AI Tools Target Rare Disease Diagnosis as Global Burden Reaches $8.6 Trillion
New AI platforms are emerging to address rare disease diagnosis and epidemiology, as global costs reach up to $8.6 trillion annually affecting 300 million patients worldwide across more than 7,000 distinct conditions.
Rare diseases affect an estimated 300 million people worldwide and touch the lives of over one billion when families and caregivers are included, with global costs estimated at between $7.2 and $8.6 trillion each year. More than 7,000 distinct rare disorders exist, with about 80% having genetic origins, and around 70% of these conditions begin in childhood.
Researchers from China have developed an artificial intelligence tool called DeepRare designed to accelerate and improve the diagnosis of rare diseases, according to a study published Feb. 18 in the international journal Nature. These conditions are notoriously hard to diagnose due to their diverse symptoms, low prevalence, and limited clinician expertise, often leaving patients waiting for years — sometimes more than five — for an accurate identification.
DeepRare employs a large language model at its core, integrated into an agentic framework. Its architecture incorporates over 40 AI agents and tools that handle distinct tasks, including extracting symptoms from notes, matching symptoms with diseases, searching medical literature for similar cases, and analysing genetic variants from sequencing data. The tool processes inputs to generate a list of ranked diagnoses and provides transparent information backed by verifiable references from medical sources.
In testing on 6,401 cases, DeepRare demonstrated performance that was better than 15 existing diagnostic tools. When incorporating genetic data, it could accurately diagnose 69% of patients in one study cohort and surpassed popular tools like Exomiser (around 56%). In 163 challenging real-world cases, DeepRare correctly identified the disease on its first attempt in 64.4% of instances, compared to 54.6% for five experienced clinicians — each with over a decade in rare disease practice. Overall, it succeeded in about 79% of cases versus 66% for the experts.
DelveInsight has unveiled DelveEpiAI, an AI-powered epidemiology database covering 500+ disease indications with 10-year epidemiology forecasts across the seven major markets (US, EU4, UK, and Japan). The platform integrates patient population insights along with incidence/prevalence forecasts into a single interface, ranging from major oncology diseases to rare niche orphan conditions. The platform features advanced interactive dashboards that enable teams to visualize comprehensive 7MM epidemiology analysis, compare historical and forecasted patient populations, and explore detailed disease-specific segmentation.
A World Economic Forum white paper, "Making Rare Diseases Count: How Better Data Can Unlock a Multi‑Trillion‑Dollar Opportunity," calls for stronger data systems to make the burden of rare diseases visible and actionable. Its roadmap centres on five priorities: defining a minimum dataset across countries, strengthening patient registries, expanding screening and diagnostic capacity, enabling trusted data sharing, and using AI and digital tools to close evidence gaps.
This agenda builds on the pivotal 2025 World Health Assembly resolution on rare diseases, sponsored by Egypt and Spain and co‑sponsored by 39 other UN Member States, which includes a call to member states to support education for healthcare providers on rare diseases. Today, only 19% of physicians feel confident diagnosing rare diseases, and many patients wait years for an accurate diagnosis.
Rare disease programmes have shown that clinicians who participate in accredited training, as compared with similar cohorts who do not, order significantly more genetic tests and code more often for rare diseases. Data reveal striking gaps in physician awareness of rare disease prevalence, with a majority reporting that they never or rarely (defined as 1-2x per year) see rare disease patients despite evidence suggesting they should be seeing several per week.