Greenwood's survival variance estimator revisited

Featuring Paul Talsma, Principal Statistician at PHASTAR

Abstract: 'Greenwood's survival variance estimator revisited'.

Borkowf (2005) introduced two new variance estimators for the Kaplan-Meier survival function. He compared these with the well-known Greenwood and Peto estimators, using simulated and real-life data. Contrary to conventional wisdom (Collett, 2003), he found that both the Greenwood and Peto estimators tend to underestimate the variance in the extreme parts of the distribution, especially for small sample sizes. He claims that his estimators provide a better match with simulated variance estimates. In spite of this, his estimators have not been used very much and have not been incorporated in SAS PROC LIFETES, probably because of their complexity. Instead we propose to use the Greenwood variance estimator, which is currently the most used and readily available in current software, and correct its bias with a straightforward modelling approach. It is shown that this approach offers a very close match with simulated variance estimates. Examples are provided, as is SAS code to perform the modelling.

About the speaker:

Paul Talsma joined PHASTAR as principal statistician in 2019. He has a PhD from the university of Nijmegen, the Netherlands, and worked as a university teacher methodology and statistics for 3.5 years, before joining the pharmaceutical industry in various functions, both CRO and pharma. He is a registered biostatistician in the Netherlands.

Interested in statistical services? Click on the button to read about our expertise

We are experts in study design, statistical analysis, data science, data capture and reporting for clinical trials