When more specific isn’t always more accurate
AI can recognize a diagnosis. But can it determine how specific that diagnosis should be without going beyond what the documentation supports?
This two-minute demo addresses that question, demonstrating the value of grounding AI models in clinical evidence.
The video considers a 65-year-old patient with hypertension and chronic kidney disease (CKD). The clinical note documents stage 3 CKD and an eGFR of 45. There may be more specific CKD concepts available, including stage 3A and stage 3B, but the documentation doesn’t necessarily support choosing one.
That’s an important distinction; an AI system needs clinical context to understand which level of specificity the evidence actually supports.
This is where a healthcare knowledge graph can help.
IMO Health’s Knowledge Graph connects clinical concepts through curated relationships, giving AI richer context for reasoning about a diagnosis. In this scenario, an AI agent can normalize CKD to a precise clinical concept, explore related refinements, and use the documented stage and eGFR to determine the appropriate level of specificity.
Just as importantly, it can recognize where to stop.
Although the graph contains more granular CKD concepts, the agent doesn’t apply additional specificity when the note doesn’t provide sufficient evidence. Instead, it can surface the supported refinement while making the reasoning and documentation gap visible.
This helps AI become appropriately specific while ensuring its reasoning stays grounded in what’s documented.
