The impact of clinical vs administrative claims coding on hospital risk-adjusted outcomes

Abstract

Background: Comorbid condition and hospital risk-adjusted outcomes prevalence were compared based on clinical registry vs administrative claims data. Hypothesis: Risk-adjusted outcomes will vary depending on the source of comorbidity data used. Methods: Clinical data from hospitalized Can Rapid Risk Stratification of Unstable Angina Patients Suppress Adverse Outcomes with Early Implementation of the American College of Cardiology/American Heart Association (ACC/AHA) Guidelines (CRUSADE) non-ST-segment elevation myocardial infarction (NSTEMI) patients ≥65 years was linked to Medicare claims. Eight common comorbid conditions were coded and compared between registry data (derived from medical record review) and claims data; hospital-level observed vs expected ratios and outlier status for 30-day readmission and mortality were calculated using logistic generalized estimating equations for clinical vs claims data. Results: Of 68 199 NSTEMI patients, 48.1% were female, 86.9% were white, and median age was 78. Degree of agreement between data sources for comorbid condition prevalence was 67.8% for myocardial infarction and 89.3% for diabetes. Overall, multivariable model performance was similar: Medicare mortality c-statistics is 0.69 vs CRUSADE is 0.71; readmission c-statistics is 0.59 for both. Hospital ratings were similar regardless of data source (mortality, R2 = 0.97863; readmission, R2 = 0.97858). Eighty-two hospitals were mortality outliers in claims-based models; of these, 70 were outliers in registry-based models. Forty-five hospitals were readmission outliers in claims-based models; of these, 39 were outliers in registry-based models. Conclusions: There were significant differences in individual comorbid condition prevalence when derived from registries vs claims, but hospital-level outcomes were comparable.