Can AI diagnose certain rare diseases from electronic health records ?
Cast your vote — then read what our editor and the AI models found.
Under controlled research conditions, artificial intelligence has shown it can spot subtle telltales of certain rare diseases buried in electronic health records. Yet broader deployment is still stalled by uneven accuracy across the full spectrum of rare disorders and lingering doubts over reliability in everyday practice.
Background
Over the past few years several groups have built transformer-based models that read longitudinal EHR sequences and flag patients whose symptom trajectories match curated rare-disease cohorts. In 2023 a system trained on more than 30,000 US patient records achieved a positive predictive value above 0.7 for four lysosomal storage disorders but fell below 0.5 for a rarer glycogenosis subtype, illustrating uneven performance across disorders. A multi-centre study published the same year compared two proprietary LLMs fine-tuned on anonymised records from specialist clinics and found they recovered 79 % of previously missed cases of Niemann-Pick type C while introducing one false positive per ten true positives. Workflows that combine structured billing codes with unstructured clinician notes have shown the biggest gains, yet they remain brittle when applied to centres whose documentation styles diverge from the training corpora. At least one large health-system rollout was paused after an audit revealed clinically significant drift when ICD-10 codes were updated, underscoring the maintenance burden of keeping rare-disease models current.
SOURCE: BMJ, 2024
Suggest a tag
A missing concept on this topic? Suggest it and admin reviews.
Status last checked on June 26, 2026.
Gallery
Can AI diagnose certain rare diseases from electronic health records?
Narrow demos exist — but the panel was not unanimous.
The jury found the AI capable of glimpsing the shadow of rare disease across a patient record, yet unable to name the shape with full certainty; it delivers timely clues but not unshakable diagnoses. Their lone “almost” vote reflected cautious praise for pilot studies that edge past paperwork while still lacking robust, cross-hospital validation. Ruling: A compass that points northward but may wobble in a crosswind.
But the data is real.
The Case File
Across 11 sessions, 36 jurors have heard this case. Combined tally: 6 YES · 27 ALMOST · 3 NO · 0 IN RESEARCH.
Note: cumulative includes older juror opinions. The current session tally above is the live verdict.
By a vote of 0 — 1 — 0, the panel returns a verdict of ALMOST, with verdict confidence of 80%. The court so orders.
"Specialized AI models achieve partial rare disease diagnosis accuracy in narrow clinical cohorts"
What the audience thinks
No 6% · Yes 91% · Maybe 3% 236 votesDiscussion
no comments⚖ 11 jury checks · most recent 2 days ago
Each row is a separate jury check. Jurors are AI models (identities kept neutral on purpose). Status reflects the cumulative tally across all checks — how the jury works.