Visual extraction of individual patient data from published kaplan-meier survival plots using a web-based application offering different reconstruction methods (V-Exploder).

Abstract

Background: Individual patient-level survival data (IPD) is required for various types of medical research, such as meta-analyses and cost-effectiveness analyses. As IPD from clinical trials are usually not publicly available, secondary data utilization relies on IPD reconstruction from published Kaplan-Meier (KM) plots. In general, the established toolchains for IPD reconstruction are cumbersome and rarely offer possibilities to assess reconstruction accuracy.

Implementation: We have therefore developed V-Exploder, an academic, non-profit, data-local tool that integrates the complete workflow of IPD reconstruction into a single browser application ( www.v-exploder.com ). Implemented methods include the iterative algorithms described by Guyot et al. and Rogula et al., and a novel, direct method for the positioning of individual events and censorings. A key feature of V-Exploder is its immediate visual feedback: recreated KM graphs are continuously constructed and superimposed onto the original figure. This enables more precise IPD reconstruction than previously possible.

Methods: We conducted a study to evaluate the IPD reconstruction performance of V-Exploder. Ten investigators each reconstructed data from four of a total of twelve figures, each showing the survival of two randomly generated cohorts (20, 50, 100, or 200 subjects in each cohort). Metrics frequently used in survival analysis were calculated for the original and reconstructed data. Median absolute errors (AE), their interquartile ranges (IQR), and root square mean errors (RMSE) are reported.

Results: Survival metrics from reconstructed IPD showed high similarity to metrics calculated from the reference data across all datasets, investigators and used methods of IPD reconstruction. Regarding log HRs, median AE (IQR) & RMSE with the direct, Rogula, and Guyot method were 5.93 × 10^- 4 (7.61 × 10^- 5, 1.65 × 10^- 3) & 6.26 × 10^- 3, 5.28 × 10^- 4 (3.21 × 10^- 5, 2.59 × 10^- 3) & 4.05 × 10^- 3, and 4.17 × 10^- 3 (1.84 × 10^- 3, 1.96 × 10^- 2) & 2.38 × 10^- 2, respectively. IPD reconstruction validity and reliability were lower with the Guyot method compared to the other methods, especially when used on smaller datasets. However, errors were considered acceptable for most research applications regardless of the method used.

Conclusion: V-Exploder allows IPD reconstruction from published KM plots with high accuracy and reliability. Given its user-friendly visual interface, V-Exploder will facilitate secondary analysis of survival data and help improving analytic rigor.