COIOSDeep risk insights through multi-disciplinary research across the globe
server room aisle with metal equipment racks
Item new Preprint · not peer-reviewed Added

Model trained to forecast patient records predicts five-year risk of a first cancer

A model trained to forecast what happens next in the records of millions of patients predicted a first cancer within five years, although no accuracy figures are given.

The model was trained on the full visit histories of millions of patients, including the irregular gaps between visits, to forecast what came next. Further training on patients whose cancer outcomes were known improved five-year prediction of a first cancer in five record sets, with no accuracy figures given. The authors present it as retrospective support for prospective testing as a way to prioritise screening for pancreatic and ovarian cancer.

Why it is interesting: A general model of patient histories, rather than a cancer-specific risk equation, tested against first cancer diagnosis.

Source
medRxiv, 19 September 2026
Paper
Identifying cohorts at elevated risk of cancers using generative modeling of patient health states
Authors
Khan A, Forster DT, Harsh M, Zheng C, Warner ET, Ritter D et al.
DOI
10.64898/2026.09.09.26362676
Type
Preprint
Design
Retrospective development and validation of a generative EHR foundation model (GenEHR) with supervised adaptation, five large EHR cohorts, millions of patients, five-year horizon
Verdict
New finding
Driver
AI in medicine

Other drivers in Data, models and systems