Clinical Decision Support for Colonoscopy Surveillance Using Natural Language Processing

Abstract

Colorectal cancer is the second leading cause of cancer-related deaths in the United States. However, 41% of patients do not receive adequate screening, since the surveillance guidelines for colonoscopy are complex and are not easily recalled by health care providers. As a potential solution, we developed a guideline based clinical decision support system (CDSS) that can interpret relevant freetext reports including indications, pathology and procedure notes. The CDSS was evaluated by comparing its recommendations with those of a gastroenterologist for a test set of 53 patients. The CDSS made the optimal recommendation in 48 cases, and helped the gastroenterologist revise the recommendation in 3 cases. We performed an error analysis for the 5 failure cases, and subsequently were able to modify the CDSS to output the correct recommendation for all the test cases. Results indicate that the system has a high potential for clinical deployment, but further evaluation and optimization is required. Limitations of our study are that the study was conducted at a single institution and with a single expert, and the evaluation did not include rare decision scenarios. Overall our work demonstrates the utility of natural language processing to enhance clinical decision support.