Recent offline reinforcement learning (RL) studies report policies that outperform physician decisions on clinical outcomes. We conduct a systematic, partially crossed evaluation of five offline RL algorithm families and 14 reward designs in 44,894 post-2018 acute ischemic stroke patients from a nationwide registry (N = 129,033). Standard Fitted Q-Evaluation (FQE) yields an apparent policy-improvement estimate of +0.0069; adding an Early Neurological Deterioration penalty increases it to +0.0101. We identify reward-embedded confounding, in which a proxy terminal reward encodes baseline severity and prognosis as well as treatment efficacy. A 2 x 2 factorial analysis finds that terminal reward confounding accounts for 218.6% of the observed signal change, so its removal overshoots the null. After DML-inspired GBM reward residualization, the FQE estimate attenuates to +0.0033 (p = 0.132), and full deconfounding yields +0.0025 (p = 0.291). FQE-based diagnostics, T-learner analyses, and direct recurrence analyses converge away from a clinically meaningful aggregate improvement. A 1-year mRS factorial analysis replicates the attenuation. We provide an empirically motivated six-step evaluation checklist. NIHSS-stratified heterogeneity is hypothesis-generating for prospective trial design; hospital-level disagreement does not persist after full reward deconfounding.
Jiacheng Zheng, Chang Guo, Zixuan Wang +2cs.LG q-bio.QM
Molecular property models are commonly evaluated by holding out Bemis-Murcko scaffolds, yet a scaffold identifier is only one notion of chemical unfamiliarity. We introduce a label-free structural-frontier split that reserves the sparsest and most physicochemically remote scaffold groups, and evaluate it on six public experimental or curated ADMET tasks. Against a 70/10/20 scaffold control with identical acyclic grouping, the frontier inflates equally weighted primary error with a taskwise median of 87.0% and a skew-sensitive mean of 130.3% (descriptive task/seed bootstrap interval, 52.1-246.0%). The mean falls to 75.9% once BBB is removed; that endpoint is the one whose score ranking inverts at the frontier. A message-passing graph-network control still shows a large gap (mean 82.8% over four tasks) and does not invert, so a low-capacity head does not explain the effect. We also test Multi-View Frontier Risk Extrapolation (MV-FREX), a count-adjusted tail-risk penalty over four molecular views, and treat it as a falsifiable probe. It changes normalized frontier error by only 0.16% relative to empirical risk minimization for the perceptron head (interval, -0.43-0.84%) and by -1.9% for the graph network; three fixed robust-penalty controls are likewise inconclusive. Against the published Lo-Hi and DataSAIL splitters, the frontier inflates error more on average, though no split is uniformly hardest. An audit of 31,561 marine natural products further shows that OOD status and agreement with legacy ADMET predictions depend on the molecular view, endpoint, and teacher coverage. Split construction and label provenance are important evaluation constraints in their own right, and the tested training penalties do not resolve the frontier failures we observe.
Self-supervised learning (SSL) is increasingly used in medical imaging to reduce annotation requirements, but representation quality is often judged using a single downstream evaluation setting. For dense clinical tasks, this can confound representation quality with the capacity of the downstream model used to recover task-relevant information. We present a systematic evaluation of self-supervised representations for left-ventricular segmentation and ejection fraction (EF) estimation from apical four-chamber echocardiography on EchoNet-Dynamic. Rather than relying on a single downstream probe, we compare a hierarchy of extraction strategies with increasing expressivity: heuristic extraction without mask-supervised training, frozen linear probes, frozen lightweight decoder probes, and partial fine-tuning. We apply this framework to two complementary representation families: generic frozen self-DIstillation with NO labels (DINOv3) features and a task-adapted dense self-supervised representation, Bootstrap Your Own Segmentation (BYOS). In both families, heuristic extraction substantially understated what was recoverable from the frozen representation. For DINOv3, performance improved from Dice 0.684 and EF mean absolute error (MAE) 13.01 under heuristic extraction to Dice 0.906 and EF MAE 9.65 with a frozen lightweight decoder, approaching a supervised U-Net baseline (Dice 0.915, EF MAE 9.72). For BYOS, performance improved from Dice 0.687 and EF MAE 17.83 under heuristic extraction to Dice 0.902 and EF MAE 8.74 with a frozen lightweight decoder. These results show that conclusions about self-supervised representation quality in dense echocardiographic analysis depend strongly on the downstream extraction strategy used for evaluation. We therefore argue that multi-strategy evaluation is an important methodological consideration for SSL in dense medical image analysis.