What if we could learn more from old cancer trials than what was originally published? A new computational tool, called MD-JoPiGo, tries to do just that. It's designed to reconstruct the hidden, multi-dimensional stories of patients—like how different factors like age or tumor markers interacted—using only the one-dimensional survival curves that are commonly published in research papers.
The framework was tested on simulated data and two real-world groups of patients: 228 people with lung cancer and 929 with colon cancer. In these evaluations, the tool was able to accurately recover the complex, unobserved dynamics between different patient variables. It was also applied to data from a major immunotherapy trial (CheckMate 227) and reconstructed patterns of treatment efficacy that matched the known clinical truth.
However, the tool's success isn't guaranteed. Its ability to faithfully reconstruct the full patient profile depends heavily on the underlying 'causal topology'—the hidden web of cause-and-effect relationships between all the different factors. For simpler relationships, it works well. For more complex, interdependent structures, the tool needs extra hints about how those factors might be connected to avoid getting the wrong answer. This is a proof-of-concept showing a possible new way to mine historical data, not a finished product ready for doctors to use.