Mo1050 Using an Automated Diagnostic Algorithm That Utilizes Electronic Health Records and Natural Language Processing to Define a Population With Cirrhosis

Background: Identification of a population with cirrhosis is challenging given manual data collection is time-intensive and laborious, while use of International Classification of Diseases Ninth Revision (ICD-9) codes can be inaccurate. Natural language processing (NLP), a novel computerized approach to analyzing electronic free text, has been used to automatically identify other patient cohorts with gastrointestinal pathologies such as inflammatory bowel disease and intraductal papillary mucinous neoplasms. Aim: To utilize NLP as a supplement to ICD-9 codes and laboratory values to better define and risk-stratify patients with cirrhosis. Methods: A cohort of patients with ICD-9 codes for chronic liver disease was identified during March 2013 to September 2014 from an academic medical center's administrative data. Patients with cirrhosis were further characterized using an algorithm incorporating NLP of radiology reports, ICD-9 codes and laboratory data. Patients who met any of the inclusion criteria were determined to have cirrhosis. Charts were manually reviewed at random to confirm cases of cirrhosis. The reviewer was blinded to the algorithm's results. Positive predictive value (PPV), negative predictive value (NPV), sensitivity and specificity were calculated. Results: Of 4292 patients with chronic liver disease, the algorithm yielded 801 patients with cirrhosis. Of the 174 manually reviewed charts, 97 patients had cirrhosis. The algorithm had a PPV of 0.76, NPV of 0.97, sensitivity of 0.97 and specificity of 0.77. Conclusion: Combining NLP, ICD-9 codes and laboratory data is a powerful and sensitive way to detect patients with cirrhosis within a large population. Our algorithm mimics clinicians' manual review process, and therefore shows promise as a tool for automated identification of patients with cirrhosis for both clinical and research use.