Selection Bias Requires Selection: The Case of Collider Stratification Bias.

In epidemiology, collider stratification bias, the bias resulting from conditioning on a common effect of two causes, is oftentimes considered a type of selection bias, regardless of the conditioning methods employed. In this commentary, we distinguish between two types of collider stratification bias: collider restriction bias due to restricting to one level of a collider (or a descendant of a collider) and collider adjustment bias through inclusion of a collider (or a descendant of a collider) in a regression model. We argue that categorizing collider adjustment bias as a form of selection bias may lead to semantic confusion, as adjustment for a collider in a regression model does not involve selecting a sample for analysis. Instead, we propose that collider adjustment bias can be better viewed as a type of overadjustment bias. We further provide two distinct causal diagram structures to distinguish collider restriction bias and collider adjustment bias. We hope that such a terminological distinction can facilitate easier and clearer communication.

Glucosamine and lower mortality and cancer incidence: Selection bias in the observational studies.

BACKGROUND: Glucosamine is a widely used supplement to treat joint pain and osteoarthritis despite inconclusive randomized trial results on its effectiveness. In contrast, observational studies associate glucosamine with significant reductions in mortality and cancer incidence. We evaluated the extent of bias, particularly selection bias, to explain these surprising beneficial effects. METHODS: We searched the literature to identify all observational studies reporting on the effect of glucosamine use on major outcomes. RESULTS: We identified 11 observational studies, reporting a mean 16% reduction in all-cause mortality (hazard ratio [HR] 0.84, 95% CI: 0.81-0.87) with glucosamine use, as well as significant reductions in cancer incidence and other major diseases including cardiovascular, respiratory and diabetes. We show that these significant effects can result from selection bias due to collider stratification, as all studies used "prevalent" cohorts, where glucosamine use started before cohort entry, and where subjects agreed to join the cohorts. Our illustration of the bias using the UK Biobank publication involving a half-million subjects shows how a true rate ratio of mortality of 1.0 in the population can result in a biased rate ratio of 0.82 in the prevalent cohort. CONCLUSIONS: The observational studies reporting significant reductions in mortality, cancer incidence and other outcomes with glucosamine were affected by selection bias from collider stratification. In the absence of properly conducted observational studies that circumvent this bias by considering "new users", the studies to date cannot support the prescription of this supplement as a preventive measure for mortality, cancer, and other chronic diseases.

Obesity paradox in heart failure: statistical artifact, or impetus to rethink clinical practice?

The "obesity paradox" in heart failure (HF) is a phenomenon of more favorable prognosis, especially better survival, in obese versus normal-weight HF patients. Various explanations for the paradox have been offered; while different in their details, they typically share the premise that obesity per se is not actually the cause of reduced mortality in HF. Even so, there is a lingering question of whether clinicians should refrain from, or at least soft-pedal on, encouraging weight loss among their obese HF patients. Against the backdrop of recent epidemiological analysis by Banack and Kaufman, which speculates that collider stratification bias may generate the obesity paradox, we seek to address the aforementioned question. Following a literature review, which confirms that obese HF patients are demographically and clinically different from their normal-weight counterparts, we present four hypothetical data sets to illustrate a spectrum of possibilities regarding the obesity-mortality association. Importantly, these hypothetical data sets become indistinguishable from each other when a crucial variable is unmeasured or unreported. While thorough, the discussion of these data sets is intended to be accessible to a wide audience, especially including clinicians, without a prerequisite of familiarity with advanced epidemiology. We also furnish intuitive visual diagrams which depict a version of the obesity paradox. These illustrations, along with reflection on the distinction between weight and weight loss (and, furthermore, between voluntary and involuntary weight loss), lead to our recommendation for clinicians regarding the encouragement of weight loss. Finally, our conclusion explicitly addresses the questions posed in the title of this article.

Confounding, causality, and confusion: the role of intermediate variables in interpreting observational studies in obstetrics.

Prospective and retrospective cohorts and case-control studies are some of the most important study designs in epidemiology because, under certain assumptions, they can mimic a randomized trial when done well. These assumptions include, but are not limited to, properly accounting for 2 important sources of bias: confounding and selection bias. While not adjusting the causal association for an intermediate variable will yield an unbiased estimate of the exposure-outcome's total causal effect, it is often that obstetricians will want to adjust for an intermediate variable to assess if the intermediate is the underlying driver of the association. Such a practice must be weighed in light of the underlying research question and whether such an adjustment is necessary should be carefully considered. Gestational age is, by far, the most commonly encountered variable in obstetrics that is often mislabeled as a confounder when, in fact, it may be an intermediate. If, indeed, gestational age is an intermediate but if mistakenly labeled as a confounding variable and consequently adjusted in an analysis, the conclusions can be unexpected. The implications of this overadjustment of an intermediate as though it were a confounder can render an otherwise persuasive study downright meaningless. This commentary provides an exposition of confounding bias, collider stratification, and selection biases, with applications in obstetrics and perinatal epidemiology.

A Simulation Platform for Quantifying Survival Bias: An Application to Research on Determinants of Cognitive Decline.

Bias due to selective mortality is a potential concern in many studies and is especially relevant in cognitive aging research because cognitive impairment strongly predicts subsequent mortality. Biased estimation of the effect of an exposure on rate of cognitive decline can occur when mortality is a common effect of exposure and an unmeasured determinant of cognitive decline and in similar settings. This potential is often represented as collider-stratification bias in directed acyclic graphs, but it is difficult to anticipate the magnitude of bias. In this paper, we present a flexible simulation platform with which to quantify the expected bias in longitudinal studies of determinants of cognitive decline. We evaluated potential survival bias in naive analyses under several selective survival scenarios, assuming that exposure had no effect on cognitive decline for anyone in the population. Compared with the situation with no collider bias, the magnitude of bias was higher when exposure and an unmeasured determinant of cognitive decline interacted on the hazard ratio scale to influence mortality or when both exposure and rate of cognitive decline influenced mortality. Bias was, as expected, larger in high-mortality situations. This simulation platform provides a flexible tool for evaluating biases in studies with high mortality, as is common in cognitive aging research.

Selecting on treatment: a pervasive form of bias in instrumental variable analyses.

Instrumental variable (IV) methods are increasingly being used in comparative effectiveness research. Studies using these methods often compare 2 particular treatments, and the researchers perform their IV analyses conditional on patients' receiving this subset of treatments (while ignoring the third option of "neither treatment"). The ensuing selection bias that occurs due to this restriction has gone relatively unnoticed in interpretations and discussions of these studies' results. In this paper we describe the structure of this selection bias with examples drawn from commonly proposed instruments such as calendar time and preference, illustrate the bias with causal diagrams, and estimate the magnitude and direction of possible bias using simulations. A noncausal association between the proposed instrument and the outcome can occur in analyses restricted to patients receiving a subset of the possible treatments. This results in bias in the numerator for the standard IV estimator; the bias is amplified in the treatment effect estimate. The direction and magnitude of the bias in the treatment effect estimate are functions of the distribution of and relationships between the proposed instrument, treatment values, unmeasured confounders, and outcome. IV methods used to compare a subset of treatment options are prone to substantial biases, even when the proposed instrument appears relatively strong.

Assessing the possible direct effect of birth weight on childhood blood pressure: a sensitivity analysis.

To estimate the possible direct effect of birth weight on blood pressure, it is conventional to condition on the mediator, current weight. Such conditioning can induce bias. Our aim was to assess the potential biasing effect of U, an unmeasured common cause of current weight and blood pressure, on the estimate of the controlled direct effect of birth weight on blood pressure, with the help of sensitivity analyses. We used data from a school-based study conducted in Switzerland in 2005-2006 (n = 3,762; mean age = 12.3 years). A small negative association was observed between birth weight and systolic blood pressure (linear regression coefficient ßbw = -0.3 mmHg/kg, 95% confidence interval: -0.9, 0.3). The association was strengthened upon adjustment for current weight (ßbw|C = -1.5 mmHg/kg, 95% confidence interval: -2.1, -0.9). Sensitivity analyses revealed that the negative conditional association was explained by U only if U was relatively strongly associated with blood pressure and if there was a large difference in the prevalence of U between low-birth weight and normal-birth weight children. This weakens the hypothesis that the negative relationship between birth weight and blood pressure arises only from collider-stratification bias induced by conditioning on current weight.

Selection Bias in Health Research: Quantifying, Eliminating, or Exacerbating Health Disparities?

Purpose of review: To summarize recent literature on selection bias in disparities research addressing either descriptive or causal questions, with examples from dementia research. Recent findings: Defining a clear estimand, including the target population, is essential to assess whether generalizability bias or collider-stratification bias are threats to inferences. Selection bias in disparities research can result from sampling strategies, differential inclusion pipelines, loss to follow-up, and competing events. If competing events occur, several potentially relevant estimands can be estimated under different assumptions, with different interpretations. The apparent magnitude of a disparity can differ substantially based on the chosen estimand. Both randomized and observational studies may misrepresent health disparities or heterogeneity in treatment effects if they are not based on a known sampling scheme. Conclusion: Researchers have recently made substantial progress in conceptualization and methods related to selection bias. This progress will improve the relevance of both descriptive and causal health disparities research.

Rethinking animal attrition in preclinical research: Expressing causal mechanisms of selection bias using directed acyclic graphs.

Animal attrition in preclinical experiments can introduce bias in the estimation of causal treatment effects, as the treatment-outcome association in surviving animals may not represent the causal effect of interest. This can compromise the internal validity of the study despite randomization at the outset. Directed Acyclic Graphs (DAGs) are useful tools to transparently visualize assumptions about the causal structure underlying observed data. By illustrating relationships between relevant variables, DAGs enable the detection of even less intuitive biases, and can thereby inform strategies for their mitigation. In this study, we present an illustrative causal model for preclinical stroke research, in which animal attrition induces a specific type of selection bias (i.e., collider stratification bias) due to the interplay of animal welfare, initial disease severity and negative side effects of treatment. Even when the treatment had no causal effect, our simulations revealed substantial bias across different scenarios. We show how researchers can detect and potentially mitigate this bias in the analysis phase, even when only data from surviving animals are available, if knowledge of the underlying causal process that gave rise to the data is available. Collider stratification bias should be a concern in preclinical animal studies with severe side effects and high post-randomization attrition.

Alzheimer's disease genetic risk variants beyond APOE ε4 predict mortality.

INTRODUCTION: We hypothesized that, like apolipoprotein E (APOE), other late-onset Alzheimer's disease (LOAD) genetic susceptibility loci predict mortality. METHODS: We used a weighted genetic risk score (GRS) from 21 non-APOE LOAD risk variants to predict survival in the Adult Changes in Thought and the Health and Retirement Studies. We meta-analyzed hazard ratios and examined models adjusted for cognitive performance or limited to participants with dementia. For replication, we assessed the GRS-longevity association in the Cohorts for Heart and Aging Research in Genomic Epidemiology, comparing cases surviving to age ≥90 years with controls who died between ages 55 and 80 years. RESULTS: Higher GRS predicted mortality (hazard ratio = 1.05; 95% confidence interval: 1.00-1.10, P = .04). After adjusting for cognitive performance or restricting to participants with dementia, the relationship was attenuated and no longer significant. In case-control analysis, the GRS was associated with reduced longevity (odds ratio = 0.64; 95% confidence interval: 0.41-1.00, P = .05). DISCUSSION: Non-APOE LOAD susceptibility loci confer risk for mortality, likely through effects on dementia incidence.

Application of directed acyclic graphs in identification and control of selection bias in causal inference / 中华疾病控制杂志

In the etiology study of epidemiology, selection bias will lead to the fact that the research sample cannot represent the general population, the association between exposure and outcome among those selected for analysis differs from the association among those eligible, and the true causal association cannot be inferred. Directed acyclic graphs (DAGs) could visualize complex causality, introduce the Collider-stratification bias using simple graphics language, provide a simple and intuitive way to identify Selection bias, different types of selection bias are verified by the graphic structure of the Collider-stratification bias. In practical studies, there may be multiple biases at the same time, improper adjustment of the collider will lead to Collider-stratification bias, open a backdoor path, even change the size and direction of the confounding bias. In order to obtain an unbiased estimate of the exposure to the outcome, it is necessary to identify the collider and avoid the adjustment to prevent the occurrence of Collider-stratification bias by using DAGs.

Application of directed acyclic graphs in evidence-based medicine / 中国医师杂志

Evidence-based medicine (EBM) is a kind of clinic practice where clinicians use the best and the latest available evidence to diagnose and treat patients, and both evidence providers and users need to identify and control different kinds of biases in medical research.Directed acyclic graphsis is a tool to explore the causal relationship.The possible biases in the study can be revealed in a simple graphical language.The use of directed acyclic graphs could avoid the occurrence of bias and improve the quality of medical research and better guide clinical practice.