Biased Exposure-Health Effect Estimates from Selection in Cohort Studies: Are Environmental Studies at Particular Risk?

Abstract

Background: The process of creating a cohort or cohort substudy may induce misleading exposure-health effect associations through collider stratification bias (i.e., selection bias) or bias due to conditioning on an intermediate. Studies of environmental risk factors may be at particular risk.

Objectives: We aimed to demonstrate how such biases of the exposure-health effect association arise and how one may mitigate them.

Methods: We used directed acyclic graphs and the example of bone lead and mortality (all-cause, cardiovascular, and ischemic heart disease) among 835 white men in the Normative Aging Study (NAS) to illustrate potential bias related to recruitment into the NAS and the bone lead substudy. We then applied methods (adjustment, restriction, and inverse probability of attrition weighting) to mitigate these biases in analyses using Cox proportional hazards models to estimate adjusted hazard ratios (HRs) and 95% confidence intervals (CIs).

Results: Analyses adjusted for age at bone lead measurement, smoking, and education among all men found HRs (95% CI) for the highest versus lowest tertile of patella lead of 1.34 (0.90, 2.00), 1.46 (0.86, 2.48), and 2.01 (0.86, 4.68) for all-cause, cardiovascular, and ischemic heart disease mortality, respectively. After applying methods to mitigate the biases, the HR (95% CI) among the 637 men analyzed were 1.86 (1.12, 3.09), 2.47 (1.23, 4.96), and 5.20 (1.61, 16.8), respectively.

Conclusions: Careful attention to the underlying structure of the observed data is critical to identifying potential biases and methods to mitigate them. Understanding factors that influence initial study participation and study loss to follow-up is critical. Recruitment of population-based samples and enrolling participants at a younger age, before the potential onset of exposure-related health effects, can help reduce these potential pitfalls.

Citation: Weisskopf MG, Sparrow D, Hu H, Power MC. 2015. Biased exposure-health effect estimates from selection in cohort studies: are environmental studies at particular risk? Environ Health Perspect 123:1113-1122; http://dx.doi.org/10.1289/ehp.1408888.

Figure 1

(A) Introduction to causal DAGs: A, B, C, D, E, F, G represent variables, or nodes, and directional arrows indicate causal relationships between these variables. A variable with two arrows pointing into it (a common effect of the two variables, e.g., D) is referred to as a collider. (B) Conditioning on a variable (either by restriction, stratification, or adjustment in a model) is indicated by drawing a box around the variable, as shown for variables C and D. See text for details on how DAGs indicate causal and noncausal associations between variables.

Figure 2

DAG representation of our assumptions about the structure of the data in the study, where we are interested in estimating the causal effects of cumulative lead exposure on mortality. See “Methods” for details. Abbreviations: CV, cardiovascular symptoms; L, measured variables; Pb, lead exposure/bone lead concentration; S_KXRF, selection into the KXRF substudy; S_NAS, selection into the NAS cohort; U, unmeasured variables. Subscripts 0 and 1 refer to the time of NAS recruitment and KXRF measurement, respectively. The U and L variables are separate variables, but the structure of arrows into and out of them are the same, and so for simplicity they are shown together.

Figure 3

Illustration of selection bias (or collider stratification bias) (A) at the time of entry into the NAS and (B) due to loss to follow-up before KXRF lead measurements. See “Methods” for details. See Figure 2 for variable definitions.

Figure 4

Illustration of bias due to conditioning on an intermediate (A) at the time of entry into the NAS and (B) due to loss to follow-up before KXRF lead measurements. See “Methods” for details. See Figure 2 for variable definitions.

Figure 5

Demonstration of impact of efforts to alleviate bias due to collider stratification bias and conditioning on an intermediate (CV). (A) DAG reflecting structure of our data in the base analysis among white men adjusting for age, education, and smoking (Model 1). (B) DAG reflecting data structure after additional regression adjustment (model 2) under the assumption that we no longer have important uncontrolled variables (U), although we recognize that we cannot rule out such variables entirely. (C) DAG reflecting data structure after additionally restricting to those ≤ 45 years of age at NAS entry. (D) DAG reflecting data structure after additionally using IPW to account for loss to follow-up between cohort inception and KXRF among those ≤ 45 years of age at baseline. See “Methods” for details. See Figure 2 for variable definitions.