Key result
LLMs improve rare disease top-1 diagnostic accuracy up to ~33% over historical rates.
Why the study?
Rare cardiomyopathies cause prolonged diagnostic delays with frequent misdiagnoses, and the impact of large language models on diagnostic timelines, accuracy, and equity remains uncertain.
Do Large Language Models improve diagnostic accuracy and shorten the path to diagnosis in rare cardiomyopathies?
Do Large Language Models improve diagnostic accuracy and shorten the path to diagnosis in rare cardiomyopathies?
Large language models may improve diagnostic accuracy for rare cardiomyopathies from historical rates to up to 33.1%, potentially shortening diagnostic timelines, provided that biases and hallucinations are mitigated.
LLMs merit further investigation for rare disease diagnosis; extends preliminary data but leaves clinical utility and validation open.
Background: Rare cardiomyopathies, including hypertrophic, dilated, restrictive, and left ventricular noncompaction subtypes, often cause prolonged diagnostic delays of four to seven years or more, with frequent misdiagnoses due to overlapping symptoms, subtle early signs, and limited access to cardiology specialists. These delays can lead to progressive cardiac dysfunction, arrhythmias, thromboembolic events, and increased morbidity and mortality. Large language models (LLMs) are emerging AI tools that synthesize clinical narratives, integrating patient histories, labs, and imaging to generate differential diagnoses and suggest testing. Their impact on diagnostic timelines, accuracy, and equity remains uncertain. Objective: This narrative review evaluates LLM performance in rare disease and cardiomyopathy diagnosis, examines their potential to reduce diagnostic delays, explores inequity and bias risks—particularly in low- and middle-income countries—and identifies safeguards for safe, equitable implementation. Methods: A non-systematic PubMed and Google Scholar search (2010–2026) identified studies on LLMs in rare disease diagnosis, Undiagnosed Disease Network cardiology cohorts, and RareArena benchmarks. Narrative synthesis focused on diagnostic accuracy, model limitations, hallucinations, biases, and mitigation strategies, including human oversight and retrieval-augmented generation. Results: LLM top-1 diagnostic accuracy ranges from 10% to 33%, improving from historical rates of 5.6%–13.3%, with RareArena reporting 33.1% for GPT-4o. LLMs can generate clinically relevant differentials and support non-specialist clinicians in identifying genetic cardiomyopathies and testing strategies. Limitations include hallucinations, data inconsistencies, overreliance, and linguistic or demographic biases that may affect patients in low-resource settings. Effective implementation requires human supervision, bias mitigation, and retrieval-augmented verification. Conclusion: LLMs may shorten diagnostic timelines for rare cardiomyopathies by identifying subtle phenotypic patterns and suggesting early targeted testing. Safe integration requires hybrid approaches combining clinician expertise with AI, robust safeguards, and continuous outcome evaluation to ensure equitable, accurate, and effective patient care, potentially transforming the diagnostic journey while addressing healthcare disparities.
No takes yet. Share an insight, caveat, or question.
Talha et al. (2026) conducted a review in Rare cardiomyopathies. Large language models (LLMs) vs. Historical diagnostic rates was evaluated on Top-1 diagnostic accuracy. Large language models (LLMs) demonstrated top-1 diagnostic accuracy ranging from 10% to 33.1% for rare diseases, improving upon historical rates of 5.6% to 13.3%.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: