Are Claims Data Accurate Enough to Identify Patients for Performance Measures or Quality Improvement? The Case of Diabetes, Heart Disease, and Depression

Abstract

The objective of this study was to demonstrate a method to accurately identify patients with specific conditions from claims data for care improvement or performance measurement. In an iterative process of trial case definitions followed by review of repeated random samples of 10 to 20 cases for diabetes, heart disease, or newly treated depression, a final identification algorithm was created from claims files of health plan members. A final sample was used to calculate the positive predictive value (PPV). Each condition had unacceptably low PPVs (0.20, 0.60, and 0.65) when cases were identified on the basis of only 1 International Classification of Diseases, ninth revision, code per year. Requiring 2 outpatient codes or 1 inpatient code within 12 months (plus consideration of medication data for diabetes and extra criteria for depression) resulted in PPVs of 0.97, 0.95, and 0.95. This approach is feasible and necessary for those wanting to use administrative data for case identification for performance measurement or quality improvement.