A US National Institutes of Health-supported study has developed an AI algorithm trained on EHR data to predict rare disease, with plans to scale over time to suggest when disease may appear, and how patients will respond to treatment.
The WEakly Supervised Transformer (WEST) algorithm is reportedly capable of using “noisy”, incomplete, inaccurate, or non-informative data from EHRs to predict whether a patient is likely to have a specific rare condition.
“In particular, the model can learn from patients both with and without confirmed diagnoses, using less precise outcome information to identify diagnostic patterns,” NIH states. “This is particularly important when studying rare diseases, where knowledge may be limited and the high-quality labeled data typically needed for model training are often unavailable.”
The algorithm was initially tested using EHR data from patients at risk for two rare lung diseases: pulmonary hypertension and severe asthma, achieving the highest rated predictive performance among all baseline models in identifying those diagnosed by clinicians.
Research teams mapped the ways these two conditions related to other clinical features in EHRs and trained WEST on patient records containing those events. Zongxin Yang, a research fellow from Harvard Medical School, explains: “Our model’s performance becomes better and better, and eventually we achieve a performance that aligns closely with expert-reviewed diagnoses.”
Next steps for the algorithm will involve scaling it to analyse longitudinal patient data in order to predict when they may develop a disease or how they may respond to treatment, NIH continues, with hopes of supporting the early identification of rare diseases.
Citation: Greco, K.F., Yang, Z., Li, M. et al. A weakly supervised transformer for rare disease diagnosis and subphenotyping from EHRs with pulmonary case studies. npj Digit. Med. 9, 211 (2026). https://doi.org/10.1038/s41746-026-02406-x
Wider trend: AI in health and care
HTN was joined by a panel including Ciara Moore, EPR operations director at Bath, Salisbury and Great Western Group, Stuart Cooney, CTO at Royal Berkshire NHS Foundation Trust, and Julian Wiggins, healthcare solution director at Rackspace Technology, for a discussion focusing on cloud adoption, AI maturity, and cyber resilience. Panellists explored how healthcare organisations are tackling delivery, legacy systems, and rising digital expectations, and what this means for future strategy and plans. We also looked at the fragmented cloud landscape, integration pressures, legacy infrastructure, AI, and the growing urgency around cyber resilience, finishing by asking where NHS leaders should prioritise investment and focus in 2026.
A study exploring informed consent for ambient documentation using generative AI in outpatient care has highlighted nuances including that patients are more likely to self-censor when talking about mental and sexual health or illicit activity during consultations. The study, published in Jama Network Open, was conducted from March to December 2024 in ambulatory practices across specialities in a “large urban academic health centre”, involving 18 clinicians and 103 patients in an operational proof-of-concept.
The board of North London NHS Foundation Trust (NLFT) has shared details of work to improve efficiency, reduce running costs, and eliminate non-value adding tasks with the use of digital tools and AI. Mobilising AI solutions to free-up time to care through a ten percent reduction in time spent on non-clinical activities year-on-year is noted, and increasing the use of digital skills by upskilling the workforce to reduce time spent on non-value adding tasks by five percent year-on-year.
