IPDfromKM: reconstruct individual patient data from published Kaplan-Meier survival curves

Abstract

Background When applying secondary analysis on published survival data, it is critical to obtain each patient’s raw data, because the individual patient data (IPD) approach has been considered as the gold standard of data analysis. However, researchers often lack access to IPD. We aim to propose a straightforward and robust approach to obtain IPD from published survival curves with a user-friendly software platform. Results Improving upon existing methods, we propose an easy-to-use, two-stage approach to reconstruct IPD from published Kaplan-Meier (K-M) curves. Stage 1 extracts raw data coordinates and Stage 2 reconstructs IPD using the proposed method. To facilitate the use of the proposed method, we developed the R package IPDfromKM and an accompanying web-based Shiny application. Both the R package and Shiny application have an “all-in-one” feature such that users can use them to extract raw data coordinates from published K-M curves, reconstruct IPD from the extracted data coordinates, visualize the reconstructed IPD, assess the accuracy of the reconstruction, and perform secondary analysis on the basis of the reconstructed IPD. We illustrate the use of the R package and the Shiny application with K-M curves from published studies. Extensive simulations and real-world data applications demonstrate that the proposed method has high accuracy and great reliability in estimating the number of events, number of patients at risk, survival probabilities, median survival times, and hazard ratios. Conclusions IPDfromKM has great flexibility and accuracy to reconstruct IPD from published K-M curves with different shapes. We believe that the R package and the Shiny application will greatly facilitate the potential use of quality IPD and advance the use of secondary data to facilitate informed decision making in medical research.

Synthetizing Published Evidence on Survival by Reconstruction of Patient-Level Data and Generation of a Multi-Trial Kaplan-Meier Curve

“…In this context, the marked improvement in the performance of techniques that reconstruct individual-patient data [ 3 ] represents an important innovation, the role of which still needs to be fully evaluated. On the one hand, reconstructing individual-patient data is a mandatory prerequisite to determine the RMST, and this explains the increased use of these reconstruction techniques when a single trial has been analyzed through the RMST [ 2 ].…”

“…The various parameters mentioned above (especially HR, RMST, and median) have been investigated for many years to identify their respective advantages and disadvantages, and the literature on this issue is wide [ 2 ]. In contrast, the literature on the use of reconstructed survival curves is still in its early stages [ 3 ], and this applies particularly when multiple trials are analyzed and pooled together.…”

“…The survival graphs were obtained from Figure 1 for KEYNOTE-024 (154 vs. 151 patients; follow-up of 66 months; 226 deaths) and for KEYNOTE-042 (299 vs.300 patients; follow-up of 40 months; 356 deaths). For each Kaplan-Meier curve, the graph was digitalized and converted into x-y data pairs using Webplotdigitizer (Version 4.5, URL: https://apps.automeris.io/wpd /); then, the Shiny package (Version: 1.2.2.0; subprogram “Reconstruct Individual Patient Data”; URL: https://www.trialdesign.org/one-page-shell.html #IPDfromKM, see Reference [ 3 ]) was used to reconstruct patient-level data on the basis of the x-y data pairs, the total number of enrolled patients, and the total number of events. Finally, the pooled survival curves were generated from the reconstructed patient-level data along with the relevant Cox statistics; for this purpose, we used three packages (“coxph”, “survfit”, and “ggsurvplot”) under the R-platform.…”

“…One advantage is that the availability of these techniques permits the combination of multiple survival curves published in different trials without using any meta-analytical statistics. As an example of this approach, we applied the Shiny technique of individual patient data reconstruction [3] to the Kaplan-Meier graphs of overall survival reported in the KEYNOTE-024 [4] and KEYNOTE-042 [5] trials. Both trials were phase-III and were conducted in patients with locally advanced/advanced non-small cell lung cancer selected according to their PD-L1 expression status, not previously treated for their metastatic disease, and receiving first-line PD-(L)1 monotherapy.…”

“…In recent times, techniques that reconstruct individual patient data from the graphs of Kaplan-Meier curves have considerably improved in terms of performance and easy applicability [ 3 ]. One advantage is that the availability of these techniques permits the combination of multiple survival curves published in different trials without using any meta-analytical statistics.…”