HealthFound: a health world model for quantitative reasoning on longitudinal health profiles
- 1Fudan University
Abstract
Medical large language models (LLMs) have shown promise in medical knowledge retrieval and clinical reasoning. However, their capacity for quantitative reasoning over dense, longitudinal health data remains limited, largely due to the lack of effective training strategies that leverage large-scale, natural-language-based prospective cohorts. To address this, we present HealthFound, a medical LLM trained on 12.4 million self-constructed examples. These include 15-year longitudinal biomedical records from 501,936 participants, as well as datasets for medical knowledge and general reasoning. We propose a novel three-stage training framework comprising progressive curriculum supervised fine-tuning, rejection-sampled retention fine-tuning, and task-verifiable policy optimization. This enables the model to effectively parse heterogeneous quantitative follow-up data and perform robust quantitative reasoning. HealthFound achieves state-of-the-art performance across multiple newly constructed benchmarks. It attains the highest accuracy on eight public medical benchmarks for general medical competence. On 1,162 UK Biobank tasks predicting future disease onset from biomedical measurements, it outperforms the second-best model by over 10 percentage points on average. In two independent external validations (MIMIC-IV and NHANES), it demonstrates superior zero-shot generalization, achieving improvements of over 6% across 484 disease endpoints. The model also achieves top results on novel tasks for literature-derived question answering and quantitative indicator imputation. Furthermore, HealthFound generates interpretable reasoning rationales that are consistent with established biomedical knowledge. Collectively, our results demonstrate that HealthFound enables medical LLMs to evolve from static knowledge repositories into dynamic quantitative reasoning engines over longitudinal health data.
Related
- Skill-guided agentic reasoning enables longitudinal synthesis of myeloma records at expert-level concordance.
- Longitudinal quantitative MRI of blood-brain barrier leakage in chronic cerebral hypoperfusion rat model
- Lipid and immune profiles in children with systemic juvenile idiopathic arthritis: a case-control study with paired longitudinal comparison of disease activity
- Baseline adipokine profiles have limited predictive value for weight-loss outcomes after laparoscopic sleeve gastrectomy: a prospective longitudinal cohort study
- Longitudinal quantitative streamline tractography: Robust estimation of white matter connectivity differences.
BibTeX
@misc{guo2026healthfound,
title = {{HealthFound: a health world model for quantitative reasoning on longitudinal health profiles}},
author = {H. Guo and L. Song and P. Ren and J. Han and Y. Wang and H. Li and J. Deng and J. Wang and W. Gong and J. Feng and W. Cheng},
howpublished = {medRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.03.26364142},
url = {https://www.medrxiv.org/content/10.64898/2026.10.03.26364142v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - HealthFound: a health world model for quantitative reasoning on longitudinal health profiles
AU - H. Guo
AU - L. Song
AU - P. Ren
AU - J. Han
AU - Y. Wang
AU - H. Li
AU - J. Deng
AU - J. Wang
AU - W. Gong
AU - J. Feng
AU - W. Cheng
PY - 2026
DA - 2026/10/03
PB - medRxiv
DO - 10.64898/2026.10.03.26364142
UR - https://www.medrxiv.org/content/10.64898/2026.10.03.26364142v1
AB - Medical large language models (LLMs) have shown promise in medical knowledge retrieval and clinical reasoning. However, their capacity for quantitative reasoning over dense, longitudinal health data remains limited, largely due to the lack of effective training strategies that leverage large-scale, natural-language-based prospective cohorts. To address this, we present HealthFound, a medical LLM trained on 12.4 million self-constructed examples. These include 15-year longitudinal biomedical records from 501,936 participants, as well as datasets for medical knowledge and general reasoning. We propose a novel three-stage training framework comprising progressive curriculum supervised fine-tuning, rejection-sampled retention fine-tuning, and task-verifiable policy optimization. This enables the model to effectively parse heterogeneous quantitative follow-up data and perform robust quantitative reasoning. HealthFound achieves state-of-the-art performance across multiple newly constructed benchmarks. It attains the highest accuracy on eight public medical benchmarks for general medical competence. On 1,162 UK Biobank tasks predicting future disease onset from biomedical measurements, it outperforms the second-best model by over 10 percentage points on average. In two independent external validations (MIMIC-IV and NHANES), it demonstrates superior zero-shot generalization, achieving improvements of over 6% across 484 disease endpoints. The model also achieves top results on novel tasks for literature-derived question answering and quantitative indicator imputation. Furthermore, HealthFound generates interpretable reasoning rationales that are consistent with established biomedical knowledge. Collectively, our results demonstrate that HealthFound enables medical LLMs to evolve from static knowledge repositories into dynamic quantitative reasoning engines over longitudinal health data.
ER -