Category: Commercialization
Long diagnostic odysseys leave rare disease patients undiagnosed and invisible to the therapies that could help.
Rare diseases often take years to diagnose because symptoms are nonspecific and prevalence is low, leaving eligible patients undiagnosed and hard to see in available data. AI applied to de-identified claims and EHR data learns patterns from confirmed cases to surface patients with similar profiles for clinical review — support for awareness and evaluation, never a diagnosis.
Rare diseases are hard to spot because their signs are nonspecific, prevalence is low, and few clinicians see many cases — so patients often move through years of tests and referrals before a diagnosis. AI applied to large, de-identified claims and EHR datasets helps by learning the patterns that tend to characterize confirmed patients and then finding others whose records look similar, producing a prioritized list of people who may warrant a closer clinical look or targeted awareness efforts. These outputs are probabilistic and support awareness and evaluation; they are not diagnoses. A model can indicate that a profile resembles known cases, but confirming a rare disease requires a clinician's assessment and often specialist testing. Because the data is sensitive and the populations small, findings are best used to inform education and clinical review under applicable privacy rules — never to label an individual or bypass the diagnostic process.
Data coverage is critical for rare conditions: ask how well the underlying sources capture the population you care about, since small case numbers make patterns harder to learn and validate. Probe how the model was built and tested given those limits, how it handles false positives, and whether it can explain why a profile was flagged rather than returning an unexplained score. Because this work touches sensitive data and vulnerable patients, examine de-identification, privacy safeguards, and compliance, and be explicit about how outputs will be used — for awareness and clinical review rather than direct targeting of individuals. Confirm the findings are specific enough to act on responsibly and that they integrate with the clinical or medical workflows meant to follow up.
A careful entry point is to validate the model against a cohort of confirmed patients in the disease area, checking how many of them it would have surfaced and how much noise it produces, and whether the patterns it relies on are clinically sensible. Running this evaluation before any output informs outreach or education — and keeping it scoped to one condition — keeps both the assessment and the privacy review manageable.
Rare diseases often present with nonspecific, overlapping symptoms, and because each condition is uncommon, many clinicians rarely encounter them. Patients can spend years cycling through tests and referrals — the so-called diagnostic odyssey — which leaves eligible patients undiagnosed and difficult to identify in available data.
AI learns the patterns that tend to characterize confirmed patients — sequences of symptoms, tests, and referrals — from de-identified claims and EHR data, then finds other patients whose records resemble those patterns. The result is a prioritized list of candidates that can inform clinical review and awareness efforts.
No. The output flags patients whose profiles resemble known cases, which is a signal for further evaluation, not a diagnosis. Confirming a rare disease requires a clinician's assessment and often specialist testing, and the findings should support awareness and review — handled under applicable privacy rules — rather than label any individual.
Ask how well the data covers the population, how the model was built and validated given small case numbers, how it handles false positives, and whether it can explain why a profile was flagged. Then examine de-identification, privacy, and compliance, how outputs are intended to be used, and how findings integrate with the clinical or medical workflows meant to follow up.
Intercept Pharmaceuticals worked with Komodo Health to apply its Healthcare Map real-world data to primary biliary cholangitis, supporting patient-journey and treatment-pattern analyses for its Ocaliva franchise.
Vendor: Komodo Health · press release (Komodo Health (Business Wire)) · 2022-11-10 — Partner: Intercept Pharmaceuticals