Population inference with mortality and attrition in longitudinal studies on aging: a two-stage multiple imputation method.

A major challenge for inference regarding aging-related change in longitudinal studies is that of study attrition and population mortality. Inferences in longitudinal studies can account for attrition and mortality-related change as distinct processes, but this is made difficult when follow-up of all individuals (i.e., age at death) is not complete. This is a common problem because most longitudinal studies of aging either have incomplete follow-up or are still collecting data on subsequent outcomes, including time of death. A statistical approach is suggested for including time-to-death as a predictor in models with incomplete follow-up using a two-stage multiple-imputation procedure. An empirical example using data from the OCTO-Twin study is presented that shows the utility of his procedure for making inferences conditional on mortality when mortality data are incomplete.