|
Abstract:
|
Cognitive decline is often a source of nonignorable missingness in longitudinal studies of older adults, because cognitively impaired participants may be unable to respond to interview questions. Therefore, researchers may recruit proxy respondents, e.g., relatives or caregivers, for epidemiologic studies. Since proxies are typically recruited only to report on behalf of participants who have missing self-reported outcomes, either a proxy report or participant self-report, but not both, is available for each participant. When outcomes are binary and self-reports are the gold standard measures, substituting proxy reports for missing participant self-reports in statistical analysis can introduce misclassification error and produce biased estimates. To address this problem, we propose a pattern-mixture model that uses proxy reports to reduce selection bias from missing outcomes, and we describe a sensitivity analysis to overcome bias from differential outcome misclassification. We perform model estimation with potentially high-dimensional covariates using propensity-score stratification and multiple imputation. We perform simulations and apply the methods to the Baltimore Hip Studies.
|
