Keywords: Survival analysis, digitization, individual patient data, IPD, meta-analysis
The meta-analysis of survival rates at time t requires a survival rate estimate at t as well as its standard error from each study included in the analysis. These metrics are often unreported in the published literature in favor of odds ratios, mean survival time, or median survival time, making it seemingly impossible to perform a comprehensive analysis of rates. However, most survival studies are accompanied by one or more Kaplan-Meier survival curves, which detail the survival behavior for the patients analyzed. We demonstrate the use of digitization software to regenerate published survival curves and a publicly available algorithm to estimate individual patient data (IPD) from these curves in order to derive patient survival rates to time t along with their standard errors. From the digitized curves and estimated IPD, a comprehensive meta-analysis of survival rates can be performed. Furthermore, the IPD enables patients across studies to be combined in an overall survival curve. Methods are illustrated using a real world, large-scale meta-analysis. We discuss lessons learned and the pros and cons of using the survival digitization method.