The PAI project (patient pathway improved using artificial intelligence): study protocol of a novel tool of artificial intelligence for earlier detection of pseudoxanthoma elasticum and hereditary bradykinin-mediated angioedema, two rare conditions (Preprint)

BACKGROUND More than 300 million people worldwide are affected by a rare disease. The average time to diagnosis is estimated at 4.8 years, leading to substantial delays in appropriate medical care. Our project focuses on two rare diseases (pseudoxanthoma elasticum (PXE) and hereditary bradykinin mediated angioedema (HAE)) selected because they differ markedly in clinical presentation, disease course, and complications. OBJECTIVE We aim to enable earlier detection of patients affected by PXE or HAE by applying large language models (LLMs) to electronic health records (EHRs), in close collaboration with medical experts. METHODS This retrospective data based study involves developing disease specific models using EHR data from patients admitted to five university hospitals in western France (Angers, Brest, Nantes, Rennes, and Tours) between 2010 and 2024. Models will first be fine tuned on data from Angers University Hospital, after which their performance will be evaluated using standard metrics and validated on data from the other hospitals. Throughout the study, diagnoses will be confirmed by expert physicians at reference centers. Once validated, the models will support the development of a detection tool capable of identifying patients potentially affected by PXE or HAE for real life testing and evaluation. RESULTS The scientific and ethical evaluations of the project have been completed and received a favorable opinion, and the regulatory and contractual procedures among the participating partners have been finalized. Data from the participating institutions will be uploaded to the West Data Hub platform (trusted research environment) starting in November 2026. Descriptive analysis and model development will begin thereafter. The first results are expected by mid-2027. CONCLUSIONS Expected outcomes include a better understanding of how clinical heterogeneity and disease trajectories influence artificial intelligence model development and performance, the creation of disease specific LLMs to support earlier detection of PXE or HAE, and a validated framework for adapting this approach to other rare diseases. Ultimately, the project aims to improve patient care pathways by enabling earlier identification and reducing diagnostic delays. CLINICALTRIAL The data based study was prospectively registered at the Health Data Hub (now Plateforme des Données de santé or PDS) (reference number: 24952390) on 2025 June 25th. Institutional scientific and ethical approval was first obtained from the CSE (Scientific and Ethics Committee) of the West Data Hub (identification number: ODH-INT-Pub-011, on 2025 05 15), followed by ethical approval granted by the CESREES (Ethics and Scientific Committee for Health Research, Studies, and Evaluations) (identification number: 24952390, on 2025 10 02), and subsequently by authorization from the CNIL (“Commission nationale de l'informatique et des libertés”, French Data Protection Agency) (decision DR-2026-032, on 2026 02 09).