Early identification of rare disease using deep phenotyping of electronic health records

Abstract

Patients with rare genetic disorders often wait years for a correct diagnosis, highlighting the urgent need for early and efficient triage for those at risk. Electronic health records (EHRs) are rich in useful information, yet the extent to which this data is sufficient to generate prediagnostic signals is uncertain. We trained machine learning models on billing codes and phenotypes extracted from clinical notes and applied these across eight conditions in a longitudinal cohort of roughly 3 million patient records from the Mayo Clinic, seeking early identification of conditions. The number of patients identified early varied by condition, ranging from 10% to 89% at 99% specificity. Median lead times exceeded one year prior to the first diagnostic billing code for most conditions, including Fabry disease (1,408 days), hereditary angioedema (2,755 days), neurofibromatosis type 1 (1,067 days), and hereditary hemorrhagic telangiectasia (2,320 days). We show that ML models trained on EHR data can prioritize patients for clinician review and confirmatory genetic testing and that the integration of diverse phenotypic data sources provides superior predictive value.

An uncertainty-aware multimodal deep learning framework for Alzheimer’s disease classification using MRI and electronic health records
Naif Almusallam, Ahmad Bacha, Maqsood Hayat +2
Scientific Reports5 Oct 2026Peer-reviewedCC BY-NC-ND 4.0Full textPsychiatry & mental healthCrossref
Prevalence of motor neuron disease and its clinical subtypes in the Valencian Community, Spain: a population-based cross-sectional study using linked electronic health records
P. Ros-Arlanzon, J. F. Vazquez-Costa, A. Merchan-Cabanillas +5
medRxiv7 Oct 2026PreprintCC BY-NC-NDFull textNeurologymedRxiv
Air temperature and thrombotic cardiovascular disease in England: case time series studies using whole-population electronic health records during 2020-2022
I. J. Walter, A. Sampri, T. Bolton +16
medRxiv2 Sep 2026PreprintCC BYFull textEpidemiology & public healthmedRxiv
Prediction of incident inflammatory bowel disease from routine electronic health records: development and external validation of a longitudinal machine learning model
G. Onal, A. Ayati, M.-Y. Chen +2
medRxiv12 Oct 2026PreprintCC BY-NC-NDFull textGeneral medicinemedRxiv
Electronic Health Record-Based Machine-Learning for Cardiac Arrhythmia Identification in Adults with High-Risk Chronic Disease: Model Development and Validation Study
M. Goel, S. Matlock, Y. Karasu Benyes +6
medRxiv30 Oct 2026PreprintCC BY-NC-NDFull textCardiovascular medicinemedRxiv

Cite

BibTeX RIS

BibTeX

@misc{tunstall2026early,
  title = {{Early identification of rare disease using deep phenotyping of electronic health records}},
  author = {T. Tunstall and P. C. Ng and L. Meyers and M.-R. Coldagelli and L. Yang and N. Pereira and M. Osundiji and K. Im and A. Kumar and M. Rabinowitz},
  howpublished = {medRxiv},
  year = {2026},
  month = sep,
  doi = {10.64898/2026.09.11.26362811},
  url = {https://www.medrxiv.org/content/10.64898/2026.09.11.26362811v1},
  note = {CC BY},
}

RIS

TY  - UNPB
TI  - Early identification of rare disease using deep phenotyping of electronic health records
AU  - T. Tunstall
AU  - P. C. Ng
AU  - L. Meyers
AU  - M.-R. Coldagelli
AU  - L. Yang
AU  - N. Pereira
AU  - M. Osundiji
AU  - K. Im
AU  - A. Kumar
AU  - M. Rabinowitz
PY  - 2026
DA  - 2026/09/11
PB  - medRxiv
DO  - 10.64898/2026.09.11.26362811
UR  - https://www.medrxiv.org/content/10.64898/2026.09.11.26362811v1
AB  - Patients with rare genetic disorders often wait years for a correct diagnosis, highlighting the urgent need for early and efficient triage for those at risk. Electronic health records (EHRs) are rich in useful information, yet the extent to which this data is sufficient to generate prediagnostic signals is uncertain. We trained machine learning models on billing codes and phenotypes extracted from clinical notes and applied these across eight conditions in a longitudinal cohort of roughly 3 million patient records from the Mayo Clinic, seeking early identification of conditions. The number of patients identified early varied by condition, ranging from 10% to 89% at 99% specificity. Median lead times exceeded one year prior to the first diagnostic billing code for most conditions, including Fabry disease (1,408 days), hereditary angioedema (2,755 days), neurofibromatosis type 1 (1,067 days), and hereditary hemorrhagic telangiectasia (2,320 days). We show that ML models trained on EHR data can prioritize patients for clinician review and confirmatory genetic testing and that the integration of diverse phenotypic data sources provides superior predictive value.
ER  -

Abstract

Related

Cite