Bias and efficiency of multiple imputation compared with complete-case analysis for missing covariate values

Determinants of attaining and maintaining a low cardiovascular risk profile--the Doetinchem Cohort Study

BACKGROUND

While maintenance of a low cardiovascular risk profile is essential for cardiovascular disease (CVD) prevention, few people maintain a low CVD risk profile throughout their life. We studied the association of demographic, lifestyle, psychological factors and family history of CVD with attainment and maintenance of a low risk profile over three subsequent 5-year periods.

METHODS

Measurements of 6390 adults aged 26-65 years at baseline were completed from 1993 to 97 and subsequently at 5-year intervals until 2013. At each wave, participants were categorized into low risk profile (ideal levels of blood pressure, cholesterol and body mass index, non-smoking and no diabetes) and medium/high risk profile (all others). Multivariable-adjusted modified Poisson regression analyses were used to examine determinants of attainment and maintenance of low risk; risk ratios (RR) and 95% confidence intervals (95% CI) were obtained. Generalized estimating equations were used to combine multiple 5-year comparisons.

RESULTS

Younger age, female gender and high educational level were associated with higher likelihood of both maintaining and attaining low risk profile (P < 0.05). In addition, likelihood of attaining low risk was 9% higher with each 1-unit increment in Mediterranean diet score (RR: 1.09, 95% CI: 1.02-1.16), twice as high with any physical activity versus none (RR: 2.17, 95% CI: 1.16-4.04) and 35% higher with moderate alcohol consumption versus heavy consumption (RR: 1.35, 95% CI: 1.06-1.73).

CONCLUSION

Healthy lifestyle factors such as adherence to a Mediterranean diet, physical activity and moderate as opposed to heavy alcohol consumption were associated with a higher likelihood of attaining a low risk profile.

Using decision trees to understand structure in missing data

OBJECTIVES

Demonstrate the application of decision trees--classification and regression trees (CARTs), and their cousins, boosted regression trees (BRTs)--to understand structure in missing data.

SETTING

Data taken from employees at 3 different industrial sites in Australia.

PARTICIPANTS

7915 observations were included.

MATERIALS AND METHODS

The approach was evaluated using an occupational health data set comprising results of questionnaires, medical tests and environmental monitoring. Statistical methods included standard statistical tests and the 'rpart' and 'gbm' packages for CART and BRT analyses, respectively, from the statistical software 'R'. A simulation study was conducted to explore the capability of decision tree models in describing data with missingness artificially introduced.

RESULTS

CART and BRT models were effective in highlighting a missingness structure in the data, related to the type of data (medical or environmental), the site in which it was collected, the number of visits, and the presence of extreme values. The simulation study revealed that CART models were able to identify variables and values responsible for inducing missingness. There was greater variation in variable importance for unstructured as compared to structured missingness.

DISCUSSION

Both CART and BRT models were effective in describing structural missingness in data. CART models may be preferred over BRT models for exploratory analysis of missing data, and selecting variables important for predicting missingness. BRT models can show how values of other variables influence missingness, which may prove useful for researchers.

CONCLUSIONS

Researchers are encouraged to use CART and BRT models to explore and understand missing data.

Predictors of anemia in a multi-ethnic chronic kidney disease population: a case-control study

Anemia is a common complication of chronic kidney disease (CKD). However, risk factors of anemia in CKD patients in Singapore are not well established. Hence, a retrospective, case–control study involving non-dialysis CKD patients was conducted to determine possible predictors of anemia in the local CKD population.

Non-dialysis adult CKD patients, not receiving renal replacement therapy or erythropoiesis-stimulating-agents were included. Parameters collected included demographics e.g. age, sex and race; clinical data e.g. CKD stage and medical/medication histories; and laboratory data e.g. serum electrolytes, urinary and hematologic parameters. Patients were classified as anemic or non-anemic using a threshold hemoglobin level of 10 g/dL. The parameters were evaluated for their predictive value for anemia development using multivariate logistical regression and calculation of odds ratios. Statistical analyses were performed using STATA.

A total of 457 patients (162 anemic and 295 non-anemic) were analysed. Multivariate analysis showed that probability of developing anemia was greater for patients with stage 5 CKD (OR 16.76, p < 0.001), with hematological disorders (OR 18.61, p < 0.001) and with respiratory disorders (OR 4.54, p = 0.004). The probability of developing anemia was lower for patients with higher previous hemoglobin concentration (OR 0.32, p < 0.001) and in those receiving iron supplements (OR 0.44, p = 0.031). Gender and race were not found to be significant predictors of anemia.

Risk of anemia is increased in patients with advanced CKD, haematological disorders, respiratory disorders, and those not taking iron supplements. This study has increased our understanding of the patient subgroups at risk for anemia.

Statistical analysis and handling of missing data in cluster randomised trials: protocol for a systematic review

INTRODUCTION

Cluster randomised trials (CRTs) randomise participants in groups, rather than as individuals, and are key tools used to assess interventions in health research where treatment contamination is likely or if individual randomisation is not feasible. Missing outcome data can reduce power in trials, including in CRTs, and is a potential source of bias. The current review focuses on evaluating methods used in statistical analysis and handling of missing data with respect to the primary outcome in CRTs.

METHODS AND ANALYSIS

We will search for CRTs published between August 2013 and July 2014 using PubMed, Web of Science and PsycINFO. We will identify relevant studies by screening titles and abstracts, and examining full-text articles based on our predefined study inclusion criteria. 86 studies will be randomly chosen to be included in our review. Two independent reviewers will collect data from each study using a standardised, prepiloted data extraction template. Our findings will be summarised and presented using descriptive statistics.

ETHICS AND DISSEMINATION

This methodological systematic review does not need ethical approval because there are no data used in our study that are linked to individual patient data. After completion of this systematic review, data will be immediately analysed, and findings will be disseminated through a peer-reviewed publication and conference presentation.

Randomized Controlled Trial of Individualized Dialysate Cooling for Cardiac Protection in Hemodialysis Patients

BACKGROUND AND OBJECTIVES

Cardiovascular disease is the most common cause of death in patients on hemodialysis (HD). HD-associated cardiomyopathy is appreciated to be driven by exposure to recurrent and cumulative ischemic insults resulting from hemodynamic instability of conventionally performed intermittent HD treatment itself. Cooled dialysate reduces HD-induced recurrent ischemic injury, but whether this confers long-term protection of the heart in terms of cardiac structure and function is not known.

DESIGN, SETTING, PARTICIPANTS, & MEASUREMENTS

Between September 2009 and January 2013, 73 incident HD patients were randomly assigned to a dialysate temperature of 37°C (control) or individualized cooling at 0.5°C below body temperature (intervention) for 12 months. Cardiac structure, function, and aortic distensibility were assessed by cardiac magnetic resonance imaging. Mean between-group difference in delivered dialysate temperature was 1.2°C±0.3°C. Treatment effects were determined by the interaction of treatment group with time in linear mixed models.

RESULTS

There was no between-group difference in the primary outcome of left ventricular ejection fraction (1.5%; 95% confidence interval, -4.3% to 7.3%). However, left ventricular function assessed by peak systolic strain was preserved by the intervention (-3.3%; 95% confidence interval, -6.5% to -0.2%) as was diastolic function (measured as peak diastolic strain rate, 0.18 s(-1); 95% confidence interval, 0.02 to 0.34 s(-1)). Reduction of left ventricular dilation was demonstrated by significant reduction in left ventricular end-diastolic volume (-23.8 ml; 95% confidence interval, -44.7 to -2.9 ml). The intervention was associated with reduced left ventricular mass (-15.6 g; 95% confidence interval, -29.4 to -1.9 g). Aortic distensibility was preserved in the intervention group (1.8 mmHg(-1)×10(-3); 95% confidence interval, 0.1 to 3.6 mmHg(-1)×10(-3)). There were no intervention-related withdrawals or adverse events.

CONCLUSIONS

In patients new to HD, individualized cooled dialysate did not alter the primary outcome but was well tolerated and slowed the progression of HD-associated cardiomyopathy. Because cooler dialysate is universally applicable at no cost, the intervention warrants wider adoption or confirmation of these findings in a larger trial.

Effectiveness of biologic DMARDs in monotherapy versus in combination with synthetic DMARDs in rheumatoid arthritis: data from the Swiss Clinical Quality Management Registry

OBJECTIVES

To determine the frequency of use of biologic DMARDs (bDMARDs) in monotherapy, to describe the baseline characteristics of patients treated with bDMARDs in monotherapy and to compare the effectiveness of bDMARDs in monotherapy with that of bDMARDs in combination with synthetic DMARDs (sDMARDs).

METHODS

Using data from the Swiss RA (SCQM-RA) registry, bDMARD treatment courses (TCs) were classified either as monotherapy or as combination therapy, depending on the presence of concomitant sDMARDs. Prescription of bDMARD monotherapy was analysed using logistic regression. bDMARD retention was analysed using Kaplan-Meier and Cox models with the addition of time-varying covariate effects. Evolution of the DAS28 over time was analysed with mixed-effects models for longitudinal data.

RESULTS

A total of 4218 TCs on bDMARDs from 3111 patients were included, of which 1136 TCs (27%) were initiated as monotherapy. bDMARD monotherapy was preferentially prescribed to older, co-morbid patients with longer disease duration, lower BMI, more active disease and more previous bDMARDs. After adjusting for potential confounding factors, drug retention was significantly lower in monotherapy [hazard ratio 1.15 (95% CI: 1.03, 1.30)]. Other factors such as type of bDMARD and calendar year of prescription were associated with a stronger effect on drug retention. Response to treatment in terms of DAS28 evolution was also slightly but significantly less favourable in monotherapy (P = 0.04).

CONCLUSION

Our data suggest that bDMARD monotherapy is prescribed to more complex cases and is significantly less effective than bDMARD therapy in combination with sDMARDs, but to an extent that is clinically only marginally relevant.

A systematic review of the statistical methods in prospective cohort studies investigating the effect of medications on cognition in older people

BACKGROUND

There is increasing awareness that medications can contribute to cognitive decline. Prospective cohort studies are rich sources of clinical data. However, investigating the contribution of medications to cognitive decline is challenging because both medication exposure and cognitive impairment can be associated with attrition of study participants, and medication exposure status may change over time. The objective of this review was to investigate the statistical methods in prospective cohort studies assessing the effect of medications on cognition in older people.

METHODS

A systematic literature search was conducted to identify prospective cohort studies of at least 12 months duration that investigated the effect of common medications or medication classes (anticholinergics, antihistamines, hypnotics, sedatives, opioids, statins, estrogens, testosterone, antipsychotics, anticonvulsants, antidepressants, anxiolytics, antiparkinson agents and bronchodilators) on cognition in people aged 65 years and older. Data extraction was performed independently by two investigators. A descriptive analysis of the statistical methods was performed.

RESULTS

A total of 44 articles were included in the review. The most common statistical methods were logistic regression (24.6% of all reported methods), Cox proportional hazards regression (22.8%), linear mixed-effects models (21.1%) and multiple linear regression (14.0%). The use of advanced techniques, most notably linear mixed-effects models, increased over time. Only 6 articles (13.6%) reported methods for addressing missing data.

CONCLUSIONS

A variety of statistical methods have been used for investigating the effect of medications on cognition in older people. While advanced techniques that are appropriate for the analysis of longitudinal data, most notably linear mixed-effects models, have increasingly been employed in recent years, there is an opportunity to implement alternative techniques in future studies that could address key research questions.

The rise of multiple imputation: a review of the reporting and implementation of the method in medical research

Background

Missing data are common in medical research, which can lead to a loss in statistical power and potentially biased results if not handled appropriately. Multiple imputation (MI) is a statistical method, widely adopted in practice, for dealing with missing data. Many academic journals now emphasise the importance of reporting information regarding missing data and proposed guidelines for documenting the application of MI have been published. This review evaluated the reporting of missing data, the application of MI including the details provided regarding the imputation model, and the frequency of sensitivity analyses within the MI framework in medical research articles.

Methods

A systematic review of articles published in the Lancet and New England Journal of Medicine between January 2008 and December 2013 in which MI was implemented was carried out.

Results

We identified 103 papers that used MI, with the number of papers increasing from 11 in 2008 to 26 in 2013. Nearly half of the papers specified the proportion of complete cases or the proportion with missing data by each variable. In the majority of the articles (86%) the imputed variables were specified. Of the 38 papers (37%) that stated the method of imputation, 20 used chained equations, 8 used multivariate normal imputation, and 10 used alternative methods. Very few articles (9%) detailed how they handled non-normally distributed variables during imputation. Thirty-nine papers (38%) stated the variables included in the imputation model. Less than half of the papers (46%) reported the number of imputations, and only two papers compared the distribution of imputed and observed data. Sixty-six papers presented the results from MI as a secondary analysis. Only three articles carried out a sensitivity analysis following MI to assess departures from the missing at random assumption, with details of the sensitivity analyses only provided by one article.

Conclusions

This review outlined deficiencies in the documenting of missing data and the details provided about imputation. Furthermore, only a few articles performed sensitivity analyses following MI even though this is strongly recommended in guidelines. Authors are encouraged to follow the available guidelines and provide information on missing data and the imputation process.

Electronic supplementary material

The online version of this article (doi:10.1186/s12874-015-0022-1) contains supplementary material, which is available to authorized users.

Analytical results in longitudinal studies depended on target of inference and assumed mechanism of attrition

OBJECTIVES

To compare methods for analysis of longitudinal studies with missing data due to participant dropout and follow-up truncated by death.

STUDY DESIGN AND SETTING

We analyzed physical functioning in an Australian longitudinal study of elderly women where the missing data mechanism could either be missing at random (MAR) or missing not at random (MNAR). We assumed either an immortal cohort where deceased participants are implicitly included after death or a mortal cohort where the target of inference is surviving participants at each survey wave. To illustrate the methods a covariate was included. Simulation was used to assess the effect of the assumptions.

RESULTS

Ignoring attrition or restricting analysis to participants with complete follow up led to biased estimates. Linear mixed model was appropriate for an immortal cohort under MAR but not MNAR. Linear increment model and joint modeling of longitudinal outcome and time to death were the most robust to MNAR. For a mortal cohort, inverse probability weighting and multiple imputation could be used, but care is needed in specifying dropout and imputation models, respectively.

CONCLUSION

Appropriate analysis methodology to deal with attrition in longitudinal studies depends on the target of inference and the missing data mechanism.

Practical guidance for conducting mediation analysis with multiple mediators using inverse odds ratio weighting

Despite the recent flourishing of mediation analysis techniques, many modern approaches are difficult to implement or applicable to only a restricted range of regression models. This report provides practical guidance for implementing a new technique utilizing inverse odds ratio weighting (IORW) to estimate natural direct and indirect effects for mediation analyses. IORW takes advantage of the odds ratio's invariance property and condenses information on the odds ratio for the relationship between the exposure (treatment) and multiple mediators, conditional on covariates, by regressing exposure on mediators and covariates. The inverse of the covariate-adjusted exposure-mediator odds ratio association is used to weight the primary analytical regression of the outcome on treatment. The treatment coefficient in such a weighted regression estimates the natural direct effect of treatment on the outcome, and indirect effects are identified by subtracting direct effects from total effects. Weighting renders treatment and mediators independent, thereby deactivating indirect pathways of the mediators. This new mediation technique accommodates multiple discrete or continuous mediators. IORW is easily implemented and is appropriate for any standard regression model, including quantile regression and survival analysis. An empirical example is given using data from the Moving to Opportunity (1994-2002) experiment, testing whether neighborhood context mediated the effects of a housing voucher program on obesity. Relevant Stata code (StataCorp LP, College Station, Texas) is provided.

Small-for-gestational-age birth and maternal plasma antioxidant levels in mid-gestation: a nested case-control study

OBJECTIVE

To assess whether maternal plasma antioxidant levels in mid-pregnancy are associated with small-for-gestational-age (SGA) birth.

DESIGN

Case-control study nested within a population-based cohort study.

SETTING

Four hospitals in Montreal, Canada.

POPULATION

Pregnant women recruited before 24 weeks of gestation, whose pregnancies were not complicated by pre-eclampsia or preterm delivery.

METHODS

Blood samples were obtained at 24-26 weeks and assayed for nutritionally derived antioxidant levels in SGA cases (n = 324) and randomly selected controls with birthweights between the 25th and 75th centiles (n = 672). We performed logistic regression analyses using the standardised z-score of each antioxidant as the main independent variable, after summing highly correlated antioxidants or combining via principle component analysis. We adjusted for risk factors for SGA that were associated with antioxidant levels.

MAIN OUTCOME MEASURES

SGA, birthweight <10th centile for gestational age and sex.

RESULTS

Retinol was positively associated with risk of SGA (adjusted odds ratio [OR] 1.41; 95% confidence interval [95% CI] 1.22-1.63, per SD increase). Carotenoids (log of the sum of β-carotene, lutein/zeaxanthin, α- and β-cryptoxanthin) were negatively associated with SGA (adjusted OR 0.64; 95% CI 0.54-0.78, per SD increase). We found no significant associations between SGA and lycopene or any of the forms of vitamin E assessed, including α-tocopherol, corrected α-tocopherol (per nmol/l of low-density lipoprotein articles), or γ-tocopherol.

CONCLUSIONS

Elevated retinol may be associated with an increased risk of SGA, whereas elevated carotenoid levels may reduce the risk. A better understanding of the nature of these associations is required, however, before recommending specific nutritional interventions in an attempt to prevent SGA birth.

Higher physical activity is associated with lower aortic stiffness but not with central blood pressure: the ADDITION-Pro Study

Physical activity is associated with reduced cardiovascular disease risk. However, improvements in conventional risk factors due to physical activity do not explain its full benefit. Therefore, we examined associations of objectively measured physical activity energy expenditure and intensity with central hemodynamics to provide new insight into the link between physical activity and cardiovascular disease. We analyzed data from 1816 Danes (median age: 66 years) without cardiovascular disease. Physical activity was estimated using combined accelerometry and heart rate monitoring. Aortic stiffness was assessed by applanation tonometry, as aortic pulse wave velocity, and central blood pressure was estimated from radial waveforms. Associations between physical activity energy expenditure and central hemodynamics were examined by linear regression. Furthermore, the consequence of substituting 1 hour sedentary behavior with 1 hour light or moderate-to-vigorous physical activity on central hemodynamics was examined. Median physical activity energy expenditure was 28.0 kJ/kg/d (IQR: 19.8; 38.7). A 10 kJ/kg/d higher energy expenditure was associated with 0.75% lower aortic pulse wave velocity (CI: -1.47; -0.03). Associations with central systolic blood pressure and central pulse pressure were not statistically significant. We observed no difference in central hemodynamics when substituting 1 hour sedentary behavior with 1 hour light or moderate-to-vigorous physical activity. In this relatively inactive population, higher physical activity energy expenditure was associated with lower aortic stiffness, while there was no statistically significant association between substitution of activity intensity and central hemodynamics. This suggests that lower aortic stiffness is one of a number of health benefits attributed to higher habitual physical activity.

Cost analysis of early psychosocial intervention in Alzheimer's disease

BACKGROUND/AIM

To investigate the impact of early psychosocial intervention aimed at patients with Alzheimer's disease (AD) and their caregivers on resource use and costs from a societal perspective.

METHODS

Dyads of patients and their primary caregiver were randomised to intervention (n = 163) or control (n = 167) and followed for 3 years. Health care use was extracted from national registers, and the Resource Utilisation in Dementia questionnaire was used to measure informal care and productivity loss. Multiple imputation was used to replace missing data, and non-parametric bootstrapping was used to estimate standard errors.

RESULTS

Overall, there were no statistically significant differences because of large variation in the observations. The average additional cost of psychosocial intervention provision was estimated at EUR 3,401 per patient. This cost masked a reduced use of formal health care and an increased use of informal care.

CONCLUSIONS

Early psychosocial intervention in AD could be cost-saving from a health care perspective, whereas the opposite seems to be true from a broader societal perspective.

Chronic multisymptom illness: a comparison of Iraq and Afghanistan deployers with veterans of the 1991 Gulf War

Symptoms and illnesses reported by veterans of the 1991 Gulf War era are a cause of potential concern for those military members who have deployed to the Gulf region in support of more recent contingency operations in Iraq and Afghanistan. In the present study, we quantified self-reported symptoms from participants in the Millennium Cohort Study, a prospective study representing all US service branches, including both active duty and Reserve/National Guard components (2001-2008). Self-reported symptoms were uniquely compared with those in a cohort of subjects from the 1991 Gulf War to gain context for the present report. Symptoms were then aggregated to identify cases of chronic multisymptom illness (CMI) based on the case definition from the Centers for Disease Control and Prevention. The prevalence of self-reported CMI symptoms was compared with that collected in 1997-1999 from a study population of US Seabees from the 1991 Gulf War, as well as from deployed and nondeployed subgroups. Although overall symptom reporting was much less in the Millennium Cohort than in the 1991 Gulf War cohort, a higher prevalence of reported CMI was noted among deployed compared with nondeployed contemporary cohort members. An increased understanding of coping skills and resilience and development of well-designed screening instruments, along with appropriate clinical and psychological follow-up for returning veterans, might help to focus resources on early identification of potential long-term chronic disease manifestations.

Handling incomplete smoking history data in survival analysis

While data are unavoidably missing or incomplete in most observational studies, consequences of mishandling such incompleteness in analysis are often overlooked. When time-varying information is collected irregularly and infrequently over a long period, even precisely obtained data may implicitly involve substantial incompleteness. Motivated by an analysis to quantitatively evaluate the effects of smoking and radiation on lung cancer risks among Japanese atomic-bomb survivors, we provide a unique application of multiple imputation to incompletely observed smoking histories under the assumption of missing at random. Predicting missing values for the age of smoking initiation and, given initiation, smoking intensity and cessation age, analyses can be based on complete, though partially imputed, smoking histories. A simulation study shows that multiple imputation appropriately conditioned on the outcome and other relevant variables can produce consistent estimates when data are missing at random. Our approach is particularly appealing in large cohort studies where a considerable amount of time-varying information is incomplete under a mechanism depending in a complex manner on other variables. In application to the motivating example, this approach is expected to reduce estimation bias that might be unavoidable in naive analyses, while keeping efficiency by retaining known information.

Mind the gap: knowledge and practice of providers treating uncomplicated malaria at public and mission health facilities, pharmacies and drug stores in Cameroon and Nigeria

Background Artemisinin combination therapy (ACT) has been the first-line treatment for uncomplicated malaria in Cameroon since 2004 and Nigeria since 2005, though many febrile patients receive less effective antimalarials. Patients often rely on providers to select treatment, and interventions are needed to improve providers’ practice and encourage them to adhere to clinical guidelines.

Methods Providers’ adherence to malaria treatment guidelines was examined using data collected in Cameroon and Nigeria at public and mission facilities, pharmacies and drug stores. Providers’ choice of antimalarial was investigated separately for each country. Multilevel logistic regression was used to determine whether providers were more likely to choose ACT if they knew it was the first-line antimalarial. Multiple imputation was used to impute missing data that arose when linking exit survey responses to details of the provider responsible for selecting treatment.

Results There was a gap between providers’ knowledge and their practice in both countries, as providers’ decision to supply ACT was not significantly associated with knowledge of the first-line antimalarial. Providers were, however, more likely to supply ACT if it was the type of antimalarial they prefer. Other factors were country-specific, and indicated providers can be influenced by what they perceived their patients prefer or could afford, as well as information about their symptoms, previous treatment, the type of outlet and availability of ACT.

Conclusions Public health interventions to improve the treatment of uncomplicated malaria should strive to change what providers prefer, rather than focus on what they know. Interventions to improve adherence to malaria treatment guidelines should emphasize that ACT is the recommended antimalarial, and it should be used for all patients with uncomplicated malaria. Interventions should also be tailored to the local setting, as there were differences between the two countries in providers’ choice of antimalarial, and who or what influenced their practice.

Improving upon the efficiency of complete case analysis when covariates are MNAR

Missing values in covariates of regression models are a pervasive problem in empirical research. Popular approaches for analyzing partially observed datasets include complete case analysis (CCA), multiple imputation (MI), and inverse probability weighting (IPW). In the case of missing covariate values, these methods (as typically implemented) are valid under different missingness assumptions. In particular, CCA is valid under missing not at random (MNAR) mechanisms in which missingness in a covariate depends on the value of that covariate, but is conditionally independent of outcome. In this paper, we argue that in some settings such an assumption is more plausible than the missing at random assumption underpinning most implementations of MI and IPW. When the former assumption holds, although CCA gives consistent estimates, it does not make use of all observed information. We therefore propose an augmented CCA approach which makes the same conditional independence assumption for missingness as CCA, but which improves efficiency through specification of an additional model for the probability of missingness, given the fully observed variables. The new method is evaluated using simulations and illustrated through application to data on reported alcohol consumption and blood pressure from the US National Health and Nutrition Examination Survey, in which data are likely MNAR independent of outcome.

Doing more than Just Acknowledging Attrition at Follow-Up: A Comment on Lu, Cheng, and Chen (2013)

Lu, Cheng, and Chen (2013) faced one of the most common challenges encountered in longitudinal studies: follow-up attrition. Using a correlational prospective design, 464 volunteers completed a questionnaire that measured the constructs of the theory of planned behavior, and subsequently 154 of them provided physical activity data at a 6-month follow-up. The proportion of participants (66.8%) for whom the investigators were not able to gather information on the behavioral outcome at follow-up may reflect a form of selection bias that may affect both the validity and generalizability of study results. Lu, et al.'s (2013) study is used here to explore the implication of follow-up attrition on the results and inference, to review what information should be reported in a scientific paper in such situations, and to give practical tips to handle follow-up attrition.

Body-mass index and risk of 22 specific cancers: a population-based cohort study of 5·24 million UK adults

BACKGROUND

High body-mass index (BMI) predisposes to several site-specific cancers, but a large-scale systematic and detailed characterisation of patterns of risk across all common cancers adjusted for potential confounders has not previously been undertaken. We aimed to investigate the links between BMI and the most common site-specific cancers.

METHODS

With primary care data from individuals in the Clinical Practice Research Datalink with BMI data, we fitted Cox models to investigate associations between BMI and 22 of the most common cancers, adjusting for potential confounders. We fitted linear then non-linear (spline) models; investigated effect modification by sex, menopausal status, smoking, and age; and calculated population effects.

FINDINGS

5·24 million individuals were included; 166,955 developed cancers of interest. BMI was associated with 17 of 22 cancers, but effects varied substantially by site. Each 5 kg/m(2) increase in BMI was roughly linearly associated with cancers of the uterus (hazard ratio [HR] 1·62, 99% CI 1·56-1·69; p<0·0001), gallbladder (1·31, 1·12-1·52; p<0·0001), kidney (1·25, 1·17-1·33; p<0·0001), cervix (1·10, 1·03-1·17; p=0·00035), thyroid (1·09, 1·00-1·19; p=0·0088), and leukaemia (1·09, 1·05-1·13; p≤0·0001). BMI was positively associated with liver (1·19, 1·12-1·27), colon (1·10, 1·07-1·13), ovarian (1·09, 1.04-1.14), and postmenopausal breast cancers (1·05, 1·03-1·07) overall (all p<0·0001), but these effects varied by underlying BMI or individual-level characteristics. We estimated inverse associations with prostate and premenopausal breast cancer risk, both overall (prostate 0·98, 0·95-1·00; premenopausal breast cancer 0·89, 0·86-0·92) and in never-smokers (prostate 0·96, 0·93-0·99; premenopausal breast cancer 0·89, 0·85-0·94). By contrast, for lung and oral cavity cancer, we observed no association in never smokers (lung 0·99, 0·93-1·05; oral cavity 1·07, 0·91-1·26): inverse associations overall were driven by current smokers and ex-smokers, probably because of residual confounding by smoking amount. Assuming causality, 41% of uterine and 10% or more of gallbladder, kidney, liver, and colon cancers could be attributable to excess weight. We estimated that a 1 kg/m(2) population-wide increase in BMI would result in 3790 additional annual UK patients developing one of the ten cancers positively associated with BMI.

INTERPRETATION

BMI is associated with cancer risk, with substantial population-level effects. The heterogeneity in the effects suggests that different mechanisms are associated with different cancer sites and different patient subgroups.

FUNDING

National Institute for Health Research, Wellcome Trust, and Medical Research Council.

Predicting outcomes of neuroreflexotherapy in patients with subacute or chronic neck or low back pain

BACKGROUND CONTEXT

In the context of shared decision-making, a valid estimation of the probability that a given patient will improve after a specific treatment is valuable.

PURPOSE

To develop models that predict the improvement of spinal pain, referred pain, and disability in patients with subacute or chronic neck or low back pain undergoing a conservative treatment.

STUDY DESIGN AND SETTING

Analysis of data from a prospective registry in routine practice.

PATIENT SAMPLE

All patients who had been discharged after receiving a conservative treatment within the Spanish National Health Service (SNHS) (n=8,778).

OUTCOME MEASURES

Spinal pain, referred pain, and disability were assessed before the conservative treatment and at discharge by the use of previously validated methods.

METHODS

Improvement in spinal pain, referred pain, and disability was defined as a reduction in score greater than the minimal clinically important change. A predictive model that included demographic, clinical, and work-related variables was developed for each outcome using multivariate logistic regression. Missing data were addressed using multiple imputation. Discrimination and calibration were assessed for each model. The models were validated by bootstrap, and nomograms were developed.

RESULTS

The following variables showed a predictive value in the three models: baseline scores for pain and disability, pain duration, having undergone X-ray, having undergone spine surgery, and receiving financial assistance for neck or low back pain. Discrimination of the three models ranged from slight to moderate, and calibration was good.

CONCLUSIONS

A registry in routine practice can be used to develop models that estimate the probability of improvement for each individual patient undergoing a specific form of treatment. Generalizing this approach to other treatments can be valuable for shared decision making.

Quantifying the benefits of achieving or maintaining long-term low risk profile for cardiovascular disease: The Doetinchem Cohort Study

BACKGROUND

Studies investigating the relation between risk profiles and cardiovascular disease have measured risk at baseline only. We investigated maintenance and changes of risk profiles over time and their potential impact on incident cardiovascular disease.

DESIGN

Population-based cohort study.

METHODS

Risk factors were measured at baseline (1987-1991) among 5574 cardiovascular disease-free adults aged 20-59 years. They were classified into four risk categories according to smoking status, presence of diabetes and widely accepted cut-off values for blood pressure, total cholesterol/HDL-ratio and body mass index. Categories were subdivided (maintenance, deterioration, improvement) based on risk factor levels at six and 11 years of follow-up. Multivariable-adjusted hazard ratios (HRs) and 95% confidence intervals (95% CIs) for cardiovascular disease incidence 5-10 years following the risk-change period were fitted using Cox proportional hazards models.

RESULTS

Only 12% of participants were low risk at baseline, and only 7% maintained it. Participants who maintained a low risk profile over 11 years had seven times lower risk of cardiovascular disease (HR: 0.14, 95% CI: 0.05-0.41) than participants with long-term high risk profile, whereas those low risk at baseline whose profile deteriorated had three times lower risk (HR: 0.36, 95% CI: 0.18-0.71). Our results suggest that, within each baseline risk profile group, compared with a stable profile, improving profiles may be associated with up to two-fold lower HRs, and deteriorating profiles with about two-fold higher HRs.

CONCLUSIONS

Our study, using long-term risk profiles, demonstrates the full benefits of low risk profile. These findings underscore the importance of achieving and maintaining low risk from young adulthood onwards.

A review of published analyses of case-cohort studies and recommendations for future reporting

The case-cohort study design combines the advantages of a cohort study with the efficiency of a nested case-control study. However, unlike more standard observational study designs, there are currently no guidelines for reporting results from case-cohort studies. Our aim was to review recent practice in reporting these studies, and develop recommendations for the future. By searching papers published in 24 major medical and epidemiological journals between January 2010 and March 2013 using PubMed, Scopus and Web of Knowledge, we identified 32 papers reporting case-cohort studies. The median subcohort sampling fraction was 4.1% (interquartile range 3.7% to 9.1%). The papers varied in their approaches to describing the numbers of individuals in the original cohort and the subcohort, presenting descriptive data, and in the level of detail provided about the statistical methods used, so it was not always possible to be sure that appropriate analyses had been conducted. Based on the findings of our review, we make recommendations about reporting of the study design, subcohort definition, numbers of participants, descriptive information and statistical methods, which could be used alongside existing STROBE guidelines for reporting observational studies.

Are missing data adequately handled in cluster randomised trials? A systematic review and guidelines

BACKGROUND

Missing data are a potential source of bias, and their handling in the statistical analysis can have an important impact on both the likelihood and degree of such bias. Inadequate handling of the missing data may also result in invalid variance estimation. The handling of missing values is more complex in cluster randomised trials, but there are no reviews of practice in this field.

OBJECTIVES

A systematic review of published trials was conducted to examine how missing data are reported and handled in cluster randomised trials.

METHODS

We systematically identified cluster randomised trials, published in English in 2011, using the National Library of Medicine (MEDLINE) via PubMed. Non-randomised and pilot/feasibility trials were excluded, as were reports of secondary analyses, interim analyses, and economic evaluations and those where no data were at the individual level. We extracted information on missing data and the statistical methods used to deal with them from a random sample of the identified studies.

RESULTS

We included 132 trials. There was evidence of missing data in 95 (72%). Only 32 trials reported handling missing data, 22 of them using a variety of single imputation techniques, 8 using multiple imputation without accommodating the clustering and 2 stating that their likelihood-based complete case analysis accounted for missing values because the data were assumed Missing-at-Random.

LIMITATIONS

The results presented in this study are based on a large random sample of cluster randomised trials published in 2011, identified in electronic searches and therefore possibly missing some trials, most likely of poorer quality. Also, our results are based on information in the main publication for each trial. These reports may omit some important information on the presence of, and reasons for, missing data and on the statistical methods used to handle them. Our extraction methods, based on published reports, could not distinguish between missing data in outcomes and missing data in covariates. This distinction may be important in determining the assumptions about the missing data mechanism necessary for complete case analyses to be valid.

CONCLUSIONS

Missing data are present in the majority of cluster randomised trials. However, they are poorly reported, and most authors give little consideration to the assumptions under which their analysis will be valid. The majority of the methods currently used are valid under very strong assumptions about the missing data, whose plausibility is rarely discussed in the corresponding reports. This may have important consequences for the validity of inferences in some trials. Methods which result in valid inferences under general Missing-at-Random assumptions are available and should be made more accessible.

On shrinkage and model extrapolation in the evaluation of clinical center performance

We consider statistical methods for benchmarking clinical centers based on a dichotomous outcome indicator. Borrowing ideas from the causal inference literature, we aim to reveal how the entire study population would have fared under the current care level of each center. To this end, we evaluate direct standardization based on fixed versus random center effects outcome models that incorporate patient-specific baseline covariates to adjust for differential case-mix. We explore fixed effects (FE) regression with Firth correction and normal mixed effects (ME) regression to maintain convergence in the presence of very small centers. Moreover, we study doubly robust FE regression to avoid outcome model extrapolation. Simulation studies show that shrinkage following standard ME modeling can result in substantial power loss relative to the considered alternatives, especially for small centers. Results are consistent with findings in the analysis of 30-day mortality risk following acute stroke across 90 centers in the Swedish Stroke Register.

Prospective population-based study of the association between serum 25-hydroxyvitamin-D levels and the incidence of specific types of cancer

BACKGROUND

Observational studies have suggested an inverse association between vitamin D status and cancer. We investigated the prospective associations between vitamin D status and the total and specific type of cancer in three cohorts from the general Danish population.

METHODS

A total of 12,204 individuals 18 to 71 years old were included. The level of 25-hydroxyvitamin D was measured at baseline, and information about cancer was obtained from the Danish Cancer Registry.

RESULTS

During the 11.3-year median follow-up time, there were 1,248 incident cancers. HRs [95% confidence intervals (CI)] per 10 nmol/L higher baseline vitamin D level were: for all cancers (HR = 1.02; 95% CI, 0.99-1.04), all cancers excluding non-melanoma skin cancer, NMSC (HR = 1.00; 95% CI, 0.97-1.03), head and neck cancer (HR = 0.97; 95% CI, 0.84-1.12), colorectal cancer (HR = 0.95; 95% CI, 0.88-1.02), cancer of bronchus and lung (HR = 0.98; 95% CI, 0.91-1.05), breast cancer (HR = 1.02; 95% CI, 0.96-1.09), cancer of the uterus (HR = 1.10; 95% CI, 0.95-1.27), prostate cancer (HR = 1.00; 95% CI, 0.93-1.08), cancer of the urinary organs (HR = 1.01; 95% CI, 0.90-1.14), NMSC (HR = 1.06; 95% CI, 1.02-1.10), and malignant melanoma (HR = 1.06; 95% CI, 0.95-1.17).

CONCLUSIONS

Apart from a significantly higher risk for NMSC with higher vitamin D status, we found no statistically significant associations between vitamin D status and total or specific cancers.

IMPACT

Our results do not indicate that there is an impact of vitamin D on total cancer incidence.

The association between admission systolic blood pressure and mortality in significant traumatic brain injury: a multi-centre cohort study

INTRODUCTION

Low systolic blood pressure (SBP) is an important secondary insult following traumatic brain injury (TBI), but its exact relationship with outcome is not well characterised. Although a SBP of <90 mmHg represents the threshold for hypotension in consensus TBI treatment guidelines, recent studies suggest redefining hypotension at higher levels. This study therefore aimed to fully characterise the association between admission SBP and mortality to further inform resuscitation endpoints.

METHODS

We conducted a multicentre cohort study using data from the largest European trauma registry. Consecutive adult patients with AIS head scores >2 admitted directly to specialist neuroscience centres between 2005 and July 2012 were studied. Multilevel logistic regression models were developed to examine the association between admission SBP and 30 day inpatient mortality. Models were adjusted for confounders including age, severity of injury, and to account for differential quality of hospital care.

RESULTS

5057 patients were included in complete case analyses. Admission SBP demonstrated a smooth u-shaped association with outcome in a bivariate analysis, with increasing mortality at both lower and higher values, and no evidence of any threshold effect. Adjusting for confounding slightly attenuated the association between mortality and SBP at levels <120 mmHg, and abolished the relationship for higher SBP values. Case-mix adjusted odds of death were 1.5 times greater at <120 mmHg, doubled at <100 mmHg, tripled at <90 mmHg, and six times greater at SBP<70 mmHg, p<0.01.

CONCLUSIONS

These findings indicate that TBI studies should model SBP as a continuous variable and may suggest that current TBI treatment guidelines, using a cut-off for hypotension at SBP<90 mmHg, should be reconsidered.

Multiple imputation of covariates by fully conditional specification: Accommodating the substantive model

Missing covariate data commonly occur in epidemiological and clinical research, and are often dealt with using multiple imputation. Imputation of partially observed covariates is complicated if the substantive model is non-linear (e.g. Cox proportional hazards model), or contains non-linear (e.g. squared) or interaction terms, and standard software implementations of multiple imputation may impute covariates from models that are incompatible with such substantive models. We show how imputation by fully conditional specification, a popular approach for performing multiple imputation, can be modified so that covariates are imputed from models which are compatible with the substantive model. We investigate through simulation the performance of this proposal, and compare it with existing approaches. Simulation results suggest our proposal gives consistent estimates for a range of common substantive models, including models which contain non-linear covariate effects or interactions, provided data are missing at random and the assumed imputation models are correctly specified and mutually compatible. Stata software implementing the approach is freely available.

A toolkit for measurement error correction, with a focus on nutritional epidemiology

Exposure measurement error is a problem in many epidemiological studies, including those using biomarkers and measures of dietary intake. Measurement error typically results in biased estimates of exposure-disease associations, the severity and nature of the bias depending on the form of the error. To correct for the effects of measurement error, information additional to the main study data is required. Ideally, this is a validation sample in which the true exposure is observed. However, in many situations, it is not feasible to observe the true exposure, but there may be available one or more repeated exposure measurements, for example, blood pressure or dietary intake recorded at two time points. The aim of this paper is to provide a toolkit for measurement error correction using repeated measurements. We bring together methods covering classical measurement error and several departures from classical error: systematic, heteroscedastic and differential error. The correction methods considered are regression calibration, which is already widely used in the classical error setting, and moment reconstruction and multiple imputation, which are newer approaches with the ability to handle differential error. We emphasize practical application of the methods in nutritional epidemiology and other fields. We primarily consider continuous exposures in the exposure-outcome model, but we also outline methods for use when continuous exposures are categorized. The methods are illustrated using the data from a study of the association between fibre intake and colorectal cancer, where fibre intake is measured using a diet diary and repeated measures are available for a subset.

Introduction to multiple imputation for dealing with missing data

Missing data are common in both observational and experimental studies. Multiple imputation (MI) is a two-stage approach where missing values are imputed a number of times using a statistical model based on the available data and then inference is combined across the completed datasets. This approach is becoming increasingly popular for handling missing data. In this paper, we introduce the method of MI, as well as a discussion surrounding when MI can be a useful method for handling missing data and the drawbacks of this approach. We illustrate MI when exploring the association between current asthma status and forced expiratory volume in 1 s after adjustment for potential confounders using data from a population-based longitudinal cohort study.

Nonlinear multiple imputation for continuous covariate within semiparametric Cox model: application to HIV data in Senegal

Multiple imputation is commonly used to impute missing covariate in Cox semiparametric regression setting. It is to fill each missing data with more plausible values, via a Gibbs sampling procedure, specifying an imputation model for each missing variable. This imputation method is implemented in several softwares that offer imputation models steered by the shape of the variable to be imputed, but all these imputation models make an assumption of linearity on covariates effect. However, this assumption is not often verified in practice as the covariates can have a nonlinear effect. Such a linear assumption can lead to a misleading conclusion because imputation model should be constructed to reflect the true distributional relationship between the missing values and the observed values. To estimate nonlinear effects of continuous time invariant covariates in imputation model, we propose a method based on B-splines function. To assess the performance of this method, we conducted a simulation study, where we compared the multiple imputation method using Bayesian splines imputation model with multiple imputation using Bayesian linear imputation model in survival analysis setting. We evaluated the proposed method on the motivated data set collected in HIV-infected patients enrolled in an observational cohort study in Senegal, which contains several incomplete variables. We found that our method performs well to estimate hazard ratio compared with the linear imputation methods, when data are missing completely at random, or missing at random.

Multiple Imputation For Combined-Survey Estimation With Incomplete Regressors In One But Not Both Surveys

Within-survey multiple imputation (MI) methods are adapted to pooled-survey regression estimation where one survey has more regressors, but typically fewer observations, than the other. This adaptation is achieved through: (1) larger numbers of imputations to compensate for the higher fraction of missing values; (2) model-fit statistics to check the assumption that the two surveys sample from a common universe; and (3) specificying the analysis model completely from variables present in the survey with the larger set of regressors, thereby excluding variables never jointly observed. In contrast to the typical within-survey MI context, cross-survey missingness is monotonic and easily satisfies the Missing At Random (MAR) assumption needed for unbiased MI. Large efficiency gains and substantial reduction in omitted variable bias are demonstrated in an application to sociodemographic differences in the risk of child obesity estimated from two nationally-representative cohort surveys.

Purposeful variable selection and stratification to impute missing Focused Assessment with Sonography for Trauma data in trauma research

BACKGROUND

The Focused Assessment with Sonography for Trauma (FAST) examination is an important variable in many retrospective trauma studies. The purpose of this study was to devise an imputation method to overcome missing data for the FAST examination. Owing to variability in patients' injuries and trauma care, these data are unlikely to be missing completely at random, raising concern for validity when analyses exclude patients with missing values.

METHODS

Imputation was conducted under a less restrictive, more plausible missing-at-random assumption. Patients with missing FAST examinations had available data on alternate, clinically relevant elements that were strongly associated with FAST results in complete cases, especially when considered jointly. Subjects with missing data (32.7%) were divided into eight mutually exclusive groups based on selected variables that both described the injury and were associated with missing FAST values. Additional variables were selected within each group to classify missing FAST values as positive or negative, and correct FAST examination classification based on these variables was determined for patients with nonmissing FAST values.

RESULTS

Severe head/neck injury (odds ratio [OR], 2.04), severe extremity injury (OR, 4.03), severe abdominal injury (OR, 1.94), no injury (OR, 1.94), other abdominal injury (OR, 0.47), other head/neck injury (OR, 0.57), and other extremity injury (OR, 0.45) groups had significant ORs for missing data; the other group's OR was not significant (OR, 0.84). All 407 missing FAST values were imputed, with 109 classified as positive. Correct classification of nonmissing FAST results using the alternate variables was 87.2%.

CONCLUSION

Purposeful imputation for missing FAST examinations based on interactions among selected variables assessed by simple stratification may be a useful adjunct to sensitivity analysis in the evaluation of imputation strategies under different missing data mechanisms. This approach has the potential for widespread application in clinical and translational research, and validation is warranted.

The impact of missing data on analyses of a time-dependent exposure in a longitudinal cohort: a simulation study

BACKGROUND

Missing data often cause problems in longitudinal cohort studies with repeated follow-up waves. Research in this area has focussed on analyses with missing data in repeated measures of the outcome, from which participants with missing exposure data are typically excluded. We performed a simulation study to compare complete-case analysis with Multiple imputation (MI) for dealing with missing data in an analysis of the association of waist circumference, measured at two waves, and the risk of colorectal cancer (a completely observed outcome).

METHODS

We generated 1,000 datasets of 41,476 individuals with values of waist circumference at waves 1 and 2 and times to the events of colorectal cancer and death to resemble the distributions of the data from the Melbourne Collaborative Cohort Study. Three proportions of missing data (15, 30 and 50%) were imposed on waist circumference at wave 2 using three missing data mechanisms: Missing Completely at Random (MCAR), and a realistic and a more extreme covariate-dependent Missing at Random (MAR) scenarios. We assessed the impact of missing data on two epidemiological analyses: 1) the association between change in waist circumference between waves 1 and 2 and the risk of colorectal cancer, adjusted for waist circumference at wave 1; and 2) the association between waist circumference at wave 2 and the risk of colorectal cancer, not adjusted for waist circumference at wave 1.

RESULTS

We observed very little bias for complete-case analysis or MI under all missing data scenarios, and the resulting coverage of interval estimates was near the nominal 95% level. MI showed gains in precision when waist circumference was included as a strong auxiliary variable in the imputation model.

CONCLUSIONS

This simulation study, based on data from a longitudinal cohort study, demonstrates that there is little gain in performing MI compared to a complete-case analysis in the presence of up to 50% missing data for the exposure of interest when the data are MCAR, or missing dependent on covariates. MI will result in some gain in precision if a strong auxiliary variable that is not in the analysis model is included in the imputation model.

Seven deadly sins in trauma outcomes research: an epidemiologic post mortem for major causes of bias

BACKGROUND

Because randomized clinical trials in trauma outcomes research are expensive and complex, they have rarely been the basis for the clinical care of trauma patients. Most published findings are derived from retrospective and occasionally prospective observational studies that may be particularly susceptible to bias. The sources of bias include some common to other clinical domains, such as heterogeneous patient populations with competing and interdependent short- and long-term outcomes. Other sources of bias are unique to trauma, such as rapidly changing multisystem responses to injury that necessitate highly dynamic treatment regimens such as blood product transfusion. The standard research design and analysis strategies applied in published observational studies are often inadequate to address these biases.

METHODS

Drawing on recent experience in the design, data collection, monitoring, and analysis of the 10-site observational PRospective Observational Multicenter Major Trauma Transfusion (PROMMTT) study, 7 common and sometimes overlapping biases are described through examples and resolution strategies.

RESULTS

Sources of bias in trauma research include ignoring (1) variation in patients' indications for treatment (indication bias), (2) the dependency of intervention delivery on patient survival (survival bias), (3) time-varying treatment, (4) time-dependent confounding, (5) nonuniform intervention effects over time, (6) nonrandom missing data mechanisms, and (7) imperfectly defined variables. This list is not exhaustive.

CONCLUSION

The mitigation strategies to overcome these threats to validity require epidemiologic and statistical vigilance. Minimizing the highlighted types of bias in trauma research will facilitate clinical translation of more accurate and reproducible findings and improve the evidence-base that clinicians apply in their care of injured patients.

Family physical activity and meal practices associated with disordered weight control behaviors in a multiethnic sample of middle-school youth

OBJECTIVE

Family practices around weight-related behaviors can shape children's development of disordered weight control behaviors (DWCB), such as vomiting, taking laxatives, or taking diet pills without a prescription. This study examined family meal and physical activity (PA) practices associated with DWCB among a multiethnic sample of youth.

METHODS

We assessed self-report data on frequency of family sit-down dinners, types of parental involvement in their children's PA, and DWCB are from 15,461 6th to 8th grade girls and boys in 47 middle schools participating in the Massachusetts Healthy Choices Study at baseline (2005).

RESULTS

Youth who had family sit-down dinners every day had lower odds of DWCB (girls: odds ratio [OR] 0.3; 95% confidence interval [CI] 0.2-0.5; boys: OR 0.6; 95% CI 0.4-0.9) than youth who never had family sit-down dinners. Similar effect estimates were found for youth who had family sit-down dinners most days. Parental provision of rides to and from a PA event was also found to be protective against DWCB among girls (OR 0.7; 95% CI 0.5-0.9). In contrast, parental participation in PA with their children was associated with increased risk for DWCB (girls: OR 1.4; 95% CI 1.0-1.8; boys: OR 1.9; 95% CI 1.4-2.4). These associations did not differ by race/ethnicity or weight status.

CONCLUSIONS

Programs emphasizing the importance of family meals may be beneficial in preventing DWCB in youth of all ethnicities. Further research is needed on how various methods of parental involvement in their children's PA are associated with DWCB.

Combining multiple imputation and meta-analysis with individual participant data

Multiple imputation is a strategy for the analysis of incomplete data such that the impact of the missingness on the power and bias of estimates is mitigated. When data from multiple studies are collated, we can propose both within-study and multilevel imputation models to impute missing data on covariates. It is not clear how to choose between imputation models or how to combine imputation and inverse-variance weighted meta-analysis methods. This is especially important as often different studies measure data on different variables, meaning that we may need to impute data on a variable which is systematically missing in a particular study. In this paper, we consider a simulation analysis of sporadically missing data in a single covariate with a linear analysis model and discuss how the results would be applicable to the case of systematically missing data. We find in this context that ensuring the congeniality of the imputation and analysis models is important to give correct standard errors and confidence intervals. For example, if the analysis model allows between-study heterogeneity of a parameter, then we should incorporate this heterogeneity into the imputation model to maintain the congeniality of the two models. In an inverse-variance weighted meta-analysis, we should impute missing data and apply Rubin's rules at the study level prior to meta-analysis, rather than meta-analyzing each of the multiple imputations and then combining the meta-analysis estimates using Rubin's rules. We illustrate the results using data from the Emerging Risk Factors Collaboration.

MCAR is not necessary for the complete cases to constitute a simple random subsample of the target sample

Missing data is the norm rather than the exception in complex epidemiological studies. Complete-case analyses, which discard all subjects with some data values missing, are known to be valid under the very restrictive assumption that the response mechanism is missing completely at random (MCAR). While conditions weaker than MCAR are known under which estimators of regression coefficients are unbiased, one often comes across the view in the literature that MCAR is necessary for the complete cases to form a simple random subsample of the target sample. In this paper, we explain why this is not the case, and we distill an assumption weaker than MCAR under which the simple random subsample condition holds, which we call available at random (AAR). Moreover, we show that, unlike MCAR, AAR response mechanisms can be missing not at random (MNAR). We also suggest how approximate AAR mechanisms might arise in practice through cancellation of selection and drop-out effects, and we conclude that before pooling partially complete and complete cases into an analysis, the investigator should consider how selection might impact on the representativeness of the cases included in the pooled analysis (compared to those comprising the complete cases only).

Using full-cohort data in nested case-control and case-cohort studies by multiple imputation

In many large prospective cohorts, expensive exposure measurements cannot be obtained for all individuals. Exposure-disease association studies are therefore often based on nested case-control or case-cohort studies in which complete information is obtained only for sampled individuals. However, in the full cohort, there may be a large amount of information on cheaply available covariates and possibly a surrogate of the main exposure(s), which typically goes unused. We view the nested case-control or case-cohort study plus the remainder of the cohort as a full-cohort study with missing data. Hence, we propose using multiple imputation (MI) to utilise information in the full cohort when data from the sub-studies are analysed. We use the fully observed data to fit the imputation models. We consider using approximate imputation models and also using rejection sampling to draw imputed values from the true distribution of the missing values given the observed data. Simulation studies show that using MI to utilise full-cohort information in the analysis of nested case-control and case-cohort studies can result in important gains in efficiency, particularly when a surrogate of the main exposure is available in the full cohort. In simulations, this method outperforms counter-matching in nested case-control studies and a weighted analysis for case-cohort studies, both of which use some full-cohort information. Approximate imputation models perform well except when there are interactions or non-linear terms in the outcome model, where imputation using rejection sampling works well.

The contribution of school environmental factors to individual and school variation in disordered weight control behaviors in a statewide sample of middle schools

We investigated the contribution of school environmental factors to individual and school variation in disordered weight control behaviors (DWCB). Analyses were based on self-report data gathered from 18,567 middle-school students in 2005 and publicly available data on school characteristics. We observed large differences across schools in percent of students engaging in DWCB in the past month, ranging from less than 1% of the student body to 12%. School-neighborhood poverty was associated with higher odds of DWCB in boys. Preventive strategies need to account for wide variability across schools and environmental factors that may contribute to DWCB in early adolescence.

Loss to follow-up in cohort studies: bias in estimates of socioeconomic inequalities

BACKGROUND

Although cohort members tend to be healthy and affluent compared with the whole population, some studies indicate this does not bias certain exposure-outcome associations. It is less clear whether this holds when socioeconomic position (SEP) is the exposure of interest.

METHODS

As an illustrative example, we use data from the Avon Longitudinal Study of Parents and Children. We calculate estimates of maternal education inequalities in outcomes for which data are available on almost the whole cohort (birth weight and length, breastfeeding, preterm birth, maternal obesity, smoking during pregnancy, educational attainment). These are calculated for the full cohort (n~12,000) and in restricted subsamples defined by continued participation at age 10 years (n∼7,000) and age 15 years (n∼5,000).

RESULTS

Loss to follow-up was related both to SEP and outcomes. For each outcome, loss to follow-up was associated with underestimation of inequality, which increased as participation rates decreased (eg, mean birth-weight difference between highest and lowest SEP was 116 g [95% confidence interval = 78 to 153] in the full sample and 93 g [45 to 141] and 62 g [5 to 119] in those attending at ages 10 and 15 years, respectively).

CONCLUSIONS

Considerable attrition from cohort studies may result in biased estimates of socioeconomic inequalities, and the degree of bias may worsen as participation rates decrease. However, even with considerable attrition (>50%), qualitative conclusions about the direction and approximate magnitude of inequalities did not change among most of our examples. The appropriate analysis approaches to alleviate bias depend on the missingness mechanism.

Dietary and physical activity factors related to eating disorder symptoms among middle school youth

BACKGROUND

Dietary and physical activity (PA) behaviors can predict disordered weight control behaviors (DWCB) among youth. This study examines dietary and PA correlates of DWCB and differences by race/ethnicity and weight status in a diverse sample of youth.

METHODS

Self-reported data on dietary weight management behaviors, strengthening/toning exercises, moderate-to-vigorous physical activity, and DWCB (vomiting, taking laxatives, and/or taking diet pills without a prescription) were obtained from 15,260 sixth to eighth graders in 47 middle schools participating in the Massachusetts Healthy Choices Study at baseline (2005). Generalized estimating equations were used to estimate odds of DWCB associated with dietary and PA behaviors and to examine for differences by race/ethnicity and weight status, adjusting for covariates and clustering of individuals within schools.

RESULTS

Disordered weight control behaviors were reported by 3.6% of girls and 3.1% of boys. Youth who engaged in strengthening/toning exercises 7 days per week versus 0-3 days per week had increased odds of DWCB (girls odds ratio [OR] = 1.9; 95% confidence interval [CI] = 1.3 - 3.0; boys OR = 1.5; 95% CI = 1.0 - 2.2). Dietary weight management behaviors were associated with increased odds of DWCB (girls OR = 1.2; 95% CI = 1.1 - 1.3; boys OR = 1.3; 95% CI = 1.2 - 1.4) for each additional behavior. These associations did not differ by race/ethnicity or weight status.

CONCLUSIONS

Persons promoting healthy dietary and PA behaviors among youth should consider the co-occurrence of strengthening/toning and dietary weight management behaviors with DWCB and the consistency in these associations across racial/ethnic and weight status groups.