DAVE: Differential Diagnostic Analysis Automation and Visualization from Clinical Notes

Electronic Medical Records are integral parts of modern healthcare. Part of the records are clinical notes that healthcare providers take during encounters with patients. Notes are key to differential analysis which is the reasoning process leading to diagnosis and treatment. This paper presents DAVE, a differential analysis automation and visualization to assist healthcare professionals through the differential analysis process. DAVE takes as input clinical notes as they are being written by professionals and suggests candidate diagnostic algorithms. We digitized textbook diagnostic algorithms into directed acyclic graphs. We trained a distributional semantics model using an annotated corpora of electronic medical records and text from diagnostic algorithm descriptions. The model, boosted with PUBMed-based semantic similarity metrics, ranks the diagnostic algorithm graphs and suggests the top three. The model achieved 74.3% success rate and was highly accepted by multiple medical professionals for usability.