Background: Administrative data are used in clinical research, but the validity of ICD-9 codes to identify cirrhotic patients has not been well established.
Goals: To determine the diagnostic accuracy of ICD-9 codes for cirrhosis in clinical practice.
Study: We conducted a retrospective cohort study of patients from a safety-net hospital between 2008 and 2011. Patients were initially identified using ICD-9 codes for cirrhosis or a resultant complication. The gold-standard for diagnosis of cirrhosis was histology and/or imaging based on medical record review. Sensitivity, specificity, positive predictive values, and negative predictive values for each ICD-9 code were calculated. Diagnostic accuracy was assessed by the c-statistic using receiver operator characteristic curve analysis.
Results: We identified 2893 patients with an ICD-9 code for cirrhosis, of whom 50.2% had 1 ICD-9 code, 20.3% had 2 different codes, and 29.5% had 3 or more codes. Cirrhosis was confirmed in 44.0% of patients with 1 ICD-9 code, 82.6% with 2 codes, and 95.7% of those with at least 3 codes. Ascites had a significantly lower positive predictive values for cirrhosis than other ICD-9 codes (P<0.001). The optimal combination of ICD-9 codes to identify cirrhotic patients included all codes except that of ascites, with a c-statistic of 0.71 in our derivation cohort. The sensitivity of this combination was confirmed to be 98% in a validation cohort of 285 patients with known cirrhosis.
Conclusions: Administrative data can identify patients with cirrhosis with high accuracy, although ascites has a significantly lower positive predictive value than other ICD-9 codes.