arXiv:2604. 23904v3 Announce Type: replace-cross Abstract: Synthetic tabular data are often evaluated by distributional similarity, privacy distance, or train-on-synthetic-test-on-real predictive performance, but these criteria do not ensure validity for causal inference.
By Yichen Xu
ReClaim is a family of generative foundation models trained on 43.8 billion administrative claims from over 200 million enrollees in the MarketScan Research Databases. The models jointly learn diagnoses, procedures, medications, and expenditures, and outperform disease‑specific gradient‑boosted models and transformer baselines in predicting more than 1,000 diseases, forecasting next‑year expenditure, and identifying high‑need, high‑cost enrollees. ReClaim’s representations also improve effect estimates in real‑world evidence studies compared with algorithm‑derived covariates, demonstrating the value of nationwide insurance claims as a scalable substrate for healthcare foundation models.
By Fan Ma, Yuntian Liu, Xiang Lan, Weipeng Zhou, Jun Ni, Mauro Giuffr\`e, Lingfei Qian, Xueqing Peng, Yujia Zhou, Ruey-Ling Weng, Huan He, Lu Li, Huiyuan Wang, Qingyu Chen, Andrew Loza, Laila Rasmy, Degui Zhi, Yuan Lu, Chenjie Zeng, Joshua C Denny, Lee Schwamm, Daniella Meeker, Brian Ondov, Lucila Ohno-Machado, Yong Chen, Hua Xu
The paper introduces Patient, Place, Prior (P³), an audit framework that evaluates whether a longitudinal medical imaging forecast benefits from a patient’s own imaging history (Patient), from patient‑matched spatial support (Place), and from predictive value beyond a population‑average model (Prior). It applies this framework to Cancer JEPA, a one‑step model that predicts future breast dynamic contrast‑enhanced MRI scans during neoadjuvant therapy, incorporating a lesion‑constrained neural correction trained with an occlusion‑based latent objective. Validation shows lower forecast error when the neural correction uses the patient’s own history and matched lesion maps, though the 95% interval comparing patient‑specific to population‑average corrections includes zero, illustrating that P³ distinguishes input usage from true patient‑specific predictive advantage.
By Xingrui Gu, Hanxue Gu, Yuxiang Zhang, Yang Yang
arXiv:2610.02957v1 Announce Type: new
Abstract: World models learn state evolution from trajectories, making access to temporal context a central training requirement. Federated learning can use dist...
By Yipan Wei, Zhaokun Yan, Ziming Hong, Jiaqi Wu, Lixu Wang
arXiv:2608. 19501v1 Announce Type: cross Abstract: Diagnostic medical tests and devices provide useful information for evaluating the potential benefits and risks of therapeutic treatments.
By Wenxin Zhang, Rachael Phillips, Mark van der Laan
PRISM (Population‑Retrieved Inference for Structured Multimodal Prognosis) is a multimodal language model that integrates structured measurements, physiological waveforms, longitudinal history, and population context for clinical prognosis. It uses per‑feature numerical tokenization, recency‑weighted aggregation of ECGs, and a dual‑modality population memory bank, retrieving mean‑pooled embeddings of similar encounters without including individual records in the prompt. On the 1,443‑task MDS‑ED benchmark, PRISM achieves high AUROCs across diagnosis, deterioration, ICU admission, and mortality, outperforming baselines and demonstrating the value of structured multimodal integration and supervised external adaptation.
By Hamza Shafiq, Jialu Tang, Ting Dang, Jun Hu, Aaqib Saeed