Smith_1996_J.Biopharm.Stat_6_395

Survival analysis techniques using Cox proportional hazards regressions with time-dependent covariates, life table survival plots, and Kaplan-Meier estimates were used to evaluate the effect of long-term tacrine hydrochloride (Cognex) treatment on nursing home placement (NHP) and mortality in patients with Alzheimer's disease. Patients with probable Alzheimer's disease (NINCDS criteria) who were randomized in a 30-week double-blind, placebo-controlled, parallel-group, high-dose study of tacrine (1) were subsequently allowed to receive long-term, open-label treatment during which they could receive doses up to 160 mg/day. Using last tacrine dose, the analyses demonstrated a dose-response relationship where patients on higher tacrine doses were less likely to enter a nursing home or die than patients on lower doses. The Cox proportional hazards regression approach with time-dependent covariates is also compared to logistic regression, which looks only at the crude proportions of patients having the event. Since logistic regression does not allow for the use of time-dependent covariates, it provides somewhat less conservative estimates of the magnitude of the treatment effect.