Download E-books Directly parameterized regression conditioning on being alive: analysis of longitudinal data truncated by deaths PDF

By Kurland B.F., Heagerty P.J.

For observational longitudinal reports of geriatric populations, results equivalent to incapacity or cognitive functioning are usually censored via loss of life. Statistical research of such information may perhaps explicitly situation on both important prestige or survival time while summarizing the longitudinal reaction. for instance a patternmixture version characterizes the suggest reaction at time t conditional on loss of life at time S = s (for s > t), and hence makes use of destiny prestige as a predictor for the time t reaction. instead, we outline regression conditioning on being alive as a regression version that stipulations on survival prestige, instead of a selected survival time. Such types should be often called in part conditional because the suggest at time t is specific conditional on being alive (S > t), instead of utilizing finer stratification (S = s for s > t). We exhibit that naive use of ordinary likelihood-based longitudinal equipment and generalized estimating equations with non-independence weights could lead to biased estimation of the in part conditional suggest version. We strengthen a taxonomy for lodging of either dropout and dying, and describe estimation for binary longitudinal info that applies choice weights to estimating equations with independence operating correlation. Simulation stories and an research of per month incapacity prestige illustrate power bias in regression equipment that don't explicitly situation on survival.