CRICO Grant Spotlight: Using Artifical Intelligence to Reduce Missed Diagnoses

Through a CRICO grant-funded project Artificial Intelligence to Enhance a Cognitive Aid for Identifying Patients at Risk of Missed Diagnosis, Mitchell Feldman, MD, and Kavishwar Wagholikar, MD, PhD, developed and evaluated a cognitive aid for clinicians to identify patients at risk of a missed or delayed diagnosis.

Specifically, they developed a tool using natural language processing (NLP) to extract findings from the narrative text of an electronic health record (EHR), optimized the tool through machine learning, and created an application programming interface to integrate with the DXplain Clinical Decision Support System, a decision support system developed at the Laboratory of Computer Science at the Massachusetts General Hospital. Drs. Feldman and Wagholikar then evaluated the tool’s ability to identify high information findings in the EHR and the tool’s overall diagnostic accuracy of the disease assessment.

This project demonstrated DXplain’s potential for automating the early identification of patients with missed diagnoses. The DXplain system is available for use by the Harvard Medical School community as a quick consult for augmenting a clinician’s diagnosis.

While this NLP-enhanced system is still only a research tool, Drs. Feldman and Wagholikar continue to collaborate on its use with several partners, including WebMD and Curai.

Dennis Markovitz, MD, DABFM, with Winchester Hospital, has used this tool and states “DxPlain is a leader in the field of computer aided diagnosis. It’s incredibly important in making accurate diagnoses, avoiding excessive testing, providing cost-effective care, and most importantly improving patient safety.”

For additional information: Contact Dr. Feldman at [email protected]

Grant-related publication (which may also be requested via CRICO):
American Medical Informatics Association manuscript: Natural Language Processing to Detect High Information Findings for Patients at Risk of Missed Diagnosis