Abstract
SummaryWe propose semi-parametric methods to model cohort data where repeated outcomes may be missing due to death and non-ignorable dropout. Our focus is to obtain inference about the cohort composed of those who are still alive at any time point (partly conditional inference). We propose: i) an inverse probability weighted method that upweights observed subjects to represent subjects who are still alive but are not observed; ii) an outcome regression method that replaces missing outcomes of subjects who are alive with their conditional mean outcomes given past observed data; and iii) an augmented inverse probability method that combines the previous two methods and is double robust against model misspecification. These methods are described for both monotone and non-monotone missing data patterns, and are applied to a cohort of elderly adults from the Health and Retirement Study. Sensitivity analysis to departures from the assumption that missingness at some visit t is independent of the outcome at visit t given past observed data and time of death is used in the data application.
Highlights
In studies of the elderly, deaths occur frequently during follow-up and in most cases, truncate the outcome process
We focus on partly conditional inference; the survivor average causal effect” (SACE) is discussed in Web Appendix A
We show the double robustness of the proposed augmented IPW (AIPW) method in the monotone study by replacing U by X in the missingness or regression models at all visits, and in the non-monotone study by omitting U from the regression model at visit 4 and from the missingness model at visit 5
Summary
In studies of the elderly, deaths occur frequently during follow-up and in most cases, truncate the outcome process. Some statistical methods do not make this distinction (e.g., linear mixed-effects models, LMM), and estimate the mean or distribution of an outcome in the whole cohort, including subjects who are no longer alive. In doing so, these methods explicitly or implicitly impute post-death outcomes, as though the outcome process. Such methods are said to produce “immortal cohort inference” or “unconditional inference” (Dufouil et al, 2004). Methods that distinguish between dropout and death, and estimate the mean or distribution of the outcomes in the subjects who are alive provide “mortal cohort inference.”
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
