Title: Methods of handling cohort data with death and non-ignorable dropout
Authors: Lan Wen - Harvard University (United States) [presenting]
Shaun Seaman - University of Cambridge (United Kingdom)
Abstract: Three methods are proposed to model cohort data where repeated outcomes may be missing due to death and non-ignorable dropout. Examples arise in HIV studies where CD4 cell counts may be missing among those who are diagnosed with AIDS, or survey studies where some cognitive function outcomes are missing among the elderly. When missing data arise from death and dropout, one may want to distinguish between reasons for missingness to avoid making inferences about a cohort where no one can die. Instead, inferences based on those who are alive at any point in time might be more informative for health policy-makers. In order to obtain valid inference on those who are alive at each time point, we state the assumptions about the non-ignorable missingness process and we put forward: 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 for 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 doubly robust against model misspecification. Through simulation, we compare the bias, efficiency and coverage probability of the three methods, and apply them to a cohort of elderly adults from the Health and Retirement Study.