtained from the Psychiatric Genomics Consortium and the GWAS and Sequencing Consortium of Alcohol and Nicotine Use. MR estimates were combined using inverse variance weighted meta-analysis (IVW). Sensitivity analyses were conducted to assess the robustness of the results. Accelerometer-based average PA was associated with a lower risk of depression (b = -0.043, 95% CI: -0.071 to -0.016, effect size[OR] = 0.957) and cigarette smoking (b = -0.026; 95% CI: -0.035 to -0.017, effect size[β] = -0.022). Accelerometer-based SB decreased the risk of anorexia (b = -0.341, 95% CI: -0.530 to -0.152, effect size[OR] = 0.711) and schizophrenia (b = -0.230; 95% CI: -0.285 to -0.175, effect size[OR] = 0.795). However, we found evidence of reverse causality in the relationship between SB and schizophrenia. Further, PTSD, bipolar disorder, anorexia, and ADHD were all associated with increased PA. This study provides evidence consistent with a causal protective effect of objectively assessed but not self-reported PA on reduced depression and cigarette smoking. Objectively assessed SB had a protective relationship with anorexia. Enhancing PA may be an effective intervention strategy to reduce depressive symptoms and addictive behaviours, while promoting sedentary or light physical activities may help to reduce the risk of anorexia in at-risk individuals.

BACKGROUND: Estimating relative causal effects (i.e., "substitution effects") is a common aim of nutritional research. In observational data, this is usually attempted using 1 of 2 statistical modeling approaches: the leave-one-out model and the energy partition model. Despite their widespread use, there are concerns that neither approach is well understood in practice. OBJECTIVES: We aimed to explore and illustrate the theory and performance of the leave-one-out and energy partition models for estimating substitution effects in nutritional epidemiology. METHODS: Monte Carlo data simulations were used to illustrate the theory and performance of both the leave-one-out model and energy partition model, by considering 3 broad types of causal effect estimands: 1) direct substitutions of the exposure with a single component, 2) inadvertent substitutions of the exposure with several components, and 3) average relative causal effects of the exposure instead of all other dietary sources. Models containing macronutrients, foods measured in calories, and foods measured in grams were all examined. RESULTS: The leave-one-out and energy partition models both performed equally well when the target estimand involved substituting a single exposure with a single component, provided all variables were measured in the same units. Bias occurred when the substitution involved >1 substituting component. Leave-one-out models that examined foods in mass while adjusting for total energy intake evaluated obscure estimands. CONCLUSIONS: Regardless of the approach, substitution models need to be constructed from clearly defined causal effect estimands. Estimands involving a single exposure and a single substituting component are typically estimated more accurately than estimands involving more complex substitutions. The practice of examining foods measured in grams or portions while adjusting for total energy intake is likely to deliver obscure relative effect estimands with unclear interpretations.

Owing to mathematical coupling, statistical analyses relating change to baseline values using correlation or regression are erroneous, where the statistical procedure of testing the null hypothesis becomes invalid. Alternatives, such as Oldham’s method and the variance ratio test, have been advocated, although these are limited in the presence of measurement errors with non-constant variance. Furthermore, such methods prohibit the consideration of additional covariates (e.g., treatment group within trials) or confounders (e.g., age and gender). This study illustrates the more sophisticated approach of multilevel modelling (MLM) which overcomes these limitations and provides a comprehensive solution to the analysis of change with respect to baseline values. Although mathematical coupling is widespread throughout applied research, one particular area where several studies have suggested a strong relationship between baseline disease severity and treatment effect is guided tissue regeneration (GTR) within dental research. For illustration, we use GTR studies where the original data were found to be available in the literature for reanalysis. We contrast the results from an MLM approach and Oldham’s method with the standard (incorrect) approach that suffers from mathematical coupling. MLM provides a robust solution when relating change to baseline and is capable of simultaneously dealing with complex error structures and additional covariates and/or potential confounders.

Aims/hypothesis This study aimed to examine the relationship between average glucose levels, assessed by continuous glucose monitoring (CGM), and HbA1c levels in pregnant women with diabetes to determine whether calculations of standard estimated average glucose (eAG) levels from HbA1c measurements are applicable to pregnant women with diabetes. Methods CGM data from 117 pregnant women (89 women with type 1 diabetes; 28 women with type 2 diabetes) were analysed. Average glucose levels were calculated from 5–7 day CGM profiles (mean 1275 glucose values per profile) and paired with a corresponding (±1 week) HbA1c measure. In total, 688 average glucose–HbA1c pairs were obtained across pregnancy (mean six pairs per participant). Average glucose level was used as the dependent variable in a regression model. Covariates were gestational week, study centre and HbA1c. Results There was a strong association between HbA1c and average glucose values in pregnancy (coefficient 0.67 [95% CI 0.57, 0.78]), i.e. a 1% (11 mmol/mol) difference in HbA1c corresponded to a 0.67 mmol/l difference in average glucose. The random effects model that included gestational week as a curvilinear (quadratic) covariate fitted best, allowing calculation of a pregnancy-specific eAG (PeAG). This showed that an HbA1c of 8.0% (64 mmol/mol) gave a PeAG of 7.4–7.7 mmol/l (depending on gestational week), compared with a standard eAG of 10.2 mmol/l. The PeAG associated with maintaining an HbA1c level of 6.0% (42 mmol/mol) during pregnancy was between 6.4 and 6.7 mmol/l, depending on gestational week. Conclusions/interpretation The HbA1c–average glucose relationship is altered by pregnancy. Routinely generated standard eAG values do not account for this difference between pregnant and non-pregnant individuals and, thus, should not be used during pregnancy. Instead, the PeAG values deduced in the current study are recommended for antenatal clinical care.

Objective: With the introduction of active management of the third stage of labour in the 1960s, it became usual practice to clamp and cut the umbilical cord immediately following birth. The timing of this cord clamping is controversial, as blood may beneficially be transferred to the baby if clamping of the cord is delayed slightly. There is no agreement, however, on how long the delay should be before clamping the cord. This study aimed to establish when blood ceased to flow in the umbilical cord to determine how long to delay clamping of the umbilical cord following delivery of the term newborn to maximise placental transfusion. Methods: This observational study collected longitudinal weight measurements set in a hospital labour ward. A total of 26 mothers at term and their singleton babies participated in the study. In this reanalysis, the velocity of weight change over the first minutes of life determined by functional data analysis was estimated. Results: We found that the flow velocity in the umbilical cord was on average 0 at 125 s after placing the baby on the scales, which was typically 140 s after birth. Conclusions: To maximise placental transfusion, cord clamping should be delayed for at least 140 s following birth of the baby.

The developmental origins of health and disease hypothesis suggests that small birth size in conjunction with rapid compensatory child- hood growth might yield a greater risk of developing chronic diseases in later life. For example, there is evidence that people who de- veloped coronary heart disease and diabetes experienced different growth trajectories from those who did not develop these diseases. However, some of the methods used in these articles may have been flawed. We critically evaluate proposed approaches for identifying the growth trajectories distinctive to those developing later disease and identifying critical phases of growth during the early lifecourse. Among the approaches we examined (tracing the z-scores, lifecourse plots and models, lifecourse path analysis, conditional body size ana- lysis, multilevel analysis, latent growth curve models and growth mixture models) conditional body size analysis, multilevel analysis, latent growth curve models and growth mixture models are least prone to collinearity problems caused by repeated measures. Multilevel analysis is more flexible when body size is not measured at the same age for all cohort members. Strengths and weaknesses of each approach are illustrated using real data. Demonstrating the in- fluence of growth trajectories on later disease is complex and chal- lenging; therefore, it is likely that a combination of approaches will be required to unravel the complexity in lifecourse research.

Heart failure and left ventricular (LV) systolic dysfunction (LVSD) are common in patients with permanent pacemakers. The aim was to determine if cardiac resynchronization therapy (CRT) at the time of pulse generator replacement (PGR) is of benefit in patients with unavoidable RV pacing and LVSD. Fifty patients with unavoidable RV pacing, LVSD, and mild or no symptoms of heart failure, listed for PGR were randomized 1 : 1 to either standard RV-PGR (comparator) or CRT. The primary endpoint was the difference in change in LV ejection fraction (LVEF) between RV-PGR and CRT groups from baseline to 6 months. Secondary endpoints included peak oxygen consumption, quality of life, and N-terminal pro-B-type natriuretic peptide (NT-proBNP) levels. At 6 months there was a difference in change in median (interquartile range) LVEF [9 (6-12) vs. -1.5 (-4.5 to -0.8)%; P < 0.0001] between the CRT and RV-PGR arms. There were also improvements in exercise capacity (P = 0.007), quality of life (P = 0.03), and NT-proBNP (P = 0.007) in those randomized to CRT. After 809 (729-880) days, 17 patients had died or been hospitalized (6 in CRT group and 11 in the comparator RV-PGR group) and two patients in the RV-PGR arm had required CRT for deteriorating heart failure. Patients with standard RV-PGR had more days in hospital during follow-up than those in the CRT group [4 (2-7) vs. 11 (6-16) days; P = 0.047]. Performing CRT in pacemaker patients with unavoidable RV pacing and LVSD but without severe symptoms of heart failure, at the time of PGR, improves cardiac function, exercise capacity, quality of life, and NT-pro-BNP levels.

The utility of blood-based omic profiles for linking environmental exposures to their potential health effects was evaluated in 649 individuals, drawn from the general population, in relation to tobacco smoking, an exposure with well-characterised health effects. Using disease connectivity analysis, we found that the combination of smoking-modified, genome-wide gene (including miRNA) expression and DNA methylation profiles predicts with remarkable reliability most diseases and conditions independently known to be causally associated with smoking (indicative estimates of sensitivity and positive predictive value 94% and 84%, respectively). Bioinformatics analysis reveals the importance of a small number of smoking-modified, master-regulatory genes and suggest a central role for altered ubiquitination. The smoking-induced gene expression profiles overlap significantly with profiles present in blood cells of patients with lung cancer or coronary heart disease, diseases strongly associated with tobacco smoking. These results provide proof-of-principle support to the suggestion that omic profiling in peripheral blood has the potential of identifying early, disease-related perturbations caused by toxic exposures and may be a useful tool in hazard and risk assessment.

Within lifecourse epidemiology, there is substantial interest in relationships between exposures (X) measured longitudinally (e.g. at time 0, 1, 2; hence X0, X1, X2) and outcomes (Y) measured cross-sectionally once (e.g. Y2 or Y2 + t). Within a causal framework, modelling presents many challenges, such as how to account for time-invariant confounding. It has been demonstrated that the information contained in separate models for each exposure can be combined into an overall model using unexplained residuals; it is suggested this also reduces standard errors of estimated effect sizes. This study explored how confounders are incorporated into this framework and whether standard errors are indeed reduced. Directed acyclic graphs (DAGs) depict a range of potential causal relationships between a longitudinal exposure (BMI: body mass index, aged 2, 15 and 64 years; BMI2, BMI15 and BMI64), and a later-life outcome (CRP: C-reactive protein, a marker for cardiovascular disease) measured once around the time of the last exposure assessment (CRP64), in the presence of a time-invariant confounder (S: sex) measured only at baseline. The DAG guides plausible covariance structures in the data simulation and steers thinking about at which stage(s) in the residual increase modelling process one should incorporate confounders. Residual increase models were then contrasted to standard regression models. With standard regression we model the impact of BMI2 using CRP64~BMI2+S, the impact of BMI15 using CRP64~BMI15+BMI2+S, and the impact of BMI64 using CRP64~BMI64+BMI15+BMI2+S. In residual increase models, BMI15~BMI2+S and BMI64~BMI15+BMI2+S yield residuals E15 and E64, representing differences between expected and actual values of BMI at each time point (or the ‘residual increase’ in BMI) while controlling for the confounding effect, S. We also control for S in the overall model: CRP64~BMI2+E15+E64+S. It is shown that the residual increase model has coefficients equivalent to those produced by the separate standard regression models and this model offers a modest reduction in the standard errors of estimated coefficients. A time-invariant confounder must be accommodated when generating ‘residual increases’ and in the overall residual increase model. Incorporating confounders this way allows for the interpretation of multiple effects in a single model, though this offers no new information about variable relationships in and of themselves. It is important to use DAGs to determine how to incorporate confounders into the residual increase model framework, which offers modest reductions in the standard errors of estimated coefficients, though model interpretation is perhaps not as straightforward as with standard regression models.

Amyotrophic Lateral Sclerosis (ALS) is a progressive neurodegenerative disease with an unknown cause. Studies have reported that the incidence rate of ALS might be changing. As ALS is an age related disease, crude incidence could increase as population structure changes and overall life expectancy improves. Age-period-cohort (APC) models are frequently used to investigate trends in demographic rates such as incidence. Age-specific incidence rate for ALS from 1996 to 2014 were taken from a population-based ALS register in Ireland. To circumvent the well-known identifiability issue in APC models, we apply the method of Partial Least Squares Regression to separate the effects of Age, Period and Cohort on ALS incidence over time. This APC analysis shows no cohort effect and the initial signs of a period effect; increasing incidence of ALS in the most recently diagnosed group. As further years of data accrue to the Irish register it will become clear if this effect emerges as a strong trend in the incidence of ALS in Ireland and replication of these analyses in other populations will show if our findings on temporal patterns in ALS incidence are shared elsewhere.

AIMS: To investigate the association between radiotherapy waiting times and survival in women who have undergone breast-conserving surgery using data from two English cancer registry regions. The data were analysed using path analysis to account for the complex variable interrelationships within the data. MATERIALS AND METHODS: Cases of female invasive breast cancer diagnosed during the period 1 January 1998 to 31 December 2005 were identified and linked to an extract of Hospital Episode Statistics data. A subset of these linked records where women underwent breast-conserving surgery was extracted (n=18,158). Patient, tumour and treatment information were extracted. A path model was developed with three outcome variables: survival, time to receive radiotherapy and receipt of chemotherapy before radiotherapy. RESULTS: During the study period, the median radiotherapy waiting time in region 1 increased from 70 days to 128.5 days. In region 2, the median wait increased from 44 days in 1998 to 68 days in 2001, then decreased to 42 days by 2005. In the path model, radiotherapy waiting time was not associated with survival (hazard ratio=1.00, 95% confidence interval 0.99-1.01 per week increase in both regions). Patients receiving chemotherapy before radiotherapy waited 12.3 weeks (region 1) and 6.3 weeks (region 2) longer for their radiotherapy than those not receiving chemotherapy. Patients with stage II/III disease waited longer than patients with stage I disease. Younger age, diagnosis of stage II/III disease and presence of co-morbidities were associated with increased odds of receiving chemotherapy before radiotherapy. CONCLUSIONS: This study found no association between waiting times for radiotherapy and survival in two regions of England, despite increases in waiting times over the study period. Such an association, if real, may only become apparent after a longer period of follow-up.

Background: The obesity-psychological distress relationship remains controversial. Purpose: This study aims to assess whether adolescents' psychological distress was associated with body mass index (BMI) class membership determined by latent class analysis. Methods: Distress (anxiety, depression) and BMI were measured annually for 4 years in 1,528 adolescents. Growth mixture modeling derived latent BMI trajectory classes for models with 2-11 classes. The relationship of distress to class membership was examined in the best-fitting model using vector generalized linear regression. Results: BMI trajectories were basically flat. The five-class model [normal weight (48.8%), overweight (36.7%), obese who become overweight (3.7%), obese (9.4%), and severely obese (1.3%)] was the preferred model (Bayesian information criterion = 22789.2, df = 31; ρ = 0.84). Greater distress was associated with higher baseline BMI and, therefore, class membership. Conclusions: Psychological distress is associated with higher BMI class during adolescence. To determine whether distress "leads" to greater weight gain may require studies of younger populations. © 2011 The Society of Behavioral Medicine.

The inclusion of geographically-based information in many epidemiological studies has led to the development of statistical and estimation methods that account for spatially dependent risks of health outcomes across geographical areas. Many of these developments have been concerned with spatial modelling of aggregated count data, for example incidence rates of cancer at the small“area level. In this chapter, we consider individual time­to­event data, where the individual subjects are hierarchically nested in natural or administrative areas. The individual failure time data are modelled using proportional hazards models, which are modified to include both spatially uncorrelated and correlated area frailty random effects; the latter accounting for local spatial dependence in the data. This model is expanded to accommodate multiple failure events, where the set of within and between failure-event spatial frailty random effects are assumed to have a multivariate normal distribution. We illustrate the proposed methodology with an analysis of timing of first childbirth and timing of first marriage across health districts in South Africa for women aged between 15 and 49 years. For each failure event, the spatial dependence is modelled using a multiple membership multiple classification (MMMC) model. A multivariate version of the MMMC model is then used to obtain estimates of covariance parameters between various failure-event spatial random effects.

In biomedical research, data generated as a consequence of the count process can often possess an excess of zeros (e.g. geographical incidence rates, hospital death rates). Whilst there are strategies for analysing such data, some can be biased where the underlying data generation process is not carefully considered. This can be exacerbated where the data are also multilevel, since hierarchical extensions to zero-inflated model strategies do not always satisfy underlying model assumptions. We therefore review zero-inflated modelling strategies for single-level data and show why standard Poisson and binomial zero-inflated models (i.e. where one latent class has a central location of zero) require class membership to be predicted by covariates in the standard regression part of the model. We also introduce generic mixture models and reveal limitations in their interpretation in a number of circumstances. With nested or hierarchical count data with an excess of zeros, upper-level distributional assumptions may not be upheld for standard multilevel models, thereby requiring alternative strategies; in Chap. 7 we introduce and illustrate the semi-parametric multilevel model as a solution to this problem.

In Chap. 5 we introduced multilevel modelling, where a continuous latent variable represents variation across the levels of a natural hierarchy, yielding random effects. In Chap. 6, we introduced latent class analysis, where using a binary latent variable gave rise to a mixture model of count data to accommodate an excess of zeros relative to standard count distributions. In this chapter, we combine and develop these concepts further. By employing discrete latent variables for upper levels of a hierarchy, a richer mixture model can be represented, with mixtures at the lower level only, the higher level only, or at both levels. Where the upper level mixture has many latent classes, this model can be viewed as similar to the traditional multilevel model, though with a discrete (opposed to continuous) distribution and thereby foregoing any parametric assumptions: for instance, the normal distribution at the upper level is replaced by a number of discrete components, and for many components, this can be viewed as a semi-parametric approximation of the continuous latent variable.

Bias is inherent in epidemiology, and researchers go to great lengths to avoid introducing bias into their studies. However, some bias is inevitable, and bias due to selection is particularly common. We discuss ways to identify bias and how authors have approached removing or adjusting for bias using statistical methods.

Statistical interactions and biological interactions are different concepts. Much confusion arises where the former is used to describe the latter, particularly in the exploration of gene-environment interactions. This will be illustrated and the circumstances under which statistical interaction may be legitimately interpreted for gene-environment interactions are presented. What is address, however, has relevance to all circumstances in biomedical research where statistical interaction is to be used and interpreted to infer biological meaning.

Background Exploratory factor analysis is a commonly used statistical technique in metabolic syndrome research to uncover latent structure amongst metabolic variables. The application of factor analysis requires methodological decisions that reflect the hypothesis of the metabolic syndrome construct. These decisions often raise the complexity of the interpretation from the output. We propose two alternative techniques developed from cluster analysis which can achieve a clinically relevant structure, whilst maintaining intuitive advantages of clustering methodology. Methods Two advanced techniques of clustering in the VARCLUS and matroid methods are discussed and implemented on a metabolic syndrome data set to analyze the structure of ten metabolic risk factors. The subjects were selected from the normative aging study based in Boston, Massachusetts. The sample included a total of 847 men aged between 21 and 81 years who provided complete data on selected risk factors during the period 1987 to 1991. Results Four core components were identified by the clustering methods. These are labelled obesity, lipids, insulin resistance and blood pressure. The exploratory factor analysis with oblique rotation suggested an overlap of the loadings identified on the insulin resistance and obesity factors. The VARCLUS and matroid analyses separated these components and were able to demonstrate associations between individual risk factors. Conclusions An oblique rotation can be selected to reflect the clinical concept of a single underlying syndrome, however the results are often difficult to interpret. Factor loadings must be considered along with correlations between the factors. The correlated components produced by the VARCLUS and matroid analyses are not overlapped, which allows for a simpler application of the methodologies and interpretation of the results. These techniques encourage consistency in the interpretation whilst remaining faithful to the construct under study.

Background: The 1995 Calman-Hine plan outlined radical reform of the UK’s cancer services with the aim of improving outcomes and reducing inequalities in NHS cancer care. Its main recommendation was to concentrate care into the hands of site-specialist multi-disciplinary teams. This study aimed to determine if these teams improved processes and outcomes of care for breast cancer patients. Patients and Methods: All patients diagnosed and treated with breast cancer in the Yorkshire region of the UK between 1995 and 2000 were identified within the Northern & Yorkshire Cancer Registry & Information Service (NYCRIS) database. Changes in the use of breast-conserving surgery, adjuvant radiotherapy following breast-conserving surgery and five-year survival were assessed amongst these patients in relation to their managing breast cancer team’s degree of adherence to the manual of cancer service standards (which outlines the specification of the ‘ideal’ breast cancer team); and the extent of site-specialisation of each team’s surgeons. Results: Variation was observed in the extent to which the breast cancer teams in Yorkshire had conformed to the Calman-Hine recommendations. Increases in adherence to the recommendations in the manual of cancer service standards were associated with a reduction in the use of breast-conserving surgery (OR=0.83, 95%CI=0.70-0.98, p<0.01). Increases in both surgical specialisation (OR=1.23, 95%CI=1.00-1.55, p=0.06 ) and adherence to the manual of cancer service standards (OR=1.22, 95%CI=0.97-1.52, p=0.05) were associated with the increased use of radiotherapy following breast-conserving surgery. There was a trend towards improved five-year survival (HR=0.93, 95%CI=0.86-1.01, p=0.10) in relation to increasing surgical site specialisation. All these effects were present after adjustment for the casemix factors of age, stage of disease, socio-economic background and year of diagnosis. Conclusions: The extent of implementation of the Calman-Hine report has been variable and, based on the limited clinical and organisational information available, its recommendations appear to be associated with improvements in processes and outcomes of care for breast cancer patients.

Abstract

There is growing evidence that some chronic diseases are caused, or promoted, by infectious disease. ‘Population mixing’ has been used as a proxy for the range and dose of infectious agents circulating in a community. Given the speculation over the role of population mixing in many chronic diseases, we review the various methods used for measuring population mixing, and provide a classification of these. We recommend that authors fulfill two criteria in publications: measures are demonstrably associated with the putative risk factors for which population‐mixing is acting as a proxy and fundamental characteristics of the chosen measures are clearly defined. Pediatr Blood Cancer 2008;51:155–160. © 2008 Wiley‐Liss, Inc.

Background Hospital mortality rates are one of the most frequently selected indicators for measuring the performance of NHS Trusts. A recent article in a national newspaper named the hospital with the highest or lowest mortality in the 2005/6 financial year; a report by the organization Dr Foster Intelligence provided information with regard to the performance of all NHS Trusts in England. Methods Basic statistical theory and computer simulations were used to explore the relationship between the variations in the performance of NHS Trusts and the sizes of the Trusts. Data of hospital standardised mortality ratio (HSMR) of 152 English NHS Trusts for 2005/6 were re-analysed. Results A close examination of the information reveals a pattern which is consistent with a statistical phenomenon, discovered by the French mathematician de Moivre nearly 300 years ago, described in every introductory statistics textbook: namely that variation in performance indicators is expected to be greater in small Trusts and smaller in large Trusts. From a statistical viewpoint, the number of deaths in a hospital is not in proportion to the size of the hospital, but is proportional to the square root of its size. Therefore, it is not surprising to note that small hospitals are more likely to occur at the top and the bottom of league tables, whilst mortality rates are independent of hospital sizes. Conclusion This statistical phenomenon needs to be taken into account in the comparison of hospital Trusts performance, especially with regard to policy decisions.

BACKGROUND: To clarify the nature of the relationship between: food deprivation and undernutrition during pre- and postnatal development; and cholesterol levels in later life, this study examined the relationship between birth weight (as a marker of prenatal nutrition) and cholesterol levels among 396 Guernsey islanders (born in 1923-1937), 87 of whom (22%) had been exposed to food deprivation as children, adolescents or young adults (i.e. to postnatal undernutrition) during the 1940-45 German occupation of the Channel Islands, and 309 of whom (78%) had left or been evacuated from the islands before the occupation began. METHODS: Three sets of multiple regression models were used to investigate: Model A - the relationship between birth weight and cholesterol levels; Model B - the relationship between postnatal exposure to the occupation and cholesterol levels; and Model C - any interaction between birth weight, postnatal exposure to the occupation and cholesterol levels. Model A and Model B also tested for any interactions between: birth weight/occupation exposure and sex; and birth weight/occupation exposure and parish of residence at birth (as a marker of parish of residence during the occupation and related variation in the severity of food deprivation). RESULTS: Before (and after) adjusting for potential confounders, no statistically significant relationships were observed between either birth weight (before adjustment: 0.09 mmol/l per kg increase, 95% CI: -0.30, 0.16; after adjustment: 0.08 mmol/l per kg increase, 95%CI: -0.17, 0.34) or exposure to the occupation (before adjustment: 0.01 mmol/l for exposed group, 95%CI: -0.24, 0.27; after adjustment: 0.04 mmol/l for exposed group, 95%CI: -0.26, 0.33) and cholesterol levels in later life. There was also little evidence of significant relationships between birth weight, exposure to the occupation and cholesterol levels in later life when Model A and Model B were stratified by sex or parish of residence at birth, although there was a significant positive relationship between birth weight and cholesterol levels in women (0.44 mmol/l per kg increase, 95%CI: 0.07, 0.81). CONCLUSIONS: These analyses provide little support for the theory that birth weight is inversely related to cholesterol levels in later life. and do not offer any evidence in support of a relationship between undernutrition in childhood, adolescence and early adulthood and cholesterol levels in later life. However, further research may determine whether undernutrition at different stages of the life-course may influence cholesterol levels in later life.

Childhood acute lymphoblastic leukaemia (ALL) and Type 1 diabetes (T1D) share some common epidemiological features, including rising incidence rates and links with an infectious aetiology. Previous work has shown a significant positive correlation in incidence between the two conditions both at the international and small-area level. The aim was to extend the methodology by including shared spatial and temporal trends using a more extensive dataset among individuals diagnosed with ALL and T1D in Yorkshire (UK) aged 0-14 years from 1978-2003. Cases with ALL and T1D were ascertained from 2 high quality population-based disease registers covering the Yorkshire region of the UK and linked to an electoral ward from the 1991 UK census. A Bayesian model was fitted where similarities and differences in risk profiles of the two diseases were captured by the shared and disease-specific components using a shared-component model, with space-time interactions. The extended model revealed a positive correlation of at least 0.70 between diseases across all time periods, and an increasing risk across time for both diseases, which was more evident for T1D. Furthermore, both diseases exhibited lower rates in the more urban county of West Yorkshire and higher rates in the more rural northern and eastern part of the region. A differential effect of T1D over ALL was found in the south-eastern part of the region, which had a more pronounced association with population mixing than with population density or deprivation. Our approach has demonstrated the utility in modelling temporally and spatially varying disease incidence patterns across small geographical areas. The findings suggest searching for environmental factors that exhibit similar geographical-temporal variation in prevalence may help in the development and testing of plausible aetiological hypotheses. Furthermore, identifying environmental exposures specific to the south-eastern part of the region, especially locally varying risk factors which may differentially affect the development of T1D and ALL, may also be fruitful.

BACKGROUND & AIMS: Population-based endoscopic studies are needed to assess the epidemiology of functional dyspepsia (FD) and the newly suggested subgroups of meal-related symptoms and epigastric pain. We evaluated the prevalence of, and risk factors for, FD in the Italian general population. METHODS: A total of 1533 inhabitants of 2 villages were invited to undergo symptom evaluation using a validated questionnaire, esophagogastroduodenoscopy, and (13)C-urea breath test; 1033 subjects (67.4%) took part. RESULTS: Of the 1033 subjects, 156 (15.1%; 95% confidence interval [CI], 12.9-17.3) had dyspepsia, and of these 114 (11%; 95% CI, 9.2-12.9) had FD. Of the 114 subjects with FD, 77 (67.5%) had meal-related symptoms (postprandial fullness and/or early satiation) and 55 (48.2%) had epigastric pain. Only 18 subjects (15.8%) had both meal-related symptoms and epigastric pain; this was fewer than expected by chance alone (P < .001). Unemployment (odds ratio [OR], 5.80; 95% CI, 1.56-21.60), divorce (OR, 2.76; 95% CI, 1.10-6.91), smoking (OR, 1.74; 95% CI, 1.11-2.70), and irritable bowel syndrome (OR, 3.38; 95% CI, 1.85-6.19) were significantly associated with FD. Unemployment, divorce, and irritable bowel syndrome were associated with both meal-related symptoms and epigastric pain, while smoking was associated only with meal-related symptoms. CONCLUSIONS: FD is present in 11% of the Italian general population. Unemployment and divorce seem to increase the risk of FD, and smoking seems to be associated with meal-related symptoms. Two distinct subgroups of FD, as suggested by Rome III, seem to exist in the general population.

Over 60 years ago Ronald Fisher demonstrated a number of potential pitfalls with statistical analyses using ratio variables. Nonetheless, these pitfalls are largely overlooked in contemporary clinical and epidemiological research, which routinely uses ratio variables in statistical analyses. This article aims to demonstrate how very different findings can be generated as a result of less than perfect correlations among the data used to generate ratio variables. These imperfect correlations result from measurement error and random biological variation. While the former can often be reduced by improvements in measurement, random biological variation is difficult to estimate and eliminate in observational studies. Moreover, wherever the underlying biological relationships among epidemiological variables are unclear, and hence the choice of statistical model is also unclear, the different findings generated by different analytical strategies can lead to contradictory conclusions. Caution is therefore required when interpreting analyses of ratio variables whenever the underlying biological relationships among the variables involved are unspecified or unclear.

BACKGROUND: To clarify the nature of the relationship between food deprivation/undernutrition during pre- and postnatal development and cardiovascular disease (CVD) in later life, this study examined the relationship between birth weight (as a marker of prenatal nutrition) and the incidence of hospital admissions for CVD from 1997-2005 amongst 873 Guernsey islanders (born in 1923-1937), 225 of whom had been exposed to food deprivation as children, adolescents or young adults (i.e. postnatal undernutrition) during the 1940-45 German occupation of the Channel Islands, and 648 of whom had left or been evacuated from the islands before the occupation began. METHODS: Three sets of Cox regression models were used to investigate (A) the relationship between birth weight and CVD, (B) the relationship between postnatal exposure to the occupation and CVD and (C) any interaction between birth weight, postnatal exposure to the occupation and CVD. These models also tested for any interactions between birth weight and sex, and postnatal exposure to the occupation and parish of residence at birth (as a marker of parish residence during the occupation and related variation in the severity of food deprivation). RESULTS: The first set of models (A) found no relationship between birth weight and CVD even after adjustment for potential confounders (hazard ratio (HR) per kg increase in birth weight: 1.12; 95% confidence intervals (CI): 0.70-1.78), and there was no significant interaction between birth weight and sex (p=0.60). The second set of models (B) found a significant relationship between postnatal exposure to the occupation and CVD after adjustment for potential confounders (HR for exposed vs. unexposed group: 2.52; 95% CI: 1.54-4.13), as well as a significant interaction between postnatal exposure to the occupation and parish of residence at birth (p=0.01), such that those born in urban parishes (where food deprivation was worst) had a greater HR for CVD than those born in rural parishes. The third model (C) found no interaction between birth weight and exposure to the occupation (p=0.43). CONCLUSION: These findings suggest that the levels of postnatal undernutrition experienced by children, adolescents and young adults exposed to food deprivation during the 1940-45 occupation of the Channel Islands were a more important determinant of CVD in later life than the levels of prenatal undernutrition experienced in utero prior to the occupation.

AIM: To explore ethnic and gender variations amongst applicants to undergraduate United Kingdom medical and dental schools. METHOD: Retrospective analyses of University and College Admissions Services (UCAS) data on all students applying to study pre-clinical medicine and dentistry, during the academic years 1994/5, 1995/6 and 1996/7. Information for each medical and dental applicant included age, gender, social class and ethnic group. RESULTS: Of all applicants, just over half (50.2%) were male, though a greater proportion of applicants to dentistry were male (54.1%) than for medicine (49.3%) (OR = 1.21, 95% CI = 1.15, 1.28). Over one third (36.4%) of all students were from minority ethnic groups, a larger proportion of which were dental students (48.3%) than were medical students (33.8%) (OR = 1.83, 95% CI = 1.73, 1.94). There were also marked differences between medicine and dentistry when the ethnic groups were examined separately. The largest number of applicants from minority ethnic groups came from the Indian community, and this group increased in size annually by 4.1% (P < 0.05) for medicine, and 29% (P < 0.05) for dentistry. CONCLUSIONS: Significant inter-ethnic and gender differences are observed amongst applicants to medicine and dentistry. Dentistry appears to be relatively more attractive to minority ethnic applicants.

This study retrospectively investigated variations in the use of secondary healthcare for head injury, particularly assault. A total of 25,300 emergency head-related admissions were examined over a two-year period, of which 3756 were assaults. There were seasonal differences according to ethnic group: far more injuries, particularly assault, occurred amongst the black and minority ethnic groups during the summer months and holidays. Black males had two to three times the rate of admission for assault than any other group. Among whites, females stayed longer in hospital after a head injury. White women stayed significantly longer than South Asian women following a head injury and South Asian men stayed significantly longer than white men after an assault. There are substantial seasonal variations and differences in the length of hospital stay after a head injury, particularly assault, depending on ethnic group. These differences require clarification and more detailed studies of head injuries ought to record the patient's ethnic background.

Oral epithelial dysplasia (OED) is an important risk factor in predicting subsequent development of invasive carcinoma. Despite the malignant potential of OED, the current state of knowledge regarding aetiological risk factors associated with OED is limited. The aim of our study was to evaluate the aetiological role of alcohol consumption in non-tobacco smokers and smoking behaviour in non-drinkers, in patients presenting with OED. Data from a hospital-based case-control study of OED were used to analyse the risk associated with alcohol in non-smokers and with tobacco in non-drinkers. A total of 140 cases and 236 controls were included. In the non-drinkers, the risks of OED increased with tobacco smoking of more than 20 cigarettes per day, particularly non-filter cigarettes. The risk of OED declined following smoking cessation, with ex-smokers of 10 or more years demonstrating no excess risk relative to non-smokers. In the nonsmokers, consumption of alcohol was not a significant predictor of OED. However, there was a synergistic effect of alcohol when combined with some aspects of tobacco smoking. Our results confirm that tobacco has an independent role in the aetiology of OED. The role of alcohol, however, is principally only important in conjunction with tobacco use.

Monthly prescribing behaviour is assessed over a 3-year period, 1 April 1992 to 31 March 1995. Total monthly number of items prescribed and overall net ingredient cost are analysed for 263 general practices, serving the 1 million residents of Birmingham, UK. Patients aged over 65 years play an important role in elevated prescribing activity. Practice composition varies considerably between training and non-training practices, and between fundholding and non-fundholding practices. Accounting for these differences, fundholders expend less and prescribe fewer items than their non-fundholding counterparts. This is observed against a steady increase in prescribing activity over the study period. There are, however, marked downward shifts in both the number of items prescribed and overall monthly expenditure occurring with every new wave of fundholding. The magnitude of these changes raises doubts about the efficacy of the transition to fundholding and the impact of such large changes upon patient care.

In response to the introduction of ethnic monitoring within the U.K. hospital inpatient data set, this study investigates the variations in secondary healthcare utilization by Black and minority ethnic communities whose cause of admission is related to asthma. The study examines all residents of the West Midlands: over 5 million people, of whom 8.5% are from Black and minority ethnic groups. A retrospective study of 15,921 asthma-related hospital admissions, from 1 April 1995 to 31 March 1996, was carried out. Age-standardized admission rates were higher in all Black and minority ethnic groups studied than in the White group. There were elevated rates in Black children aged 5-14 years, and particular differences were observed for Indian and Bangladeshi men and women aged 65 years or over. Emergency admissions to hospital for asthma were strongly associated with patients' socioeconomic background but this was largely observed for Black and minority ethnic groups that also generally experience high levels of deprivation. The findings support previous studies which suggest that hospital utilization rates for asthma among people from Black and minority ethnic groups are high compared with the White group, despite little evidence in measured prevalence. This study suggests that ethnic background is more important in asthma admissions than deprivation, which raises serious concerns on the appropriateness and quality of asthma care for these patient groups within our society. Future studies need to examine pathways to care, that is the health-seeking behaviour of Black and minority ethnic groups, the type of treatment received at the primary care level and referral patterns to secondary care.

Abstract

This article discusses three statistical paradoxes that pervade epidemiological research: Simpson's paradox, Lord's paradox, and suppression. These paradoxes have important implications for the interpretation of evidence from observational studies. This article uses hypothetical scenarios to illustrate how the three paradoxes are different manifestations of one phenomenon – the reversal paradox – depending on whether the outcome and explanatory variables are categorical, continuous or a combination of both; this renders the issues and remedies for any one to be similar for all three. Although the three statistical paradoxes occur in different types of variables, they share the same characteristic: the association between two variables can be reversed, diminished, or enhanced when another variable is statistically controlled for. Understanding the concepts and theory behind these paradoxes provides insights into some controversial or contradictory research findings. These paradoxes show that prior knowledge and underlying causal theory play an important role in the statistical modelling of epidemiological data, where incorrect use of statistical models might produce consistent, replicable, yet erroneous results.

Randomized controlled trials (RCTs) are widely recommended as the most useful study design to generate reliable evidence and guidance to daily practices in medicine and dentistry. However, it is not well-known in dental research that different statistical methods of data analysis can yield substantial differences in study power. In this study, computer simulations are used to explore how using different univariate and multivariate statistical methods of analyzing change in continuous outcome variables affects study power, and the sample size required for RCTs. Results show that, in general, analysis of covariance (ANCOVA) yields greater power than other statistical methods in testing the superiority of one treatment over another, or in testing the equivalence between two treatments. Therefore, ANCOVA should be used in preference to change score or percentage change score to reduce type II error rates.

Deterministic variables are variables that are functionally determined by one or more parent variables. They commonly arise when a variable has been functionally created from one or more parent variables, as with derived variables, and in compositional data, where the 'whole' variable is determined from its 'parts'. This article introduces how deterministic variables may be depicted within directed acyclic graphs (DAGs) to help with identifying and interpreting causal effects involving derived variables and/or compositional data. We propose a two-step approach in which all variables are initially considered, and a choice is made whether to focus on the deterministic variable or its determining parents. Depicting deterministic variables within DAGs brings several benefits. It is easier to identify and avoid misinterpreting tautological associations, i.e., self-fulfilling associations between deterministic variables and their parents, or between sibling variables with shared parents. In compositional data, it is easier to understand the consequences of conditioning on the ‘whole’ variable, and correctly identify total and relative causal effects. For derived variables, it encourages greater consideration of the target estimand and greater scrutiny of the consistency and exchangeability assumptions. DAGs with deterministic variables are a useful aid for planning and interpreting analyses involving derived variables and/or compositional data.

Background In longitudinal data, it is common to create ‘change scores’ by subtracting measurements taken at baseline from those taken at follow-up, and then to analyse the resulting ‘change’ as the outcome variable. In observational data, this approach can produce misleading causal-effect estimates. The present article uses directed acyclic graphs (DAGs) and simple simulations to provide an accessible explanation for why change scores do not estimate causal effects in observational data. Methods Data were simulated to match three general scenarios in which the outcome variable at baseline was a (i) ‘competing exposure’ (i.e. a cause of the outcome that is neither caused by nor causes the exposure), (ii) confounder or (iii) mediator for the total causal effect of the exposure variable at baseline on the outcome variable at follow-up. Regression coefficients were compared between change-score analyses and the appropriate estimator(s) for the total and/or direct causal effect(s). Results Change-score analyses do not provide meaningful causal-effect estimates unless the baseline outcome variable is a ‘competing exposure’ for the effect of the exposure on the outcome at follow-up. Where the baseline outcome is a confounder or mediator, change-score analyses evaluate obscure estimands, which may diverge substantially in magnitude and direction from the total and direct causal effects. Conclusion Future observational studies that seek causal-effect estimates should avoid analysing change scores and adopt alternative analytical strategies.

OBJECTIVE: This study was designed to determine the United Kingdom public's perception of how oral health affects quality of life (QoL) and to determine socio-demographic variations in these perceptions. RESEARCH DESIGN: The vehicle for this study was the ONS Omnibus Survey in the UK. A random probability sample of 2,668 eligible addresses was selected from the British Postcode Address File. Setting The data were collected by qualitative, face-to-face interviews with respondents, nation-wide, in their homes, about how their oral health status affected their QoL. PARTICIPANTS: 1,778 adults aged 16 years or older across the UK took part in the study. RESULTS: 75% (1,340) believed their oral health either enhanced or reduced their QoL. Most frequently, this was perceived as being the result of its effect on eating. comfort and appearance. Other ways in which QoL was affected are also presented. Sociodemographic variations were apparent. For example, people from higher socio-economic backgrounds believed that their oral health enhanced their QoL to a greater degree (OR=1.46, CI=1.20, 1.78) than the lower socio-economic groups. Women claimed that their oral health had a greater negative effect on QoL than did men (OR=1.36, CI=1.11, 1.64). Younger people (16-64 years old) reported that their oral health status reduced and enhanced QoL more than older adults (65 years and over) (OR=1.59, CI=1.23, 2.04). CONCLUSIONS: The study shows that the UK public perceives oral health as affecting their QoL in a variety of physical, social and psychological ways and that significant socio-demographic variations exist in these perceptions.

Objective: To explain the concepts and application of Bayesian modelling and how it can be applied to the analysis of dental research data. Basic design: Methodological in nature, this article introduces Bayesian modelling through hypothetical dental examples. Setting: The synthesis of RCT results with previous evidence, including expert opinion, is used to illustrate full Bayesian modelling. Meta-analysis, in the form of empirical Bayesian modelling, is introduced. An example of full Bayesian modelling is described for the synthesis of evidence from several studies that investigate the success of root canal treatment. Hierarchical (Bayesian) modelling is demonstrated for a survey of childhood caries, where surface data is nested within subjects. Results: Bayesian methods enhance interpretation of research evidence through the synthesis of information from multiple sources. Conclusions: Bayesian modelling is now readily accessible to clinical researchers and is able to augment the application of clinical decision making in the development of guidelines and clinical practice.

Objective: To introduce and encourage the use of generalised linear models (GLMs) in analysing caries data that do not require the response to be treated necessarily as a sample from a normal distribution. Basic research design: At the present time, it is most likely that the sampling distribution of dmf/DMF in industrialised countries will not approximate normality. Generalised linear modelling can be conducted assuming many underlying distributions which, in fact, includes the normal distribution. In this paper three GLMs are employed (normal, Poisson, negative binomial) for modelling an example caries data set. In addition, a binomial model is used to model the dichotomous outcome of caries-free/caries-present. Clinical setting: The data comprised 871 Old Trafford, Manchester primary school children aged between 4 years 0 months and 5 years 11 months. Results: The effect of one study covariate was prominent in a normal model applied to all available dmf data but not in two non-normal models which used dmf > 0 data only. Furthermore, the same covariate was significant at the 5% level in a binomial model indicating that it influenced whether or not caries was present and not the level of dmf. Conclusion: A suitable modelling approach for caries data is to employ a Poisson or a negative binomial model for the dmf/DMF response and a binomial model for the caries-free/caries-present outcome. This allows separate estimation of those factors which influence the magnitude of caries and those factors which influence whether caries is actually present or not.

OBJECTIVE: To examine the socio-economic background of patients who underwent inpatient and day-case oral operations from 1989-1994 in the West Midlands. DESIGN: Retrospective. SETTING: National Health Services Executive West Midlands; with a population of approximately 5.5 million. SUBJECTS: A total of 4,926,438 hospital inpatient finished consultant episodes, within 56 specialties recorded in the hospital episode statistics database, of which 60,904 (1.24 per cent) were dental; oral surgery. OUTCOME MEASURE: Standardised hospital inpatient activity ratios. RESULTS: The overall levels of activity rose moderately over the five year study period, although the type of activity remained consistent. Surgical removal of teeth was the third most common activity of all surgical procedures. Patients who availed themselves of elective inpatient oral surgery were generally from a higher socio-economic group, in marked contrast to those patients admitted for emergency oral surgery. This observation was highlighted even further by the procedure 'removal of impacted wisdom tooth', which accounted for 41 per cent of all oral surgery activity. The most deprived communities (15 per cent of the regional population) utilised this service 50 per cent less than all other groups. CONCLUSIONS: The application of a deprivation index to hospital episode data enables more accurate assessment of service utilisation by differing socio-economic groups. The inequality of the oral surgery provision to the 15 per cent of the population which were most deprived adds to the overall debate on the appropriateness of this service. Further examination of the pattern of behaviour and referrals amongst primary dental care services is clearly needed.

Objectives. The aims of this retrospective study were (1) to compare the outcome of periapical surgery performed in endodontic and in oral surgery units of a teaching dental hospital and (2) to evaluate the influence of factors affecting outcome. Study design. A total of 176 teeth (endodontic unit, 83; oral surgery unit, 93) surgically treated more than 4 years previously were examined clinically and radiographically by means of strict criteria. Multiple logistic regression analysis was used. Results. The rate of complete healing for patients treated in the endodontic unit (37.4%) was significantly (P = .009) higher than that for patients treated in the oral surgery unit (19.4%). The technical quality of surgery (P < .001), placement of root-end filling (P = .039), absence of a preoperative periapical lesion (P = .042), absence of a post (P = .047), and presence of an adequate coronal restoration (P = .056, odds ratio = 3.71) had significant effects on treatment outcome. Conclusion. The technical quality of periapical surgery, the presence of a periapical lesion, and adequate apical and coronal seal are important prognostic determinants of successful periapical surgery. © 2001 Mosby, Inc.

Epidemiologists are often interested in examining the effect on a later-life outcome of an exposure measured repeatedly over the life course. When different hypotheses for this effect are proposed by competing theories, it is important to identify those most supported by observed data as a first step toward estimating causal associations. One method is to compare goodness-of-fit of hypothesized models with a saturated model, but it is unclear how to judge the “best” out of two hypothesized models that both pass criteria for a good fit. We developed a new method using the least absolute shrinkage and selection operator to identify which of a small set of hypothesized models explains most of the observed outcome variation. We analyzed a cohort study with repeated measures of socioeconomic position (exposure) through childhood, early- and mid-adulthood, and body mass index (outcome) measured in mid-adulthood. We confirmed previous findings regarding support or lack of support for the following hypotheses: accumulation (number of times exposed), three critical periods (only exposure in childhood, early- or mid-adulthood), and social mobility (transition from low to high socioeconomic position). Simulations showed that our least absolute shrinkage and selection operator approach identified the most suitable hypothesized model with high probability in moderately sized samples, but with lower probability for hypotheses involving change in exposure or highly correlated exposures. Identifying a single, simple hypothesis that represents the specified knowledge of the life course association allows more precise definition of the causal effect of interest.

Directed acyclic graphs (DAGs), which offer systematic representations of causal relationships, have become an established framework for the analysis of causal inference in epidemiology, often being used to determine covariate adjustment sets for minimizing confounding bias. DAGitty is a popular web application for drawing and analysing DAGs. Here we introduce the R package ‘dagitty’, which provides access to all of the capabilities of the DAGitty web application within the R platform for statistical computing, and also offers several new functions. We describe how the R package ‘dagitty’ can be used to: evaluate whether a DAG is consistent with the dataset it is intended to represent; enumerate ‘statistically equivalent’ but causally different DAGs; and identify exposure outcome adjustment sets that are valid for causally different but statistically equivalent DAGs. This functionality enables epidemiologists to detect causal misspecifications in DAGs and make robust inferences that remain valid for a range of different DAGs. The R package ‘dagitty’ is available through the comprehensive R archive network (CRAN) at [https://cran.r-project.org/web/packages/dagitty/]. The source code is available on github at [https://github.com/jtextor/dagitty]. The web application ‘DAGitty’ is free software, licensed under the GNU general public licence (GPL) version 2 and is available at [http:// dagitty.net/].

‘Unexplained residuals’ models have been used within lifecourse epidemiology to model an exposure measured longitudinally at several time points in relation to a distal outcome. It has been claimed that these models have several advantages, including: the ability to estimate multiple total causal effects in a single model, and additional insight into the effect on the outcome of greater-than-expected increases in the exposure compared to traditional regression methods. We evaluate these properties and prove mathematically how adjustment for confounding variables must be made within this modelling framework. Importantly, we explicitly place unexplained residual models in a causal framework using directed acyclic graphs. This allows for theoretical justification of appropriate confounder adjustment and provides a framework for extending our results to more complex scenarios than those examined in this paper. We also discuss several interpretational issues relating to unexplained residual models within a causal framework. We argue that unexplained residual models offer no additional insights compared to traditional regression methods, and, in fact, are more challenging to implement; moreover, they artificially reduce estimated standard errors. Consequently, we conclude that unexplained residual models, if used, must be implemented with great care.

Background Post hepatectomy liver failure (PHLF) remains a significant risk in patients undergoing curative liver resection for cancer, however currently available PHLF risk prediction investigations are not sufficiently accurate. The Hepatectomy risk assessment with functional magnetic resonance imaging trial (HEPARIM) aims to establish if quantitative MRI biomarkers of liver function & perfusion can be used to more accurately predict PHLF risk and FLR function, measured against indocyanine green (ICG) liver function test. Methods HEPARIM is an observational cohort study recruiting patients undergoing liver resection of 2 segments or more, prior to surgery patients will have both Dynamic Gadoxetate-enhanced (DGE) liver MRI and ICG testing. Day one post op ICG testing is repeated and R15 compared to the Gadoxetate Clearance (GC) of the future liver remnant (FLR-GC) as measure by preoperative DGE- MRI which is the primary outcome, and preoperative ICG R15 compared to GC of whole liver (WL-GC) as a secondary outcome. Data will be collected from medical records, biochemistry, pathology and radiology reports and used in a multi-variate analysis to the value of functional MRI and derive multivariant prediction models for future validation. Discussion If successful, this test will potentially provide an efficient means to quantitatively assess FLR function and PHLF risk enabling surgeons to push boundaries of liver surgery further while maintaining safe practice and thereby offering chance of cure to patients who would previously been deemed inoperable. MRI has the added benefit of already being part of the routine diagnostic pathway and as such would have limited additional burden on patients time or cost to health care systems. (Hepatectomy Risk Assessment With Functional Magnetic Resonance Imaging - Full Text View - ClinicalTrials.gov, n.d.) Trial registration ClinicalTrials.gov, ClinicalTrials.gov NCT04705194 - Registered 12th January 2021 – Retrospectively registered

There is a growing debate with regards to the appropriate methods of analysis of growth trajectories and their association with prospective dependent outcomes. Using the example of childhood growth and adult BP, we conducted an extensive simulation study to explore four two-stage and two joint modelling methods, and compared their bias and coverage in estimation of the (unconditional) association between birth length and later BP, and the association between growth rate and later BP (conditional on birth length). We show that the two-stage method of using multilevel models to estimate growth parameters and relating these to outcome gives unbiased estimates of the conditional associations between growth and outcome. Using simulations, we demonstrate that the simple methods resulted in bias in the presence of measurement error, as did the two-stage multilevel method when looking at the total (unconditional) association of birth length with outcome. The two joint modelling methods gave unbiased results, but using the re-inflated residuals led to undercoverage of the confidence intervals. We conclude that either joint modelling or the simpler two-stage multilevel approach can be used to estimate conditional associations between growth and later outcomes, but that only joint modelling is unbiased with nominal coverage for unconditional associations.

OBJECTIVES: The aims of this study were to model jointly the incidence rates of six smoking related cancers in the Yorkshire region of England, to explore the patterns of spatial correlation amongst them, and to estimate the relative weight of smoking and other shared risk factors for the relevant disease sites, both before and after adjustment for socioeconomic background (SEB). METHODS: Data on the incidence of oesophagus, stomach, pancreas, lung, kidney, and bladder cancers between 1983 and 2003 were extracted from the Northern & Yorkshire Cancer Registry database for the 532 electoral wards in the Yorkshire region. Using postcode of residence, each case was assigned an area-based measure of SEB using the Townsend index. Standardised incidence ratios (SIRs) were calculated for each cancer site and their correlations investigated. The joint analysis of the spatial variation in incidence used a Bayesian shared-component model. Three components were included to represent differences in smoking (for all six sites), bodyweight/obesity (for oesophagus, pancreas and kidney cancers) and diet/alcohol consumption (for oesophagus and stomach cancers). RESULTS: The incidence of cancers of the oesophagus, pancreas, kidney, and bladder was relatively evenly distributed across the region. The incidence of stomach and lung cancers was more clustered around the urban areas in the south of the region, and these two cancers were significantly associated with higher levels of area deprivation. The incidence of lung cancer was most impacted by adjustment for SEB, with the rural/urban split becoming less apparent. The component representing smoking had a larger effect on cancer incidence in the eastern part of the region. The effects of the other two components were small and disappeared after adjustment for SEB. CONCLUSION: This study demonstrates the feasibility of joint disease modelling using data from six cancer sites. Incidence estimates are more precise than those obtained without smoothing. This methodology may be an important tool to help authorities evaluate healthcare system performance and the impact of policies.

OBJECTIVE: To explain the concepts and application of multilevel modelling (MLM) and how it can be applied to the analysis of dental research data. BASIC DESIGN: Methodological in nature, this article introduces MLM through actual and hypothetical dental examples. SETTING: Examples of MLM in periodontal research illustrate cross-sectional and longitudinal analyses of hierarchical dental data. Multilevel multivariate modelling is illustrated by an example in orthodontic research. The potential implications of study design in the context MLM is extensively discussed, including study sample size and accounting for examiner variability. RESULTS: The fine detail and greater insight provided by MLM enables dental researchers to review and revise old hypotheses and to generate new ones that could not be addressed by single-level methods. CONCLUSIONS: MLM has the potential to increase our understanding of oral disease and health and offers the opportunity to rethink some aspects of dental research procedure.

OBJECTIVE: To introduce the concepts of random coefficient multilevel models through an application to periodontal research data. BASIC RESEARCH DESIGN: Multilevel models with random coefficients are illustrated using periodontal data that comprise four levels: repeated measurements at level-1, sites at level-2, teeth at level-3, and subjects at level-4. The study explores random coefficient models--where random variation occurs about explanatory variable coefficients. Outcomes considered are lifetime cumulative attachment loss and pocket probing depth. PARTICIPANTS: The study data were taken from a survey of periodontal disease involving 100 white male trainee engineers aged between 16 and 20 entering the apprentice training school at the Royal Air Force-Halton, UK. RESULTS: The application of multilevel modelling to longitudinal data provides a new way of exploring old problems. The multilevel random coefficient models provide an opportunity to examine the 'linear' and 'burst' theories of periodontal disease progression, leading to the postulation that both can be unified within the multilevel framework. CONCLUSIONS: The multilevel methodology illustrates how advances in the understanding of oral health can be achieved with the advent of new statistical methods

OBJECTIVE: To introduce the concepts of multilevel survival analysis through an investigation into the longevity of amalgam restorations. BASIC RESEARCH DESIGN: The multilevel Cox proportional hazard model is illustrated using amalgam restoration data comprising three levels: repeated restorations at level-1, teeth at level-2, and subjects at level-3. The outcome was duration of amalgam restoration survival. Single-level and multilevel Cox methods are contrasted. PARTICIPANTS: The data were from a survey of amalgam restorations (reported elsewhere), involving 200 RAF personnel aged between 16 and 37 years at enlistment between 1947 and 1979, having served continuously for a minimum of 16 years prior to 1994. RESULTS: Differences existed between single-level and multilevel methods; the latter being the method of choice. Initial caries experience was a good predictor of longevity. Molar teeth fared worse than pre-molars and MOD & B, MOD & L, and MOD & BL restorations experienced considerably greater risk of failure than did MOD, MO, DO and MO/DO ext types, which in turn fared worse than occlusal restorations. Root-treated and pinned teeth also experienced an elevated risk of premature failure. There was a moderate but significant increase in restoration failure amongst subjects who were seen by more dentists throughout their service. CONCLUSIONS: The application of multilevel modelling to survival analysis provides an appropriate and powerful solution to the problem of lack of independence amongst dental restorations. It is beneficial that studies undertake a multilevel analysis in preference to ignoring hierarchy or omitting swathes of information in order to perform a single-level analysis.

Objective: To explain the theory of multilevel modelling and demonstrate its application in the analysis of dental research data. Basic research design: Multilevel modelling was introduced using dental data comprising four levels: repeated measurements at level-1, sites at level-2, teeth at level-3, and subjects at level-4. Variance components models (which have no explanatory variables) were evaluated for all outcome measures. Explanatory variables were added to the models with outcomes for both lifetime cumulative attachment loss and pocket probing depth. Salient features of the multilevel models were discussed. Participants: Research data were obtained from a longitudinal survey of periodontal disease conducted on 100 white male trainee engineers aged between 16 and 20 years entering the apprentice training school at Royal Air Force Halton, England. Results: The statistical methods revealed that periodontal measures demonstrate considerable variation at all levels of the multilevel structure. Models for lifetime cumulative attachment loss and pocket probing depth illustrated that risk factors operated at more than one level. Supragingival calculus was a risk factor at the subject-level (subjects experiencing more sites with the condition had greater attachment loss and greater pocketing) whilst there was apparently a protective effect occurring at the site (sites with the condition had less attachment loss and less pocketing). Conclusions: This study demonstrates that multilevel modelling is a more powerful research tool than single-level techniques for the analysis of hierarchical dental data. Researchers using these techniques are well equipped to analyse complex hierarchical data structures, such as those often found within dentistry.

Objective: To demonstrate the use of multilevel multivariate modelling in the evaluation of multiple outcome dental cata. Basic research design: Multiple outcome dental research data are used to illustrate the problems of analysing such complex information structures - i.e. several outcomes clustered within subjects. Appropriate and statistically efficient methods of data analysis are proposed and illustrated step-by-step. The data structure is analysed using multilevel multivariate regression techniques and this process is discussed in comparison to conventional single-level multiple regression. Participants: Questionnaire data were obtained from an orthognathic study of 84 subjects seeking treatment and 106 'non-treatment' controls (full details of which are reported elsewhere). Results: Multivariate multiple regression analysis demonstrated a number of advantages over separate single-level multiple regression approaches, including a gain in statistical efficiency and greater insight into: a) the role of (significant) explanatory variables and b) outcome variable interactions. Multilevel multivariate analysis reduced the risk of both Type I and Type II statistical errors. Conclusions: The study demonstrates the benefit of multilevel multivariate modelling over conventional single-level techniques for statistical analysis of multiple outcome data. As a result of ongoing technical developments in the power, speed and memory of modern PCs, multilevel multivariate regression can now be undertaken with relative ease. Consequently, researchers are better equipped to analyse such complex data structures, particularly within dentistry where multivariate data are common.

Objective. To introduce the statistical methodology of meta-analysis within the framework of multilevel modelling (MLM) using an illustrative example. Basic research design. In meta-analysis it is important that the quantitative pooling of study results should be carried out in conjunction with careful consideration of the variation apparent between studies. If statistical heterogeneity is found to be significant, it is due, at least in part, to clinical heterogeneity. It is possible to account for clinical heterogeneity by including covariates that are thought to be responsible, using meta-regression. Clinical setting. A total of 38 studies of root canal treatment outcome were identified as being suitable for introducing the meta-analysis methodology. Two covariates were considered for modelling: a 'loose' or 'strict' (loose - incomplete radiographic healing; strict - complete radiographic healing) criterion for judging outcome of treatment and the year in which the study was performed. Results. There was considerable statistical heterogeneity between the study results. The effect of employing loose criteria for judging success significantly increased the probability of success when compared to employing strict criteria. Furthermore, the variance between studies was significantly reduced when this covariate was included in the modelling process when compared to the variation estimated in the model which did not consider covariates. Conclusion. MLM is a good facilitator for meta-analysis and meta-regression.

Previously, burst and linear theories for periodontal disease progression were proposed based on different but limited statistical methods of analysis. Multilevel modeling provides a new approach, yielding a more comprehensive model. Random coefficient models were used to analyze longitudinal periodontal data consisting of repeated measures (level 1), sites (level 2), teeth (level 3), and subjects (level 4). Large negative and highly significant correlations between random linear and quadratic time coefficients indicated that subjects and teeth with greater-than-average linear change experienced decelerated variation. Conversely, subjects and teeth with less-than-average linear change experienced accelerated variation. Change therefore exhibited a dynamic regression to the mean at the tooth and subject levels. Since no equilibrium was attained throughout the study, changes were cyclical. When considered as a multilevel system, the "linear" and "burst" theories of periodontal disease progression are a manifestation of the same phenomenon: Some sites improve while others progress, in a cyclical manner.

ABSTRACT

Background Four models are commonly used to adjust for energy intake when estimating the causal effect of a dietary component on an outcome: 1) the “standard model” adjusts for total energy intake, 2) the “energy partition model” adjusts for remaining energy intake, 3) the “nutrient density model” rescales the exposure as a proportion of total energy, and 4) the “residual model” indirectly adjusts for total energy by using a residual. It remains underappreciated that each approach evaluates a different estimand and only partially accounts for confounding by common dietary causes. Objectives We aimed to clarify the implied causal estimand and interpretation of each model and evaluate their performance in reducing dietary confounding. Methods Semiparametric directed acyclic graphs and Monte Carlo simulations were used to identify the estimands and interpretations implied by each model and explore their performance in the absence or presence of dietary confounding. Results The “standard model” and the mathematically identical “residual model” estimate the average relative causal effect (i.e., a “substitution” effect) but provide biased estimates even in the absence of confounding. The “energy partition model” estimates the total causal effect but only provides unbiased estimates in the absence of confounding or when all other nutrients have equal effects on the outcome. The “nutrient density model” has an obscure interpretation but attempts to estimate the average relative causal effect rescaled as a proportion of total energy. Accurate estimates of both the total and average relative causal effects may instead be derived by simultaneously adjusting for all dietary components, an approach we term the “all-components model.” Conclusions Lack of awareness of the estimand differences and accuracy of the 4 modeling approaches may explain some of the apparent heterogeneity among existing nutritional studies. This raises serious questions regarding the validity of meta-analyses where different estimands have been inappropriately pooled.

This article discusses three statistical paradoxes that pervade epidemiological research: Simpson's paradox, Lord's paradox, and suppression. These paradoxes have important implications for the interpretation of evidence from observational studies. This article uses hypothetical scenarios to illustrate how the three paradoxes are different manifestations of one phenomenon - the reversal paradox - depending on whether the outcome and explanatory variables are categorical, continuous or a combination of both; this renders the issues and remedies for any one to be similar for all three. Although the three statistical paradoxes occur in different types of variables, they share the same characteristic: the association between two variables can be reversed, diminished, or enhanced when another variable is statistically controlled for. Understanding the concepts and theory behind these paradoxes provides insights into some controversial or contradictory research findings. These paradoxes show that prior knowledge and underlying causal theory play an important role in the statistical modelling of epidemiological data, where incorrect use of statistical models might produce consistent, replicable, yet erroneous results. © 2008 Tu et al; licensee BioMed Central Ltd.

Epidemiological studies focusing on the etiology of childhood chronic diseases have used population mixing as a proxy for the level of infection circulating in a community. We compared different measures of population mixing (based on residential migration and commuting) and other demographic variables, derived from the United Kingdom Census, with hospital inpatient data on infections from two Government Office Regions in England (Eastern and the West Midlands) to inform the development of an infectious disease proxy for future epidemiological studies. The association between rates of infection and the population mixing measures was assessed, using incidence rate ratios across census areas, from negative binomial regression. Commuting distance demonstrated the most consistent association with admissions for infections across the two regions; areas with a higher median distance travelled by commuters leaving the area having a lower rate of hospital admissions for infections. Deprived areas and densely populated areas had a raised rate of admissions for infections. Assuming hospital admissions are a reliable indicator of common infection rates, the results from this study suggest that commuting distance is a consistent measure of population mixing in relation to infectious disease and deprivation and population density are reliable demographic proxies for infectious exposure. Areas that exhibit high levels of population mixing do not necessarily possess raised rates of hospital admissions for infectious disease.

Autoregressive and moving average models for temporally dynamic networks treat time as a series of discrete steps which assumes even intervals between data measurements and can introduce bias if this assumption is not met. Using real and simulated data from the London Underground network, this paper illustrates the use of continuous time multilevel models to capture temporal trajectories of edge properties without the need for simultaneous measurements, along with two methods for producing interpretable summaries of model results. These including extracting ‘features’ of temporal patterns (e.g. maxima, time of maxima) which have utility in understanding the network properties of each connection and summarising whole-network properties as a continuous function of time which allows estimation of network properties at any time without temporal aggregation of non-simultaneous measurements. Results for temporal pattern features in the response variable were captured with reasonable accuracy. Variation in the temporal pattern features for the exposure variable was underestimated by the models. The models showed some lack of precision. Both model summaries provided clear ‘real-world’ interpretations and could be applied to data from a range of spatio-temporal network structures (e.g. rivers, social networks). These models should be tested more extensively in a range of scenarios, with potential improvements such as random effects in the exposure variable dimension.

The aim of this article is to encourage good practice in the statistical analysis of dental research data. Our objective is to highlight the statistical problems of collinearity and multicollinearity. These are among the most common statistical pitfalls in oral health research when exploring the relationship between clinical variables using multiple regression analysis. We hope that this article will show why these problems arise and how they can be avoided and overcome. Examples from the periodontal literature will be used to illustrate how collinearity and multicollinearity can seriously distort the model development process as a result of the phenomenon of mathematical coupling. Knowledge of these problems can help to eliminate misleading results and improve any subsequent interpretations. Regression analyses are useful tools in oral health research when their limitations are recognized. However, care is required in planning and it is worthwhile seeking statistical advice when formulating the study’s research questions.

The current paradigm for causal inference in epidemiology relies primarily on the evaluation of counterfactual contrasts via statistical regression models informed by graphical causal models (often in the form of directed acyclic graphs, or DAGs) and their underlying mathematical theory. However, there have been growing calls for supplementary methods, and one such method that has been proposed is agent-based modelling due to its potential for simulating counterfactuals. However, within the epidemiological literature there currently exists a general lack of clarity regarding what exactly agent-based modelling is (and is not) and, importantly, how it differs from microsimulation modelling – perhaps its closest methodological comparator. We clarify this distinction by briefly reviewing the history of each method, which provides context for their similarities and differences, and casts light on the types of research questions that they have evolved (and thus are well-suited) to answering; we do the same for DAG-informed regression methods. The distinct historical evolutions of DAG-informed regression modelling, microsimulation modelling, and agent-based modelling have given rise to distinct features of the methods themselves, and provide a foundation for critical comparison. Not only are the three methods well-suited to addressing different types of causal questions, but in doing so they place differing levels of emphasis on fixed and random effects, and also tend to operate on different timescales and in different timeframes.