Evaluating Conformal Reliability of Pathway-Level Transcriptomic Signatures Under Cross-Cohort Shift in Sepsis Mortality Prediction
The study tested gene-level, pathway-level, and hybrid transcriptomic models for sepsis mortality prediction across different hospitals using cross-cohort validation. It found that while gene-level models offered better raw accuracy, pathway-level signatures provided greater robustness and more reliable uncertainty estimates when deployed in new settings. The research introduces a Pathway Stability Index to quantify these transferable biological signals.