Search

Epidemiologic studies routinely suffer from bias due to exposure measurement error. In this paper, the authors examine the effect of measurement error when the exposure variable of interest is constrained by a lower limit. This is an important consideration, since often in epidemiologic studies an exposure variable is constrained by a lower limit such as zero or a nonzero detection limit. In this paper, attenuation of exposure-disease associations is defined within the framework of a classical model of uncorrelated additive error. Then, the special case of nonlinearity due to the effect of a lower threshold is examined. A general model is developed to characterize the effect of random measurement error when there is a lower threshold for recorded values. Findings are illustrated under the assumption that the true exposure follows the lognormal and gamma distributions. The authors show that the direction and magnitude of bias in estimated exposure-response associations depends on the population distribution of the exposure, the magnitude of the recording threshold, the value assigned to below-threshold measurement results, and the variance in the measured exposure due to random measurement error. [ABSTRACT FROM PUBLISHER]

Measurement error is pervasive in epidemiologic research. Epidemiologists often assume that mismeasurement of study variables is nondifferential with respect to other analytical variables and then rely on the heuristic that "nondifferential misclassification will bias estimates towards the null." However, there are many exceptions to the heuristic for which bias towards the null cannot be assumed. In this paper, we compile and characterize 7 exceptions to this rule and encourage analysts to take a more critical and nuanced approach to evaluating and discussing bias from nondifferential mismeasurement. [ABSTRACT FROM AUTHOR]

Principled methods with which to appropriately analyze missing data have long existed; however, broad implementation of these methods remains challenging. In this and 2 companion papers (Am J Epidemiol. 2018;187(3):576-584 and Am J Epidemiol. 2018;187(3):585-591), we discuss issues pertaining to missing data in the epidemiologic literature. We provide details regarding missing-data mechanisms and nomenclature and encourage the conduct of principled analyses through a detailed comparison of multiple imputation and inverse probability weighting. Data from the Collaborative Perinatal Project, a multisite US study conducted from 1959 to 1974, are used to create a masked data-analytical challenge with missing data induced by known mechanisms. We illustrate the deleterious effects of missing data with naive methods and show how principled methods can sometimes mitigate such effects. For example, when data were missing at random, naive methods showed a spurious protective effect of smoking on the risk of spontaneous abortion (odds ratio (OR) = 0.43, 95% confidence interval (CI): 0.19, 0.93),while implementation of principled methods multiple imputation (OR = 1.30, 95% CI: 0.95, 1.77) or augmented inverse probability weighting (OR = 1.40, 95% CI: 1.00, 1.97) provided estimates closer to the "true" full-data effect (OR = 1.31, 95% CI: 1.05, 1.64). We call for greater acknowledgement of and attention to missing data and for the broad use of principled missing-data methods in epidemiologic research. [ABSTRACT FROM AUTHOR]

In a 1993 paper (Am J Epidemiol. 1993;137(1):1–8), Weinberg considered whether a variable that is associated with the outcome and is affected by exposure but is not an intermediate variable between exposure and outcome should be considered a confounder in etiologic studies. As an example, she examined the common practice of adjusting for history of spontaneous abortion when estimating the effect of an exposure on the risk of spontaneous abortion. She showed algebraically that such an adjustment could substantially bias the results even though history of spontaneous abortion would meet some definitions of a confounder. Directed acyclic graphs (DAGs) were introduced into epidemiology several years later as a tool with which to identify confounders. The authors now revisit Weinberg's paper using DAGs to represent scenarios that arise from her original assumptions. DAG theory is consistent with Weinberg's finding that adjusting for history of spontaneous abortion introduces bias in her original scenario. In the authors' examples, treating history of spontaneous abortion as a confounder introduces bias if it is a descendant of the exposure and is associated with the outcome conditional on exposure or is a child of a collider on a relevant undirected path. Thoughtful DAG analyses require clear research questions but are easily modified for examining different causal assumptions that may affect confounder assessment. [ABSTRACT FROM PUBLISHER]

Statistical interactions between genes and environmental exposures with respect to disease outcomes may help to identify biologic mechanisms and pathways and inform behavioral interventions. The number of persons required for a single study to have sufficient statistical power to detect such interactions may be considered prohibitively large, making a meta-analysis of published literature an apparently attractive alternative. However, meta-analysis of gene-environment interactions using published literature is challenging, with the conclusions being likely to suffer from bias and lack of generalizability. The authors highlight these challenges and biases using an illustrative example: meta-analysis of interactions between the Pro12Ala variant of the peroxisome proliferator-activated receptor γ (PPARγ) gene and various diet and lifestyle factors in the risk of diabetes. The authors conclude that literature-based meta-analysis conducted to examine gene-environment interactions is unlikely to provide a meaningful quantitative conclusion. Alternative strategies are required, including analyses in scientific consortia established to assess main genetic effects, where individual participant data can be shared, allowing both greater power and consistency of analysis methods. However, these consortia are likely to be limited by lack of standardization of the measures of environmental factors. This issue may ultimately only be resolvable by the de novo establishment of large single or multicenter cohorts using comparable methods. [ABSTRACT FROM PUBLISHER]

Recently, some authors have questioned the validity of methods which correct relative risk estimates for measurement error and misclassification when the “gold standard” used to obtain information about the measurement error process is itself imperfect. When such an “alloyed” gold standard is used to validate the usual exposure measurement, the bias in the “regression calibration” (Rosner et al., Stat Med 1989; 8: 1051–69) measurement-error correction factor for relative risks estimated from logistic regression models is derived. This quantity is a function of the correlations of the “alloyed” gold standard (X) and the usual exposure assessment method (Z) with the truth, of the ratio of the variances of X and Z, and of the correlation between the errors in the “alloyed” gold standard and the errors in the usual exposure assessment method. In this paper, it is proven that if the errors between Z and X are uncorrelated, the regression calibration method has no bias even when the gold standard is “alloyed.” When a third method of exposure assessment is available and it is reasonable to assume that the errors in this method are uncorrelated with the errors in the other two exposure assessment methods, point and interval estimates of the correlation between the errors in X and Z are derived. These methods are illustrated here with data on the measurement of physical activity, vitamins A and E, and poly- and monounsaturated fat. In addition, when a third exposure assessment method is available, a modification of standard regression calibration is derived which can be used to calculate point and interval estimates of relative risk that are corrected for measurement error in both X and Z. This new method is illustrated here with data from the Health Professionals Follow-up Study, a study investigating the associations between physical activity and colon cancer incidence and between vitamin E intake and coronary heart disease. It is shown that in these examples, correlations of the errors in X and Z tended to be small. Even when moderate, estimates of relative risk corrected for error in both X and Z were not very different from the estimates which assumed that X was a true gold standard. Am J Epidemiol 1997; 145: 184–96. [ABSTRACT FROM AUTHOR]

Cohort studies are often enriched for a primary exposure of interest to improve cost-effectiveness, which presents analytical challenges not commonly discussed in epidemiology. In this paper, we use causal diagrams to represent exposure-enriched cohort studies, illustrate a scenario wherein the risk ratio for the effect of a secondary exposure on an outcome is biased, and propose an analytical method for correcting for such bias. In our motivating example, maternal smoking (Z) is a cause of fetal growth restriction (X), which subsequently affects preterm birth (Y) (i.e., Z → X → Y); strong positive associations exist between both Z, X and X, Y; and enrichment for X increases its prevalence from 10% to 50%. In the X-enriched cohort, unadjusted and X-adjusted analyses lead to bias in the risk ratio for the total effect of Z on Y. After application of inverse probability weights, the bias is corrected, with a small loss of efficiency in comparison with a same-sized study without X-enrichment. With increasing interest in conducting secondary analyses to reduce research costs, caution should be employed when analyzing studies that have already been enriched, intentionally or unintentionally, for a primary exposure of interest. Causal diagrams can help identify scenarios in which secondary analyses may be biased. Inverse probability weights can be used to remove the bias. [ABSTRACT FROM AUTHOR]

Consider a study in which the effect of a binary exposure on an outcome operates partly through a binary mediator but measurement of the mediator is nondifferentially misclassified. Suppose that an investigator wishes to estimate the direct and indirect effects of the exposure on the outcome. In this paper, the authors describe a mathematical correspondence between the empirical expressions for the natural direct effect and the effect of exposure among the unexposed standardized by a binary confounder. They then exploit this correspondence to prove that the direction of the bias due to nondifferential measurement error in estimating the natural direct and indirect effects is to overestimate the natural direct effect and underestimate the natural indirect effect. [ABSTRACT FROM AUTHOR]

Dialysis has been used in the treatment of patients with end-stage renal disease since the 1960s. Recently, several large observational studies have been conducted to assess whether early initiation of dialysis prolongs survival, as compared with late initiation. However, these studies have used analytic approaches which are likely to suffer from either lead-time bias or immortal-time bias. In this paper, the authors demonstrate that recently developed methods in the causal inference literature can be used to avoid both types of bias and accurately estimate the ideal time for dialysis initiation from observational data. This is illustrated using data from a nationwide population-based cohort of patients with chronic kidney disease in Sweden (1996–2003). [ABSTRACT FROM PUBLISHER]

Selection bias is a concern in cohort studies in which selection into the cohort is related to the studied outcome. An example is chronic infection with hepatitis C virus, where the initial infection may be asymptomatic for decades. This problem leads to selection of more severely ill individuals into registers of such infections. Cohort studies often adjust for this bias by introducing a time window between entry into the cohort and entry into the study. This paper describes and assesses a novel method to improve adjustment for this type of selection bias. The size of the time window is decided by calculating a standardized incidence ratio as a continuous function of the size of the time window. The resulting graph is used to decide on an appropriate window size. The method is evaluated by using the Swedish register of hepatitis C virus infections for 1990–2006. The complications studied were non-Hodgkin lymphoma and liver cancer. Selection bias differed for the studied outcomes, and a time window of a minimum of 2 months and 12 months, respectively, was judged to be appropriate. The novel method may have advantages compared with an interval-based method, especially in cohort studies with small numbers of events. [ABSTRACT FROM PUBLISHER]

Immortal time is a span of cohort follow-up during which, because of exposure definition, the outcome under study could not occur. Bias from immortal time was first identified in the 1970s in epidemiology in the context of cohort studies of the survival benefit of heart transplantation. It recently resurfaced in pharmacoepidemiology, with several observational studies reporting that various medications can be extremely effective at reducing morbidity and mortality. These studies, while using different cohort designs, all involved some form of immortal time and the corresponding bias. In this paper, the author describes various cohort study designs leading to this bias, quantifies its magnitude under different survival distributions, and illustrates it by using data from a cohort of lung cancer patients. The author shows that for time-based, event-based, and exposure-based cohort definitions, the bias in the rate ratio resulting from misclassified or excluded immortal time increases proportionately to the duration of immortal time. The bias is more pronounced with a decreasing hazard function for the outcome event, as illustrated with the Weibull distribution compared with a constant hazard from the exponential distribution. In conclusion, observational studies of drug benefit in which computerized databases are used must be designed and analyzed properly to avoid immortal time bias. [ABSTRACT FROM PUBLISHER]

The authors propose a new statistical procedure that utilizes measurement error models to estimate missing exposure data in health effects assessment. The method detailed in this paper follows a Bayesian framework that allows estimation of various parameters of the model in the presence of missing covariates in an informative way. The authors apply this methodology to study the effect of household-level long-term air pollution exposures on lung function for subjects from the Southern California Children's Health Study pilot project, conducted in the year 2000. Specifically, they propose techniques to examine the long-term effects of nitrogen dioxide (NO2) exposure on children's lung function for persons living in 11 southern California communities. The effect of nitrogen dioxide exposure on various measures of lung function was examined, but, similar to many air pollution studies, no completely accurate measure of household-level long-term nitrogen dioxide exposure was available. Rather, community-level nitrogen dioxide was measured continuously over many years, but household-level nitrogen dioxide exposure was measured only during two 2-week periods, one period in the summer and one period in the winter. From these incomplete measures, long-term nitrogen dioxide exposure and its effect on health must be inferred. Results show that the method improves estimates when compared with standard frequentist approaches. [ABSTRACT FROM AUTHOR]

Biomarker use in exposure assessment is increasingly common, and consideration of related issues is of growing importance. Exposure quantification may be compromised when measurement is subject to a lower threshold. Statistical modeling of such data requires a decision regarding the handling of such readings. Various authors have considered this problem. In the context of linear regression analysis, Richardson and Ciampi (Am J Epidemiol 2003;157:355–63) proposed replacement of data below a threshold by a constant equal to the expectation for such data to yield unbiased estimates. Use of such an imputation has some limitations; distributional assumptions are required, and bias reduction in estimation of regression parameters is asymptotic, thereby presenting concerns about small studies. In this paper, the authors propose distribution-free methods for managing values below detection limits and evaluate the biases that may result when exposure measurement is constrained by a lower threshold. The authors utilize an analytical approach and a simulation study to assess the effects of the proposed replacement method on estimates. These results may inform decisions regarding analytical plans for future studies and provide a possible explanation for some amount of the discordance seen in extant literature. [ABSTRACT FROM AUTHOR]

Often, data on important confounders are not available in cohort studies. Sensitivity analyses based on the relation of single, but not multiple, unmeasured confounders with an exposure of interest in a separate validation study have been proposed. In this paper, the authors controlled for measured confounding in the main cohort using propensity scores (PS's) and addressed unmeasured confounding by estimating two additional PS's in a validation study. The “error-prone” PS exclusively used information available in the main cohort. The “gold standard” PS additionally included data on covariates available only in the validation study. Based on these two PS's in the validation study, regression calibration was applied to adjust regression coefficients. This propensity score calibration (PSC) adjusts for unmeasured confounding in cohort studies with validation data under certain, usually untestable, assumptions. The authors used PSC to assess the relation between nonsteroidal antiinflammatory drugs (NSAIDs) and 1-year mortality in a large cohort of elderly persons. “Traditional” adjustment resulted in a hazard ratio for NSAID users of 0.80 (95% confidence interval (CI): 0.77, 0.83) as compared with an unadjusted hazard ratio of 0.68 (95% CI: 0.66, 0.71). Application of PSC resulted in a more plausible hazard ratio of 1.06 (95% CI: 1.00, 1.12). Until the validity and limitations of PSC have been assessed in different settings, the method should be seen as a sensitivity analysis. [ABSTRACT FROM AUTHOR]

In research on the determinants of change in health status, a crucial analytic decision is whether to adjust for baseline health status. In this paper, the authors examine the consequences of baseline adjustment, using for illustration the question of the effect of educational attainment on change in cognitive function in old age. With data from the US-based Assets and Health Dynamics Among the Oldest Old survey (n = 5,726; born before 1924), they show that adjustment for baseline cognitive test score substantially inflates regression coefficient estimates for the effect of schooling on change in cognitive test scores compared with models without baseline adjustment. To explain this finding, they consider various plausible assumptions about relations among variables. Each set of assumptions is represented by a causal diagram. The authors apply simple rules for assessing causal diagrams to demonstrate that, in many plausible situations, baseline adjustment induces a spurious statistical association between education and change in cognitive score. More generally, when exposures are associated with baseline health status, this bias can arise if change in health status preceded baseline assessment or if the dependent variable measurement is unreliable or unstable. In some cases, change-score analyses without baseline adjustment provide unbiased causal effect estimates when baseline-adjusted estimates are biased. [ABSTRACT FROM AUTHOR]

The Flint Men's Health Study is an ongoing population-based study of African-American men designed to address questions related to prostate cancer and urologic symptoms. The initial phase of the study was conducted in 1996-1997 in two stages: an interviewer-administered survey followed by a clinical examination. The response rate in the clinical examination phase was 52%. Thus, some data were missing for clinical examination variables, diminishing the generalizability of the results to the general population. This paper is a case study demonstrating the application of multiple imputation to address important questions related to prostate cancer and urologic symptoms in a data set with missing values. On the basis of the observed clinical examination data, the American Urological Association Symptoms Score showed a surprising reduction in symptoms in the oldest age group, but after multiple imputation there was a monotonically increasing trend with age. It appeared that multiple imputation corrected for nonresponse bias associated with the observed data. For other outcome measures--namely, the age-adjusted 95th percentile of prostate-specific antigen level and the association between urologic symptoms and prostate volume--results from the observed data and the multiply imputed data were similar. [ABSTRACT FROM AUTHOR]

The etiology, particularly the genetic basis, of multifactorial late-onset diseases is the subject of many genetic epidemiologic studies. The authors' aim in this paper was to investigate the circumstances under which competing risks can lead to bias in studies of genetic susceptibility to late-onset diseases. The authors used a model built in an epidemiologic framework to show that when a genetic risk factor and an environmental risk factor interact to cause a frequent competing risk of death, the measure of the association between the disease under investigation and the genetic risk factor will be biased if the environmental risk factor is also associated with the latter disease and is omitted from the analysis. This is an example of confounding bias, and it is the consequence of an association between the genetic risk factor and the environmental risk factor that appears over time. Numerical examples show that under certain conditions this bias can be substantial. The authors present several patterns of association in favor of such a bias. Because competing risks of death are likely to be present in older subjects, researchers studying the etiology of late-onset diseases should be aware of the possibility of this bias. [ABSTRACT FROM PUBLISHER]

Regression calibration is a technique that corrects biases in regression results in situations where exposure variables are measured with error. The existence of a calibration substudy, where accurate and crude measurement methods are related by a second regression analysis, is assumed. The cost of measurement error in multivariate analyses is loss of statistical power. In this paper, calibration data from California Seventh-day Adventists are used to simulate study populations and new calibration studies. Applying regression calibration logistic analyses, the authors estimate power for pairs of nutritional variables. The results demonstrate substantial loss of power if variables measured with error are strongly correlated. Biases in estimated effects in cases where regression calibration is not performed can be large and are corrected by regression calibration. When the true coefficient has zero value, the corresponding coefficient in a crude analysis will usually have a nonzero expected value. Then type I error probabilities are not nominal, and the erroneous appearance of statistical significance can readily occur, particularly in large studies. Major determinants of power with use of regression calibration are collinearity between the variables measured with error and the size of correlations between crude and corresponding true variables. Where there is important collinearity, useful gains in power accrue with calibration study size up to 1,000 subjects. [ABSTRACT FROM PUBLISHER]

Methods of adjusting for bias in estimates due to mismeasured or missing covariates and outcomes through the use of validation sets have been developed in many types of health studies. These methods can be employed for the efficient design and analysis of vaccine studies as well. On the one hand, nonspecific case definitions can lead to attenuated efficacy and effectiveness estimates, but confirmation by culture or a quick test of the infectious agent is also expensive and difficult. On the other hand, data on exposure to infection can influence estimates of vaccine efficacy, but good data on exposure are difficult to obtain. In this paper, the authors show how use of small validation sets can correct the bias of the estimates obtained from a large main study while maintaining efficiency. They illustrate the approach for outcomes using the example of influenza vaccine efficacy and effectiveness trials and illustrate the approach for exposure to infection using the example of a human immunodeficiency virus vaccine trial. The authors discuss challenges posed by infectious diseases in the use of currently available methods. Development of these efficient designs and methods of analysis for vaccine field studies will improve estimation of vaccine efficacy for both susceptibility and infectiousness, as well as estimation of indirect and overall effects of vaccination in community trials. [ABSTRACT FROM AUTHOR]

Regression calibration is a popular approach for correcting biases in estimated regression parameters when exposure variables are measured with error. This approach involves building a calibration equation to estimate the value of the unknown true exposure given the error-prone measurement and other covariates. The estimated, or calibrated, exposure is then substituted for the unknown true exposure in the health outcome regression model. When used properly, regression calibration can greatly reduce the bias induced by exposure measurement error. Here, we first provide an overview of the statistical framework for regression calibration, specifically discussing how a special type of error, called Berkson error, arises in the estimated exposure. We then present practical issues to consider when applying regression calibration, including: 1) how to develop the calibration equation and which covariates to include; 2) valid ways to calculate standard errors of estimated regression coefficients; and 3) problems arising if one of the covariates in the calibration model is a mediator of the relationship between the exposure and outcome. Throughout, we provide illustrative examples using data from the Hispanic Community Health Study/Study of Latinos (United States, 2008–2011) and simulations. We conclude with recommendations for how to perform regression calibration. [ABSTRACT FROM AUTHOR]

In this paper, the authors report on the design of a population-based case-control study of family history as a risk factor for coronary heart disease (CHD). They studied the characteristics of subjects who completed a detailed family history questionnaire in 1992–1994 as well as the accuracy of recall of family history in order to quantify both selection and recall biases. Coronary disease cases were enrolled through the Newcastle MONICA Project (Monitoring Trends and Determinants in Cardiovascular Disease), which registered all suspected heart attacks and sudden cardiac deaths in the Lower Hunter region of New South Wales, Australia, between August 1984 and March 1994. Controls were selected at random from the New South Wales electoral roll. The response rate was 76% in cases and 62% in controls; the major factor associated with participation in the study was perceived family history of CHD, more so in the control series than in the case series. Accuracy was determined by comparing information obtained from the proband with that recorded on death certificates. In first-degree relatives, sensitivity of CHD recall was 85% (95% confidence interval (Cl) 74–92%) in cases and 95% (95% Cl 84–99%) in controls, while specificity was 59% (95% Cl 49–69%) and 74% (95% Cl 65–82%), respectively. The net bias in both selection and recall is toward the null and hence the comparisons provide a conservative estimate of risk of CHD associated with a positive family history. Am J Epidemiol 1998; 147:1123–32. [ABSTRACT FROM PUBLISHER]

Conditional logistic regression was developed to avoid “sparse-data” biases that can arise in ordinary logistic regression analysis. Nonetheless, it is a large-sample method that can exhibit considerable bias when certain types of matched sets are infrequent or when the model contains too many parameters. Sparse-data bias can cause misleading inferences about confounding, effect modification, dose response, and induction periods, and can interact with other biases. In this paper, the authors describe these problems in the context of matched case-control analysis and provide examples from a study of electrical wiring and childhood leukemia and a study of diet and glioma. The same problems can arise in any likelihood-based analysis, including ordinary logistic regression. The problems can be detected by careful inspection of data and by examining the sensitivity of estimates to category boundaries, variables in the model, and transformations of those variables. One can also apply various bias corrections or turn to methods less sensitive to sparse data than conditional likelihood, such as Bayesian and empirical-Bayes (hierarchical regression) methods. Am J Epidemiol 2000;151:531–9. [ABSTRACT FROM PUBLISHER]

Many epidemiologic studies use metabolomics for discovery-based research. The degree to which sample handling may influence findings, however, is poorly understood. In 2016, serum samples from 13 volunteers from the US Department of Agriculture's Beltsville Human Nutrition Research Center were subjected to different clotting (30 minutes/120 minutes) and refrigeration (0 minutes/24 hours) conditions, as well as different numbers (0/1/4) and temperatures (ice/refrigerator/room temperature) of thaws. The median absolute percent difference (APD) between metabolite levels and correlations between levels across conditions were estimated for 628 metabolites. The potential for handling artifacts to induce false-positive associations was estimated using variable hypothetical scenarios in which 1%–100% of case samples had different handling than control samples. All handling conditions influenced metabolite levels. Across metabolites, the median APD when extending clotting time was 9.08%. When increasing the number of thaws from 0 to 4, the median APD was 10.05% for ice and 5.54% for room temperature. Metabolite levels were correlated highly across conditions (all r 's ≥ 0.84), indicating that relative ranks were preserved. However, if handling varied even modestly by case status, our hypotheticals showed that results can be biased and can result in false-positive findings. Sample handling affects levels of metabolites, and special care should be taken to minimize effects. Shorter room-temperature thaws should be preferred over longer ice thaws, and handling should be meticulously matched by case status. [ABSTRACT FROM AUTHOR]

Confounding by indication is a term used when a variable is a rise factor for a disease among nonexposed persons and is associated with the exposure of interest in the population form which the cases derive, without being an intermediate step in the causal pathway between the exposure and the disease. However, in the literature, the term confounding by indicationis not always used consistently. The authors found three different situations in which the term has been applied or might have been used but was not: confounding by indication as protopathic bias, as confounding by severity, or as a form of selection bias. It might be helpful to limit use of the term confounding by indication to the situation in which the disease that forms the indication acts as a confounder irrespective of its severity and to applly the term confounding by severity if the severity of this disease acts as a confounder. Protopathic bias and selection bias should not be confused with these terms. The used of appropriate terms ultimately will improve communication among researchers and contribute to the clarity of their papers. Am J Epidemiol 1999; 149: 981–3. [ABSTRACT FROM PUBLISHER]

Mendelian randomization (MR) is used to answer a variety of epidemiologic questions. One stated advantage of MR is that it estimates a "lifetime effect" of exposure, though this term remains vaguely defined. Instrumental variable analysis, on which MR is based, has focused on estimating the effects of point or time-fixed exposures rather than "lifetime effects." Here we use an empirical example with data from the Rotterdam Study (Rotterdam, the Netherlands, 2009–2013) to demonstrate how confusion can arise when estimating "lifetime effects." We provide one possible definition of a lifetime effect: the average change in outcome measured at time t when the entire exposure trajectory from conception to time t is shifted by 1 unit. We show that MR only estimates this type of lifetime effect under specific conditions—for example, when the effect of the genetic variants used on exposure does not change over time. Lastly, we simulate the magnitude of bias that would result in realistic scenarios that use genetic variants with effects that change over time. We recommend that investigators in future MR studies carefully consider the effect of interest and how genetic variants whose effects change with time may impact the interpretability and validity of their results. [ABSTRACT FROM AUTHOR]

Loss to follow-up is an endemic feature of time-to-event analyses that precludes observation of the event of interest. To our knowledge, in typical cohort studies with encounters occurring at regular or irregular intervals, there is no consensus on how to handle person-time between participants' last study encounter and the point at which they meet a definition of loss to follow-up. We demonstrate, using simulation and an example, that when the event of interest is captured outside of a study encounter (e.g., in a registry), person-time should be censored when the study-defined criterion for loss to follow-up is met (e.g., 1 year after last encounter), rather than at the last study encounter. Conversely, when the event of interest must be measured within the context of a study encounter (e.g., a biomarker value), person-time should be censored at the last study encounter. An inappropriate censoring scheme has the potential to result in substantial bias that may not be easily corrected. [ABSTRACT FROM AUTHOR]

The clustering of human papillomavirus (HPV) infections in some individuals is often interpreted as the result of common risk factors rather than biological interactions between different types of HPV. The intraindividual correlation between times-at-risk for all HPV infections is not generally considered in the analysis of epidemiologic studies. We used a deterministic transmission model to simulate cross-sectional and prospective epidemiologic studies measuring associations between 2 HPV types. When we assumed no interactions, the model predicted that studies would estimate odds ratios and incidence rate ratios greater than 1 between HPV types even after complete adjustment for sexual behavior. We demonstrated that this residual association is due to correlation between the times-at-risk for different HPV types, where individuals become concurrently at risk for all of their partners' HPV types when they enter a partnership and are not at risk when they are single. This correlation can be controlled in prospective studies by restricting analyses to susceptible individuals with an infected sexual partner. The bias in the measured associations was largest in low-sexual-activity populations, cross-sectional studies, and studies which evaluated infection with a first HPV type as the exposure. These results suggest that current epidemiologic evidence does not preclude the existence of competitive biological interactions between HPV types. [ABSTRACT FROM AUTHOR]

Electronic health records (EHRs) are an increasingly utilized resource for clinical research. While their size allows for many analytical opportunities, as with most observational data there is also the potential for bias. One of the key sources of bias in EHRs is what we term informed presence--the notion that inclusion in an EHR is not random but rather indicates that the subject is ill, making people in EHRs systematically different from those not in EHRs. In this article, we use simulated and empirical data to illustrate the conditions under which such bias can arise and how conditioning on the number of health-care encounters can be one way to remove this bias. In doing so, we also show when such an approach can impart M bias, or bias from conditioning on a collider. Finally, we explore the conditions under which number of medical encounters can serve as a proxy for general health. We apply these methods to an EHR data set from a university medical center covering the years 2007-2013. [ABSTRACT FROM AUTHOR]

In time-to-event analyses of observational studies of drug effectiveness, incorrect handling of the period between cohort entry and first treatment exposure during follow-up may result in immortal time bias. This bias can be eliminated by acknowledging a change in treatment exposure status with time-dependent analyses, such as fitting a time-dependent Cox model. The prescription time-distribution matching (PTDM) method has been proposed as a simpler approach for controlling immortal time bias. Using simulation studies and theoretical quantification of bias, we compared the performance of the PTDM approach with that of the time-dependent Cox model in the presence of immortal time. Both assessments revealed that the PTDM approach did not adequately address immortal time bias. Based on our simulation results, another recently proposed observational data analysis technique, the sequential Cox approach, was found to be more useful than the PTDM approach (Cox: bias = -0.002, mean squared error = 0.025; PTDM: bias = -1.411, mean squared error = 2.011). We applied these approaches to investigate the association of β-interferon treatment with delaying disability progression in a multiple sclerosis cohort in British Columbia, Canada (Long-Term Benefits and Adverse Effects of Beta-Interferon for Multiple Sclerosis (BeAMS) Study, 1995-2008). [ABSTRACT FROM AUTHOR]

Longitudinal observational data are required to assess the association between exposure to β-interferon medications and disease progression among relapsing-remitting multiple sclerosis (MS) patients in the “real-world” clinical practice setting. Marginal structural Cox models (MSCMs) can provide distinct advantages over traditional approaches by allowing adjustment for time-varying confounders such as MS relapses, as well as baseline characteristics, through the use of inverse probability weighting. We assessed the suitability of MSCMs to analyze data from a large cohort of 1,697 relapsing-remitting MS patients in British Columbia, Canada (1995–2008). In the context of this observational study, which spanned more than a decade and involved patients with a chronic yet fluctuating disease, the recently proposed “normalized stabilized” weights were found to be the most appropriate choice of weights. Using this model, no association between β-interferon exposure and the hazard of disability progression was found (hazard ratio = 1.36, 95% confidence interval: 0.95, 1.94). For sensitivity analyses, truncated normalized unstabilized weights were used in additional MSCMs and to construct inverse probability weight-adjusted survival curves; the findings did not change. Additionally, qualitatively similar conclusions from approximation approaches to the weighted Cox model (i.e., MSCM) extend confidence in the findings. [ABSTRACT FROM PUBLISHER]

Land-use regression (LUR) models are increasingly used to estimate air pollution exposure in epidemiologic studies. These models use air pollution measurements taken at a small set of locations and modeling based on geographical covariates for which data are available at all study participant locations. The process of LUR model development commonly includes a variable selection procedure. When LUR model predictions are used as explanatory variables in a model for a health outcome, measurement error can lead to bias of the regression coefficients and to inflation of their variance. In previous studies dealing with spatial predictions of air pollution, bias was shown to be small while most of the effect of measurement error was on the variance. In this study, we show that in realistic cases where LUR models are applied to health data, bias in health-effect estimates can be substantial. This bias depends on the number of air pollution measurement sites, the number of available predictors for model selection, and the amount of explainable variability in the true exposure. These results should be taken into account when interpreting health effects from studies that used LUR models. [ABSTRACT FROM PUBLISHER]

We employed the parametric G formula to analyze lung cancer mortality in a cohort of textile manufacturing workers who were occupationally exposed to asbestos in South Carolina. A total of 3,002 adults with a median age of 24 years at enrollment (58% male, 81% Caucasian) were followed for 117,471 person-years between 1940 and 2001, and 195 lung cancer deaths were observed. Chrysotile asbestos exposure was measured in fiber-years per milliliter of air, and annual occupational exposures were estimated on the basis of detailed work histories. Sixteen percent of person-years involved exposure to asbestos, with a median exposure of 3.30 fiber-years/mL among those exposed. Lung cancer mortality by age 90 years under the observed asbestos exposure was 9.44%. In comparison with observed asbestos exposure, if the facility had operated under the current Occupational Safety and Health Administration asbestos exposure standard of <0.1 fibers/mL, we estimate that the cohort would have experienced 24% less lung cancer mortality by age 90 years (mortality ratio = 0.76, 95% confidence interval: 0.62, 0.94). A further reduction in asbestos exposure to a standard of <0.05 fibers/mL was estimated to have resulted in a minimal additional reduction in lung cancer mortality by age 90 years (mortality ratio = 0.75, 95% confidence interval: 0.61, 0.92). [ABSTRACT FROM PUBLISHER]

Collider-stratification bias arises from conditioning on a variable (collider) which opens a path from exposure to outcome. M bias occurs when the collider-stratification bias is transmitted through ancestors of exposure and outcome. Previous theoretical work, but not empirical data, has demonstrated that M bias is smaller than confounding bias. The authors simulated data for large cohort studies with binary exposure, an outcome, a collider, and 2 predictors of the collider. They created 178 scenarios by changing the frequencies of variables and/or the magnitudes of associations among the variables. They calculated the effect estimate, percentage bias, and mean squared error. M bias in these realistic scenarios ranged from −2% to −5%. When the authors increased one or both relative risks for the relation between the collider and unmeasured factors to ≥8, the negative bias was more substantial (>15%). The result was substantially biased (e.g., >20%) if an unmeasured confounder that was also a collider was not adjusted to avoid M bias. In scenarios resembling those the authors examined, M bias had a small impact unless associations between the collider and unmeasured confounders were very large (relative risk > 8). When a collider is itself an important confounder, controlling for confounding would take precedence over avoiding M bias. [ABSTRACT FROM AUTHOR]

Measurement error in both the exposure and the outcome is a common problem in epidemiologic studies. Measurement errors in the exposure and the outcome are said to be independent of each other if the measured exposure and the measured outcome are statistically independent conditional on the true exposure and true outcome (and dependent otherwise). Measurement error is said to be nondifferential if measurement of the exposure does not depend on the true outcome conditional on the true exposure and vice versa; otherwise it is said to be differential. Few results on differential and dependent measurement error are available in the literature. Here the authors use formal rules governing associations on signed directed acyclic graphs (DAGs) to draw conclusions about the presence and sign of causal effects under differential and dependent measurement error. The authors apply these rules to 4 forms of measurement error: independent nondifferential, dependent nondifferential, independent differential, and dependent differential. For a binary exposure and outcome, the authors generalize Weinberg et al.’s (Am J Epidemiol. 1994;140(6):565–571) result for nondifferential measurement error on preserving the direction of a trend to settings which also allow measurement error in the outcome and to cases involving dependent and/or differential error. [ABSTRACT FROM PUBLISHER]

The psychometric properties of instruments used to measure self-reported experiences of discrimination in epidemiologic studies are rarely assessed, especially regarding construct validity. The authors used 2000–2001 data from the Coronary Artery Risk Development in Young Adults (CARDIA) Study to examine differential item functioning (DIF) in 2 versions of the Experiences of Discrimination (EOD) Index, an index measuring self-reported experiences of racial/ethnic and gender discrimination. DIF may confound interpretation of subgroup differences. Large DIF was observed for 2 of 7 racial/ethnic discrimination items: White participants reported more racial/ethnic discrimination for the “at school” item, and black participants reported more racial/ethnic discrimination for the “getting housing” item. The large DIF by race/ethnicity in the index for racial/ethnic discrimination probably reflects item impact and is the result of valid group differences between blacks and whites regarding their respective experiences of discrimination. The authors also observed large DIF by race/ethnicity for 3 of 7 gender discrimination items. This is more likely to have been due to item bias. Users of the EOD Index must consider the advantages and disadvantages of DIF adjustment (omitting items, constructing separate measures, and retaining items). The EOD Index has substantial usefulness as an instrument that can assess self-reported experiences of discrimination. [ABSTRACT FROM PUBLISHER]

In choosing covariates for adjustment or inclusion in propensity score analysis, researchers must weigh the benefit of reducing confounding bias carried by those covariates against the risk of amplifying residual bias carried by unmeasured confounders. The latter is characteristic of covariates that act like instrumental variables—that is, variables that are more strongly associated with the exposure than with the outcome. In this issue of the Journal (Am J Epidemiol. 2011;174(11):1213–1222), Myers et al. compare the bias amplification of a near-instrumental variable with its bias-reducing potential and suggest that, in practice, the latter outweighs the former. The author of this commentary sheds broader light on this comparison by considering the cumulative effects of conditioning on multiple covariates and showing that bias amplification may build up at a faster rate than bias reduction. The author further derives a partial order on sets of covariates which reveals preference for conditioning on outcome-related, rather than exposure-related, confounders. [ABSTRACT FROM PUBLISHER]

Recent theoretical studies have shown that conditioning on an instrumental variable (IV), a variable that is associated with exposure but not associated with outcome except through exposure, can increase both bias and variance of exposure effect estimates. Although these findings have obvious implications in cases of known IVs, their meaning remains unclear in the more common scenario where investigators are uncertain whether a measured covariate meets the criteria for an IV or rather a confounder. The authors present results from two simulation studies designed to provide insight into the problem of conditioning on potential IVs in routine epidemiologic practice. The simulations explored the effects of conditioning on IVs, near-IVs (predictors of exposure that are weakly associated with outcome), and confounders on the bias and variance of a binary exposure effect estimate. The results indicate that effect estimates which are conditional on a perfect IV or near-IV may have larger bias and variance than the unconditional estimate. However, in most scenarios considered, the increases in error due to conditioning were small compared with the total estimation error. In these cases, minimizing unmeasured confounding should be the priority when selecting variables for adjustment, even at the risk of conditioning on IVs. [ABSTRACT FROM PUBLISHER]

The food frequency questionnaire approach to dietary assessment is ubiquitous in nutritional epidemiology research. Food records and recalls provide approaches that may also be adaptable for use in large epidemiologic cohorts, if warranted by better measurement properties. The authors collected (2007–2009) a 4-day food record, three 24-hour dietary recalls, and a food frequency questionnaire from 450 postmenopausal women in the Women’s Health Initiative prospective cohort study (enrollment, 1994–1998), along with biomarkers of energy and protein consumption. Through comparison with biomarkers, the food record is shown to provide a stronger estimate of energy and protein than does the food frequency questionnaire, with 24-hour recalls mostly intermediate. Differences were smaller and nonsignificant for protein density. Food frequencies, records, and recalls were, respectively, able to “explain” 3.8%, 7.8%, and 2.8% of biomarker variation for energy; 8.4%, 22.6%, and 16.2% of biomarker variation for protein; and 6.5%, 11.0%, and 7.0% of biomarker variation for protein density. However, calibration equations that include body mass index, age, and ethnicity substantially improve these numbers to 41.7%, 44.7%, and 42.1% for energy; 20.3%, 32.7%, and 28.4% for protein; and 8.7%, 14.4%, and 10.4% for protein density. Calibration equations using any of the assessment procedures may yield suitable consumption estimates for epidemiologic study purposes. [ABSTRACT FROM PUBLISHER]

The case-crossover method is an efficient study design for evaluating associations between transient exposures and the onset of acute events. In one common implementation of this design, odds ratios are estimated using conditional logistic or stratified Cox proportional hazards models, with data stratified on each individual event. In environmental epidemiology, where aggregate time-series data are often used, combining strata with identical exposure histories may be computationally convenient. However, when the SAS software package (SAS Institute Inc., Cary, North Carolina) is used for analysis, users can obtain biased results if care is not taken to properly account for multiple cases observed at the same time. The authors show that fitting a stratified Cox model with the “Breslow” option for handling tied failure times (i.e., ties = Breslow) provides unbiased health-effects estimates in case-crossover studies with shared exposures. The authors’ simulations showed that using conditional logistic regression—or equivalently a stratified Cox model with the “ties = discrete” option—in this setting leads to health-effect estimates which can be biased away from the null hypothesis of no association by 22%–39%, even for small simulated relative risks. All methods tested by the authors yielded unbiased results under a simulated scenario with a relative risk of 1.0. This potential bias does not arise in R (R Foundation for Statistical Computing, Vienna, Austria) or Stata (Stata Corporation, College Station, Texas). [ABSTRACT FROM PUBLISHER]

The authors use recent methodology in causal inference to disentangle the direct and indirect effects that operate through a mediator in an exposure-response association paradigm. They demonstrate how total effects can be partitioned into direct and indirect effects even when the exposure and mediator interact. The impact of bias due to unmeasured confounding on the exposure-response association is assessed through a series of sensitivity analyses. These methods are applied to a problem in perinatal epidemiology to examine the extent to which the effect of abruption on perinatal mortality is mediated through preterm delivery. Data on over 26 million US singleton births (1995–2002) were utilized. Risks of mortality among abruption and nonabruption births were 102.7 and 6.2 per 1,000 births, respectively. Risk ratios of the natural direct and indirect (preterm delivery-mediated) effects of abruption on mortality were 10.18 (95% confidence interval: 9.80, 10.58) and 1.35 (95% confidence interval: 1.33, 1.38), respectively. The proportion of increased mortality risk mediated through preterm delivery was 28.1%, with even higher proportions associated with deliveries at earlier gestational ages. Sensitivity analyses underscore that the qualitative conclusions of some mediated effects and substantial direct effects are reasonably robust to unmeasured confounding of a fairly considerable magnitude. [ABSTRACT FROM AUTHOR]

In longitudinal studies of developmental and disease processes, participants are followed prospectively with intermediate milestones identified as they occur. Frequently, studies enroll participants over a range of ages including ages at which some participants’ milestones have already passed. Ages at milestones that occur prior to study entry are left censored if individuals are enrolled in the study or left truncated if they are not. The authors examined the bias incurred by ignoring these issues when estimating the distribution of age at milestones or the time between 2 milestones. Methods that account for left truncation and censoring are considered. Data on the menopausal transition are used to illustrate the problem. Simulations show that bias can be substantial and that standard errors can be severely underestimated in naïve analyses that ignore left truncation. Bias can be reduced when analyses account for left truncation, although the results are unstable when the fraction truncated is high. Simulations suggest that a better solution, when possible, is to modify the study design so that information on current status (i.e., whether or not a milestone has passed) is collected on all potential participants, analyzing those who are past the milestone at the time of recruitment as left censored rather than excluding such individuals from the analysis. [ABSTRACT FROM PUBLISHER]

Previous injury is believed to be a causal risk factor for subsequent injury. Using empirical data on circus artists (n = 1,281 artists) between 2004 and 2008 in Montreal, Canada, as a motivating example, the authors use patient vector plots to demonstrate that a bias away from the null must always occur in the typical analyses cited as evidence (i.e., survival analysis, Poisson regression), except in the improbable context where all subjects have the same inherent risk independent of previous injury. In addition, using simulated data, the authors demonstrate that a simple method that conditions on the individual will approximate conclusions from more complex analytical methods. By using the typical analysis of the authors’ empirical data, Kaplan-Meier curves and Cox regression suggested increasing injury rates for both the second and third injuries compared with the first injury. However, conditional analyses using a matched population (i.e., time to first, second, and third injuries among artists with 3 or more injuries) showed that injury rates were unchanged for both the second and third injuries compared with the first injury. These results suggest that previous injury should not be evaluated as a causal risk factor unless one conditions on the individual in some way. [ABSTRACT FROM PUBLISHER]

Repeat-biomarker measurement error models accounting for systematic correlated within-person error can be used to estimate the correlation coefficient (ρ) and deattenuation factor (λ), used in measurement error correction. These models account for correlated errors in the food frequency questionnaire (FFQ) and the 24-hour diet recall and random within-person variation in the biomarkers. Failure to account for within-person variation in biomarkers can exaggerate correlated errors between FFQs and 24-hour diet recalls. For 2 validation studies, ρ and λ were calculated for total energy and protein density. In the Automated Multiple-Pass Method Validation Study (n = 471), doubly labeled water (DLW) and urinary nitrogen (UN) were measured twice in 52 adults approximately 16 months apart (2002—2003), yielding intraclass correlation coefficients of 0.43 for energy (DLW) and 0.54 for protein density (UN/DLW). The deattenuated correlation coefficient for protein density was 0.51 for correlation between the FFQ and the 24-hour diet recall and 0.49 for correlation between the FFQ and the biomarker. Use of repeat-biomarker measurement error models resulted in a ρ of 0.42. These models were similarly applied to the Observing Protein and Energy Nutrition Study (1999—2000). In conclusion, within-person variation in biomarkers can be substantial, and to adequately assess the impact of correlated subject-specific error, this variation should be assessed in validation studies of FFQs. [ABSTRACT FROM AUTHOR]

Propensity-score matching is increasingly being used to estimate the effects of treatments using observational data. In many-to-one (M:1) matching on the propensity score, M untreated subjects are matched to each treated subject using the propensity score. The authors used Monte Carlo simulations to examine the effect of the choice of M on the statistical performance of matched estimators. They considered matching 1–5 untreated subjects to each treated subject using both nearest-neighbor matching and caliper matching in 96 different scenarios. Increasing the number of untreated subjects matched to each treated subject tended to increase the bias in the estimated treatment effect; conversely, increasing the number of untreated subjects matched to each treated subject decreased the sampling variability of the estimated treatment effect. Using nearest-neighbor matching, the mean squared error of the estimated treatment effect was minimized in 67.7% of the scenarios when 1:1 matching was used. Using nearest-neighbor matching or caliper matching, the mean squared error was minimized in approximately 84% of the scenarios when, at most, 2 untreated subjects were matched to each treated subject. The authors recommend that, in most settings, researchers match either 1 or 2 untreated subjects to each treated subject when using propensity-score matching. [ABSTRACT FROM PUBLISHER]

Frailty, a poorly measured confounder in older patients, can promote treatment in some situations and discourage it in others. This can create unmeasured confounding and lead to nonuniform treatment effects over the propensity score (PS). The authors compared bias and mean squared error for various PS implementations under PS trimming, thereby excluding persons treated contrary to prediction. Cohort studies were simulated with a binary treatment T as a function of 8 covariates X. Two of the covariates were assumed to be unmeasured strong risk factors for the outcome and present in persons treated contrary to prediction. The outcome Y was simulated as a Poisson function of T and all X’s. In analyses based on measured covariates only, the range of PS's was trimmed asymmetrically according to the percentile of PS in treated patients at the lower end and in untreated patients at the upper end. PS trimming reduced bias due to unmeasured confounders and mean squared error in most scenarios assessed. Treatment effect estimates based on PS range restrictions do not correspond to a causal parameter but may be less biased by such unmeasured confounding. Increasing validity based on PS trimming may be a unique advantage of PS's over conventional outcome models. [ABSTRACT FROM PUBLISHER]

Properly planned and conducted randomized clinical trials remain susceptible to a lack of external validity. The authors illustrate a model-based method to standardize observed trial results to a specified target population using a seminal human immunodeficiency virus (HIV) treatment trial, and they provide Monte Carlo simulation evidence supporting the method. The example trial enrolled 1,156 HIV-infected adult men and women in the United States in 1996, randomly assigned 577 to a highly active antiretroviral therapy and 579 to a largely ineffective combination therapy, and followed participants for 52 weeks. The target population was US people infected with HIV in 2006, as estimated by the Centers for Disease Control and Prevention. Results from the trial apply, albeit muted by 12%, to the target population, under the assumption that the authors have measured and correctly modeled the determinants of selection that reflect heterogeneity in the treatment effect. In simulations with a heterogeneous treatment effect, a conventional intent-to-treat estimate was biased with poor confidence limit coverage, but the proposed estimate was largely unbiased with appropriate confidence limit coverage. The proposed method standardizes observed trial results to a specified target population and thereby provides information regarding the generalizability of trial results. [ABSTRACT FROM PUBLISHER]

To estimate the net effect of imperfectly measured highly active antiretroviral therapy on incident acquired immunodeficiency syndrome or death, the authors combined inverse probability-of-treatment-and-censoring weighted estimation of a marginal structural Cox model with regression-calibration methods. Between 1995 and 2007, 950 human immunodeficiency virus–positive men and women were followed in 2 US cohort studies. During 4,054 person-years, 374 initiated highly active antiretroviral therapy, 211 developed acquired immunodeficiency syndrome or died, and 173 dropped out. Accounting for measured confounders and determinants of dropout, the weighted hazard ratio for acquired immunodeficiency syndrome or death comparing use of highly active antiretroviral therapy in the prior 2 years with no therapy was 0.36 (95% confidence limits: 0.21, 0.61). This association was relatively constant over follow-up (P = 0.19) and stronger than crude or adjusted hazard ratios of 0.75 and 0.95, respectively. Accounting for measurement error in reported exposure using external validation data on 331 men and women provided a hazard ratio of 0.17, with bias shifted from the hazard ratio to the estimate of precision as seen by the 2.5-fold wider confidence limits (95% confidence limits: 0.06, 0.43). Marginal structural measurement-error models can simultaneously account for 3 major sources of bias in epidemiologic research: validated exposure measurement error, measured selection bias, and measured time-fixed and time-varying confounding. [ABSTRACT FROM PUBLISHER]

In alcohol epidemiology surveys, there is a tradition of measuring alcohol-related consequences using respondents’ attribution of alcohol as the cause. The authors aimed to compare the prevalence and frequency of self-attributed consequences to consequences without self-attribution using alcohol-attributable fractions (AAF). In 2007, a total of 7,174 Swiss school students aged 13–16 years reported the numbers of 6 alcohol-related adverse consequences (e.g., fights, injuries) they had incurred in the past 12 months. Consequences were measured with and without attribution of alcohol as the cause. The alcohol-use measures were frequency and volume of drinking in the past 12 months and number of risky single-occasion (≥5 drinks) drinking episodes in the past 30 days. Attributable fractions were derived from logistic (≥1 incident) and Poisson (number of incidents) regression analyses. Although relative risk estimates were higher when alcohol-attributed consequences were compared with nonattributed consequences, the use of AAFs resulted in more alcohol-related consequences (10,422 self-attributed consequences vs. 24,520 nonattributed consequences determined by means of AAFs). The likelihood of underreporting was higher among drinkers with intermediate frequencies than among either rare drinkers or frequent drinkers. Therefore, the extent of alcohol-related adverse consequences among adolescents may be underestimated when using self-attributed consequences, because of differential attribution processes, especially among infrequent drinkers. [ABSTRACT FROM PUBLISHER]

According to the authors, time-modified confounding occurs when the causal relation between a time-fixed or time-varying confounder and the treatment or outcome changes over time. A key difference between previously described time-varying confounding and the proposed time-modified confounding is that, in the former, the values of the confounding variable change over time while, in the latter, the effects of the confounder change over time. Using marginal structural models, the authors propose an approach to account for time-modified confounding when the relation between the confounder and treatment is modified over time. An illustrative example and simulation show that, when time-modified confounding is present, a marginal structural model with inverse probability-of-treatment weights specified to account for time-modified confounding remains approximately unbiased with appropriate confidence limit coverage, while models that do not account for time-modified confounding are biased. Correct specification of the treatment model, including accounting for potential variation over time in confounding, is an important assumption of marginal structural models. When the effect of confounders on either the treatment or outcome changes over time, time-modified confounding should be considered. [ABSTRACT FROM PUBLISHER]

Several observational studies have found a higher risk of recurrence/relapse of depression for patients who discontinue antidepressant use compared with those who continue. This study demonstrated that measurement of follow-up time can be subject to immortal and neglected time bias. Data were obtained from the 2001 Second Dutch National Survey of General Practice. The study population was composed of antidepressant users with a registered depression diagnosis, divided into early discontinuers and continuing users. Two methods were used to measure time to relapse/recurrence. Method 1, used in previously mentioned studies, measured the beginning of follow-up 6 months after starting antidepressant therapy. Method 2 constructed individual treatment episodes for each patient and measured follow-up from actual end-of-treatment episode. The Cox proportional hazards model produced a risk ratio of 1.58 (95% confidence interval: 1.02, 2.45) for method 1, suggesting a higher risk of relapse/recurrence for early discontinuers. In method 2, a statistically nonsignificant risk ratio of 0.77 (95% confidence interval: 0.49, 1.21) was produced, indicating no difference in risk of relapse/recurrence. The authors found the method used in previous studies subject to bias. Applying a different method, accounting for immortal and neglected time bias, eliminated the protective effects of longer treatments. [ABSTRACT FROM PUBLISHER]