Generative model of patient health states and pan-cancer risk stratification
- 1Harvard Medical School
Abstract
While large language models are powerful generators of new text, forecasting disease progression from longitudinal health histories remains a challenging problem. We introduce GenEHR, an autoregressive generative model trained on electronic health records (EHRs) from millions of patients in diverse health systems. The model learns from real-world multimodal patient histories, including diagnoses, medications, lab values, and procedures while explicitly representing irregular intervals between visits to forecast future clinical events. We combine the general-purpose patient representation learned during foundational training with parameter-efficient supervised adaptation for the task of risk stratification for 17 cancer types. In five large EHR cohorts, supervised adaptation substantially improved prediction performance of a first cancer diagnosis within a five year horizon window. Our retrospective results support the prospective evaluation of GenEHR-CancerRisk as a clinical decision-support tool for prioritizing high-risk patients for screening and prevention of aggressive cancer types, such as pancreatic and ovarian cancer.
Related
- Molecular stratification of SPOP-mutant prostate cancer demonstrates Wnt signaling and replication stress signatures in diverse patient populations
- Biomarker discovery and patient stratification in pancreatic cancer using incomplete multi-omics data
- Baseline Patient and Tumour Characteristics Associated with Mental Health in Men Undergoing Active Surveillance for Prostate Cancer
- Evolution of colorectal cancer screening technologies and health economic evaluation: optimization strategies based on an economic gradient and dynamic risk stratification
- Associations between health-related social needs and colorectal cancer screening among United States Hispanic/Latino and non-Hispanic white adults
BibTeX
@misc{khan2026generative,
title = {{Generative model of patient health states and pan-cancer risk stratification}},
author = {A. Khan and D. T. Forster and M. Harsh and C. Zheng and E. T. Warner and D. Ritter and A. Chang and Q. Wei and T. K. Sorensen and D. S. Marks and L. V. Sequist and J. J. Hadlock and N. R. Fillmore and C. Sander},
howpublished = {medRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.09.26362676},
url = {https://www.medrxiv.org/content/10.64898/2026.09.09.26362676v1},
note = {CC BY-NC},
}
RIS
TY - UNPB
TI - Generative model of patient health states and pan-cancer risk stratification
AU - A. Khan
AU - D. T. Forster
AU - M. Harsh
AU - C. Zheng
AU - E. T. Warner
AU - D. Ritter
AU - A. Chang
AU - Q. Wei
AU - T. K. Sorensen
AU - D. S. Marks
AU - L. V. Sequist
AU - J. J. Hadlock
AU - N. R. Fillmore
AU - C. Sander
PY - 2026
DA - 2026/09/09
PB - medRxiv
DO - 10.64898/2026.09.09.26362676
UR - https://www.medrxiv.org/content/10.64898/2026.09.09.26362676v1
AB - While large language models are powerful generators of new text, forecasting disease progression from longitudinal health histories remains a challenging problem. We introduce GenEHR, an autoregressive generative model trained on electronic health records (EHRs) from millions of patients in diverse health systems. The model learns from real-world multimodal patient histories, including diagnoses, medications, lab values, and procedures while explicitly representing irregular intervals between visits to forecast future clinical events. We combine the general-purpose patient representation learned during foundational training with parameter-efficient supervised adaptation for the task of risk stratification for 17 cancer types. In five large EHR cohorts, supervised adaptation substantially improved prediction performance of a first cancer diagnosis within a five year horizon window. Our retrospective results support the prospective evaluation of GenEHR-CancerRisk as a clinical decision-support tool for prioritizing high-risk patients for screening and prevention of aggressive cancer types, such as pancreatic and ovarian cancer.
ER -