Finding Undiagnosed Cancer Cachexia Patients Using Guideline-based AI
Poster Presentation
Learn how, in collaboration with NHS Lothian, Pangaea’s AI platform was configured on four cachexia guidelines (Fearon, Evans, GLIM and mGPS criteria) and outperformed baseline LLM prompting, achieving 100% precision and 92.3% sensitivity for identifying patients with undiagnosed cancer cachexia and 89.7% precision and 96.3% sensitivity for documented cancer cachexia.
The results demonstrate that embedding guidelines directly into the AI, patient records can be systematically assessed without clinician prompting, generating repeatable, actionable insights to support timely clinical assessment and intervention.