AI's Role in Diagnosing Rare and Undiagnosed Diseases: A Critical Perspective
What the Researchers Tested
The NEJM Catalyst Insights Council has articulated a forward-looking perspective on the integration of artificial intelligence into the diagnostic pathway for rare and undiagnosed diseases. This involves leveraging AI's capacity for advanced pattern recognition, machine learning algorithms, and natural language processing to analyze diverse data types, including genomic sequencing data, electronic health records, imaging studies, and clinical notes. The core mechanism involves identifying subtle, often non-obvious, associations between patient symptoms, laboratory findings, and genetic markers that might escape conventional diagnostic approaches due to their rarity or atypical presentation. AI systems, particularly those incorporating deep learning, can be trained on vast, multimodal datasets to detect patterns indicative of specific rare diseases, even with incomplete or noisy input. For instance, an AI could cross-reference an unusual constellation of symptoms with known disease phenomaps, or analyze genomic variants for pathogenic significance. The Council's view is that this technological capability, while powerful, must be contextualized within a broader diagnostic ecosystem that supports data standardization, interoperability, and human expert oversight, rather than viewing AI as a standalone diagnostic tool.
What the Study Found
AI is identified as holding 'great promise' for accelerating the diagnosis of rare and undiagnosed diseases by the NEJM Catalyst Insights Council.
The collective burden of individually rare diseases is acknowledged, emphasizing that these conditions are 'not rare' in aggregate.
The Council advocates for AI as a critical component, but not the sole solution, within a comprehensive diagnostic strategy.
Effective AI integration requires foundational improvements in data infrastructure, clinical workflows, and multidisciplinary collaboration.
Why This Matters to Clinical Practice
For practicing clinicians, particularly those in primary care, pediatrics, and specialized fields grappling with complex cases, the diagnostic challenge of rare diseases is immense. Patients often endure years of misdiagnoses, specialist hopping, and invasive procedures before receiving a definitive answer. This 'diagnostic odyssey' not only places an enormous emotional and financial burden on patients and their families but also strains healthcare resources through repeated consultations and ineffective treatments. Integrating AI into the diagnostic process has the potential to significantly shorten this odyssey, leading to earlier intervention, improved patient outcomes, and a reduction in associated healthcare costs. By flagging potential rare disease candidates more rapidly, AI can guide clinicians toward appropriate specialist referrals, targeted genetic testing, and personalized therapeutic strategies, ultimately transforming the management paradigm for these historically challenging conditions. The ability to identify these patients faster could prevent irreversible organ damage, improve quality of life, and optimize resource allocation within health systems.
Clinical & Workflow Takeaways
Clinicians should anticipate an increasing role for AI-powered diagnostic support tools in identifying rare and undiagnosed conditions. These tools will likely integrate into existing electronic health record systems or serve as standalone platforms capable of analyzing complex patient data. Physicians should familiarize themselves with the capabilities and limitations of such AI systems, understanding that they function best as decision support aids rather than autonomous diagnostic entities. Collaboration with bioinformatics specialists and genetic counselors will become even more critical as AI facilitates the earlier identification of potential genetic or rare disease etiologies, necessitating expert interpretation and patient counseling. Hospitals and health systems should begin investing in robust, interoperable data infrastructures to maximize the utility of AI in this domain.
Limitations and What Remains Uncertain
While the promise of AI in rare disease diagnostics is substantial, several limitations persist. Current AI models are heavily reliant on the quality and completeness of training data; rare diseases inherently suffer from sparse and heterogeneous data, which can lead to biases and inaccuracies. Ethical considerations, including data privacy, algorithmic transparency, and the potential for perpetuating existing health disparities if training data is not diverse, remain paramount. Furthermore, regulatory frameworks for AI in rare disease diagnosis are still evolving, and real-world implementation will require rigorous validation beyond initial studies. The 'black box' nature of some advanced AI models can also challenge clinical interpretability and physician trust, necessitating robust explainable AI (XAI) approaches.
Explore the Work Behind Clinical AI
MedXchange Research Fellowship
Publish Clinical AI Research With Us
Our 10-week fellowship mentors physicians and researchers to build foundation models, analyze multimodal health datasets, and co-author peer-reviewed clinical studies.
Related Clinical AI Stories
LLM Chatbot Offers Promising, Yet Flawed, Allergen Avoidance Guidance in Contact Dermatitis
A recent quantitative evaluation study explored the utility of a Large Language Model (LLM) chatbot, specifically GPT-4, in providing allergen-free product recommendations for patients with contact dermatitis. The study assessed the LLM's ability to recommend products free of 30 common allergen groups found in international standard patch tests. While the chatbot demonstrated a high proportion of valid responses, particularly for cosmetics and medications, clinicians must recognize its limitations, especially concerning complex allergens prone to cross-reactivity. Over a period from September 2023 to January 2024, GPT-4 was prompted to suggest five allergen-free products for each allergen group, utilizing three distinct question formats. Dermatologists manually reviewed and categorized 2250 responses for validity, invalidity, cross-reactions, unknown status, and unresponsiveness. The findings indicate a significant potential for LLM tools to assist in allergen avoidance counseling, albeit with a crucial need for human verification and oversight.
Mitigating AI Hallucinations in LLMs: A Scoping Review for Clinical and Educational Integration
Large Language Models (LLMs) are increasingly integrated into medical education and healthcare, yet their utility is significantly hampered by 'AI hallucinations' â outputs that are plausible but factually incorrect, unsupported, or fabricated, including erroneous citations. These inaccuracies pose substantial risks to educational integrity, clinical decision-making, and patient safety. A comprehensive scoping review, adhering to Joanna Briggs Institute and Arksey and O'Malley methodologies, has systematically mapped the evidence on the nature, detection, mitigation, and governance implications of these hallucinations. The review identified 21 relevant sources from a pool of 2,051 records published between January 2018 and July 2026. This selection included five empirical studies evaluating LLM performance and 16 conceptual or review-based articles. The synthesis categorized findings into five major themes: conceptualization of hallucinations, detection and validation methods, mitigation strategies, educational implications, and governance frameworks. Key strategies to reduce hallucination risks include retrieval-augmented generation (RAG), structured prompt engineering, robust output verification, and essential human oversight, emphasizing that these are not just technical issues but multidimensional challenges requiring a comprehensive, multilayered approach for safe implementation.
Gene Therapy for Genetic Epilepsy: A Paradigm Shift from Symptomatic Control to Mechanistic Correction
Genetic epilepsies, often monogenic with well-defined molecular mechanisms, are poised for a significant therapeutic shift from seizure control to mechanism-driven correction via gene therapy. A recent review comprehensively analyzed five nucleic acid-based strategies: gene supplementation, endogenous gene expression enhancement, gene silencing, DNA editing, and RNA editing. The selection of an appropriate intervention is critically dependent on variant-specific functional consequences, differentiating between loss-of-function (LOF) and gain-of-function (GOF) mutations. This analytical framework aims to guide clinical decision-making by evaluating the mechanisms, suitability, cell-type specificity, delivery vectors, and safety profiles of each approach. The review integrated evidence from animal models, induced pluripotent stem cell (iPSC)-derived neurons, organoids, and ongoing clinical trials. It highlighted that antisense oligonucleotides (ASOs) and regulatory adeno-associated viral (AAV) approaches currently demonstrate the most advanced clinical validation, primarily due to mature delivery platforms and manageable safety profiles. While other strategies like DNA and RNA editing hold promise for durable disease modification, they require further research into delivery efficiency and long-term safety, underscoring the ongoing translational challenges in this rapidly evolving field.