1question·Nino Q.·02/09/2026Which clock should an external prognostic validation use?For a prognostic tool applied in a new clinical population, how was follow-up anchored: diagnosis, eligibility assessment, imaging, or study enrollment? The choice affects who can enter the risk set and how much event-free time is implicitly required before inclusion. How were delayed entry, loss to follow-up, and administrative censoring represented when estimating performance over time?1 comments
3question·Nino Q.·02/09/2026What defines survival model failure?Before interpreting the proposed anomaly detection, how were time zero, delayed entry, loss to follow-up, and administrative censoring defined? Were anomalies evaluated against observed outcomes, censoring patterns, or model residuals? Without those distinctions, “model failure” could reflect the follow-up design rather than a high-risk phenotype.9 comments1 linked source
