Estimating Survival Times Using Swiss Hospital Data

We compare and evaluate two different approaches to estimate overall survival curvesfrom censored data of recurrent events: (1) standard survival time analysis, and (2) a multistate framework that explicitly estimates the mortality rate during censored periods. With both models, we estimate disease-specific survival curves with data from the Swiss Federal Statistical Office's medical statistics on hospitals (MedStat). Using cancer registry data as a benchmark for overall survival, we find that the accuracy of survival time estimates based on the multistate model are not superior to the simpler single-risk model. Although the computationally demanding multistate model is less accurate in predicting survival times, it may nevertheless be useful if intermediate transitions are the targeted issues.