RareAlert: Aligning heterogeneous large language model reasoning for early rare disease risk screening

Dimensions overall score 8.0

GitHub Code Pulse

No public code linked for this paper yet.

• RareAlert achieved an AUC of 0.917 on an independent test set for rare disease risk screening.result 95%
• RareAlert outperformed all evaluated individual LLMs including GPT-5, DeepSeek-R1, Claude-3.7-Sonnet, o3-mini, Gemini-2.5-Pro, and Qwen3-235B.result 90%
• RareAlert integrates reasoning from ten different large language models and distills them into a single locally deployable model.method 90%
• The system was developed and evaluated using RareBench, a real-world dataset of 158,666 cases covering 33 Orphanet disease categories and more than 7,000 rare conditions.method 95%
• RareAlert enables privacy-preserving rare disease risk screening suitable for large-scale local deployment.technical 85%
• The system's effectiveness depends on high-quality input data from initial clinical visits, and inaccuracies or missing information could reduce efficacy.limitation 80%
• Rare disease identification can be reconceptualized as a universal uncertainty resolution process applied to the general patient population.result 75%
• RareAlert is based on a Qwen3-4B model trained with calibrated reasoning signals.technical 90%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card