Abstract

Accurate identification of patients with cirrhosis is important for research using administrative databases. We aimed to examine the accuracy of several major ICD-10 codes for cirrhosis diagnosis in a large and diverse patient cohort; there is little existing research on this topic.Using data from 3396 patients with chronic liver disease (hepatitis B, C or non-alcoholic fatty liver disease) from one university and several community medical centers, we calculated sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and area under the receiver operating characteristic curve (AUROC) for several major ICD-10 codes for cirrhosis, which was verified by individual chart review. We performed a secondary validation in a general cohort of 1560 randomly selected patients.While each of the individual study ICD-10 codes were specific (98.08 - 100%), none of the codes were sufficiently sensitive (0.27 - 55.70%). PPVs were high in the chronic liver disease cohort (88.41 - 100%) but lower in the general population (55.53 - 66.76%). The AUROC for having at least one code was higher (0.79) than any code alone (0.50 - 0.65).Individual ICD-10 codes are suboptimal for identifying patients with cirrhosis in the general patient population. We recommend conditioning ICD-10 code searches with a chronic liver disease diagnosis code and/or combining diagnostic codes to maximize performance.

View details for DOI 10.1159/000510981

View details for PubMedID 32814313