Computational framework reconstructs multidimensional clinical profiles from published Kaplan-Meier images

This methodological research developed and evaluated MD-JoPiGo, a computational framework designed to reconstruct multidimensional clinical profiles from published 1D Kaplan-Meier curves. The evaluation used simulated data and empirical cohorts including 228 lung cancer patients and 929 colon cancer patients. The framework's primary outcome was reconstruction fidelity, which was found to depend on the underlying causal topology of the data.

In evaluations, the framework accurately recovered unobserved multivariable dynamics in both simulated data and empirical cohorts. It also successfully reconstructed latent intersectional efficacy from CheckMate 227 trial reports, with results consistent with clinical ground truth. No effect sizes, absolute numbers, or statistical measures were reported for these reconstructions.

No safety or tolerability data were reported as this was a methodological study. Key limitations were not explicitly listed in the provided evidence. The practice relevance is that this approach enables secondary analysis of historical randomized controlled trials, potentially supporting individual patient data meta-analyses and synthetic trial emulations. However, clinicians should recognize that reconstruction fidelity varies based on causal structures, with parallel predictors resolved unconditionally while interdependent structures require minimal structural priors to resolve unidentifiability.