From Holmes to House: Visualizing Patient Complexity Through Evolving Clinical Pathways

Session Track: Graph + AI in Production

Session Time: 01:30 pm - 02:00 pm April 15

Session description

Medicine and forensic investigation are surprisingly similar—both involve forming hypotheses from incomplete information and recognizing patterns across isolated observations. It shouldn’t surprise you, then, that Sherlock Holmes was based on an actual physician—Dr. Joseph Bell, Arthur Conan Doyle’s professor at Edinburgh Medical School.

Bell’s approach was remarkable: observing a patient’s hands and gait, he could deduce their profession, exposures, and risk factors. He didn’t see isolated symptoms—he saw narratives. This describes how good medicine works, yet our information systems fail at it, excelling at recording isolated events rather than continuous processes.

This is why the speaker’s team introduced clinical pathways as a medical information organization tool—representing patient journeys as evolving processes rather than collections of isolated encounters, treating medical events the way Bell treated observations: as interconnected pieces of a larger story.

The technical foundation rests on three complementary technologies: knowledge graphs, large language models, and visualization.

Knowledge graphs provide the structural backbone, capturing networks of connections rather than isolated records. Graph algorithms offer precision—finding patterns, identifying conflicts, and detecting information gaps through deterministic analysis. Large language models transform unstructured clinical text into structured entities, enable natural language queries, and generate readable reports—providing flexibility where graph algorithms provide precision.

Interactive visualization ties this together. The human visual cortex processes complex relationships orders of magnitude faster visually than through text, making it essential for comprehending interconnected pathway data.

The presentation will explore the philosophy and technical architecture behind this approach. You will learn when knowledge graphs make sense as an architectural choice, where AI excels and fails, and why visualization becomes crucial for complex data. The speaker will demonstrate these concepts with real-world examples using anonymized patient data.

The techniques the speaker will present aren’t limited to medical data: the same principles can be applied to any domain that requires a process understanding.