← allmeta

Kaplan-Meier IPD Reconstructor

Reconstruct approximate individual patient data (event times + censoring indicators) from a digitized K-M curve plus at-risk table. Implements a simplified Guyot-style heuristic — events are placed at curve-point midpoints, not the full iterative product-limit reconciliation of the original Guyot et al. BMC Med Res Methodol 2012;12:9. Always verify total events / median / log-rank p match the published trial; never claim IPD-level accuracy from reconstructed data.

Input

Reconstruction

Reconstructed N0
Total events0
Median survival—

K-M overlay: digitized vs reconstructed

How it works: Between each pair of adjacent digitized points this tool estimates events from the survival drop (d ≈ n·(1 − Sk/Sk−1)) and assigns the residual decrease in the at-risk count to censoring, so the reconstructed K-M tracks the digitized curve and the at-risk totals approximately follow the table. This is a simplified heuristic, not Guyot et al.'s full iterative product-limit reconciliation. The output is one row per subject (event time + event/censor indicator); always verify total events / median / log-rank p against the published trial.