About Time: When Times Are Missing Management of Missing Data in Discrete-Time Survival Analyses

ABSTRACT Survival analysis is a method of analysis used to study event occurrence. Missing periods in discrete-time survival analyses are problematic, since whether an event occurs determines whether the subject is followed up upon. Seven strategies that can be used when missingness occurs (case deletion, deletion upon missing, single imputation, multiple imputation, remembrance, the Non-Event-Strategy and the Event-Strategy) are evaluated using four criteria: effect size bias, standard error bias, power and coverage rate of confidence intervals. Single imputation, multiple imputation and the Non-Event Strategy show good results. Single imputation performs slightly better, yet the Non-Event Strategy is easier to implement. Keywords Simulation study, discrete-time survival analysis, event occurrence, missing data, single imputation, multiple Imputation