Adjustment for selection bias in observational studies with application to the analysis of autopsy data

Abstract

Background: The interpretation of neuropathological studies of dementia and Alzheimer's disease is complicated by potential selection mechanisms that can drive whether or not a study participant is observed to undergo autopsy. Notwithstanding this, there appears to have been little emphasis placed on potential selection bias in published reports from population-based neuropathological studies of dementia.

Methods: We provide an overview of methodological issues relating to the identification of and adjustment for selection bias. When information is available on factors that govern selection, inverse-probability weighting provides an analytic approach to adjust for selection bias. The weights help alleviate bias by serving to bridge differences between the population from which the observed data may be viewed as a representative sample and the target population, identified as being of scientific interest.

Results: We illustrate the methods with data obtained from the Adult Changes in Thought study. Adjustment for potential selection bias yields substantially strengthened association between neuropathological measurements and risk of dementia.

Conclusions: Armed with analytic techniques to adjust for selection bias and to ensure generalizability of results from population-based neuropathological studies, researchers should consider incorporating information related to selection into their data collection schemes.

Fig. 1

Directed acyclic graph illustrating potential selection bias arising from Berkson's paradox.

Fig. 2

Directed acyclic graph illustrating potential selection bias in the context of a case-control study.

Fig. 3

Flowchart indicating the progression of ACT enrollees from baseline to status at last known follow-up. Among those with at least 1 follow-up visit, individuals are grouped according to vital status, whether or not they had withdrawn and, among those that died, whether or not an autopsy was performed. Number of patients with a clinical diagnosis of dementia during follow-up is given in parentheses.

Fig. 4

Directed acyclic graph showing the conditional independencies for the dementia and selection models. Note, NP risk factors (X) are only observed on the autopsy sample; all other covariates are observed on all individuals in ACT. Confounders are split into two groups: those that influence selection (C₀) and those that do not (C₁).

Fig. 5

Estimated odds ratio association (with pointwise 95% CI) between age and risk of ‘selection’, based on the saturated selection model. The functional form of the association is modeled via a natural smoothing spline (4 degrees of freedom), with age 85 years taken as the referent age group.