When Prognostic Compatibility Does Not Guarantee Transfer Utility in Event-Limited Cross-Species Survival Modeling

Abstract

Cross-species molecular transfer may improve prognostic modeling in rare cancers, but under severe target-event scarcity the same data can easily be used to fit, choose, and evaluate adaptation, increasing the risk of negative transfer and model-selection bias. We evaluated canine-to-human osteosarcoma survival transfer using a chronologically frozen design. DOG2 served as the canine source domain, 50 MSigDB Hallmark modules defined the shared representation, and a known-truth benchmark comprised 180 scenarios and 21,600 replicates spanning 5 - 40 target events and multiple transport regimes. A frozen safety rule compared low-rank and module-selective adaptation; later diagnostics and a 6,000-replicate disjoint-seed mechanism experiment could not alter that decision. Neither selectable architecture met the negative-transfer limit (0.215 and 0.486 versus 0.10), although oracle analyses showed that the threshold was attainable. Prespecified post-HOLD weighting and threshold sensitivities showed that benchmark composition amplified the magnitude of A3 failure but did not create its instability, whereas A2 was comparatively insensitive to weighting. Training-set prognostic compatibility identified generator regimes (AUROC 0.980) but did not reliably order transfer benefit versus harm. Prediction-time gate hardening improved discrimination and reduced negative transfer, although its composite criterion was not met. Outcome-blind DOG2 - TARGET analysis showed heterogeneous Hallmark preservation. In 86 TARGET cases with 29 events, the frozen canine classical model had IBS 0.159 versus 0.166 for a fold-local no-covariate Kaplan - Meier reference, while the human interpretation remained unresolved. Measurable prognostic compatibility therefore did not guarantee safe architecture-level transfer utility.