The effect of survival bias on case-control genetic association studies of highly lethal diseases

Abstract

Background: Survival bias is the phenomenon by which individuals are excluded from analysis of a trait because of mortality related to the expression of that trait. In genetic association studies, variants increasing risk for disease onset as well as risk of disease-related mortality (lethality) could be difficult to detect in genetic association case-control designs, possibly leading to underestimation of a variant's effect on disease risk.

Methods and results: We modeled cohorts for 3 diseases of high lethality (intracerebral hemorrhage, ischemic stroke, and myocardial infarction) using existing longitudinal data. Based on these models, we simulated case-control genetic association studies for genetic risk factors of varying effect sizes, lethality, and minor allele frequencies. For each disease, erosion of detected effect size was larger for case-control studies of individuals of advanced age (age >75 years) and/or variants with very high event-associated lethality (genotype relative risk for event-related death >2.0). We found that survival bias results in no more than 20% effect size erosion for cohorts with mean age <75 years, even for variants that double lethality risk. Furthermore, we found that increasing effect size erosion was accompanied by depletion of minor allele frequencies in the case population, yielding a "signature" of the presence of survival bias.

Conclusions: Our simulation provides formulas to allow estimation of effect size erosion given a variant's odds ratio of disease, odds ratio of lethality, and minor allele frequencies. These formulas will add precision to power calculation and replication efforts for case-control genetic studies. Our approach requires validation using prospective data.

Figure 1

AIS = Acute Ischemic Stroke. ICH = Intracerebral Hemorrhage. MI = Myocardial Infarction. Bars represent the incidence and mortality estimates for each disease for the age ranges specified, according to the published reports for each cohort. Lines represent the interpolated incidence and mortality curves used to create the disease models for simulation of longitudinal and case-control study designs.

Figure 2

Simulation design for longitudinal study model (A) and simulation design for case-control study model (B). For each panel, solid circles represent the non-diseased population from which controls are recruited, and dashed circles represent the diseased population from which cases are recruited. In Panel B, the dashed square represents the individuals who experienced disease-related mortality, and are therefore not present in the pool of diseased individuals for case recruitment.

Figure 3

Effect size erosion in case-control design is plotted according to varying cohort ages and genetic relative risk of lethality (GRR-L) for acute ischemic stroke. The median (thick line) and range (box-plot with whiskers) for each GRR-L simulation reflects the ranges of minor allele frequency and genetic relative risk of disease. The top panel assumes 0% ascertainment of lethal cases, while the bottom panel assumes 40% ascertainment of lethal cases.

Figure 4

Effect size erosion in case-control design is plotted according to varying cohort ages and genetic relative risk of lethality (GRR-L) for intracerebral hemorrhage. The median (thick line) and range (box-plot with whiskers) for each GRR-L simulation reflects the ranges of minor allele frequency and genetic relative risk of disease. The top panel assumes 0% ascertainment of lethal cases, while the bottom panel assumes 40% ascertainment of lethal cases.

Figure 5

Effect size erosion in case-control design is plotted according to varying cohort ages and genetic relative risk of lethality (GRR-L) for myocardial infarction. The median (thick line) and range (box-plot with whiskers) for each GRR-L simulation reflects the ranges of minor allele frequency and genetic relative risk of disease. The top panel assumes 0% ascertainment of lethal cases, while the bottom panel assumes 40% ascertainment of lethal cases.