Foundation Models to Unlock Real-World Evidence from Nationwide Medical Claims
Foundation Models to Unlock Real-World Evidence from Nationwide Medical Claims
Authors: Fan Ma, Yuntian Liu, Xiang Lan, Weipeng Zhou, Jun Ni, Mauro Giuffrè, Lingfei Qian, Xueqing Peng, Yujia Zhou, Ruey-Ling Weng, Huan He, Lu Li, Qingyu Chen, Andrew Loza, Laila Rasmy, Degui Zhi, Yuan Lu, Chenjie Zeng, Joshua C. Denny, Lee Schwamm, Daniella Meeker, Lucila Ohno-Machado, Yong Chen, Hua Xu Date: 2026-05-04 Paper ID: openalex:2605.02740
Summary
ReClaim is a generative transformer foundation model trained on massive-scale longitudinal administrative claims data. It effectively learns complex representations of patient trajectories, encompassing diagnoses, procedures, and medications. Empirical evaluation shows substantial performance gains over existing baselines across 1,000+ disease-onset prediction tasks and enhanced utility in healthcare expenditure forecasting and trial emulation. The model demonstrates strong generalizability and scaling laws across diverse medical event types.
Key Contributions
- Introduced ReClaim, a generative transformer trained on 43.8 billion medical events from 200 million patients in the MarketScan database.
- Achieved a mean AUC of 75.6% across 1,000+ disease-onset prediction tasks, significantly outperforming LightGBM (66.3%) and Delphi (69.4%).
- Demonstrated superior utility in downstream clinical tasks, including improved healthcare expenditure forecasting and a 72% reduction in systematic bias for target trial emulation.
Open Questions & Future Work
Key Concepts
- reclaim: A generative transformer foundation model trained on large-scale administrative claims data for longitudinal healthcare trajectory modeling.
Archivist Review
ReClaim is approved as a key healthcare foundation model architecture specifically targeting longitudinal administrative claims, which is a domain-specific paradigm deserving recognition. MarketScan is approved as a central dataset in this field. The open question regarding clinical phenotype definitions is approved as it addresses a fundamental limitation in benchmarking high-stakes clinical models. The multimodal integration question was rejected as being a generic ‘more data’ future work direction common to many model papers.
Approved Concepts
- ReClaim: It is a foundational model specifically architecture for administrative claims data, demonstrating state-of-the-art performance in disease prediction and healthcare expenditure forecasting.
Approved Open Questions
- Clinical Phenotype Definitions for Foundation Models: Using suboptimal disease definitions limits the clinical validity and precision of disease-onset predictions, potentially leading to representations that capture coding patterns rather than true pathophysiology.
Datasets
Links
Metadata & Links
- url
- https://arxiv.org/abs/2605.02740
- paper_id
- 2605.02740
- paper_source
- openalex
- domain
- nlp
- tags
-
- architectures
-
- datasets
- MarketScan
- concept_slugs
- reclaim
- dataset_slugs
- marketscan
- skill
- TimeSeriesSkill
- created_at
- 2026-05-07T06:04:37Z