Abstract:
|
Survival analysis is a very common methodology in biomedical and epidemiological research, where the interest is in understanding time to an event. Secondary analysis of survival studies for pooling or meta-analyses are often hamstrung by the unavailability of primary individual-level data. Often, stratified rather than overall results are published. This work starts with published survival analysis results like Kaplan-Meier curves and derivative summary survival estimates and re-creates cohorts that would conform to the reported survival experience. This involves leveraging and inverting the construction of the Kaplan-Meier curve to estimate the timing and number of events and censored observations along the timeline. Unlike other similar methods, our approach requires only the total size of the cohort at the beginning of the study (i.e., at time 0) and a digitized Kaplan-Meier curve. Testing our method on over 150 published Kaplan-Meier curves, we find that the Kaplan-Meier curves generated from the constructed cohorts accurately mimic the original curves with very small errors towards the right end of the curve when there is naturally more uncertainty.
|