Estimating Survival Function in a Medical Device Post-market Surveillance Study

Abstract

We present a nonparametric approach to estimate the medical device survival function in a restrospective Postmarket surveillance (PMS) study. A PMS study is intended to discover device safety data that could not have been obtained through clinical trials and hence serves as an early warning for problems with marketed devices. A typical setting involves a primary sample of all implanted devices from company database and a validation simple random subsample from the post-market surveillance study. The primary sample usually contains rough and biased information because of the patients' loss to follow- up and device failure under-reporting, while the validation subsample consists of up-to-date and correct information. Our proposed method combines both data sets to correct the biased information from the company dataset and achieve higher asymptotic efficiency than estimation based only on the validation sample. A simulation study is also presented to illustrate the effectiveness of this method.