Comparison of risk-adjustment models using administrative or clinical data for outcome prediction in patients after myocardial infarction or coronary bypass surgery in Korea

Summary Objective:  The objectives of this study were to compare the performance indicators of risk-adjustment models based on administrative and clinical data in Korea, and to assess whether administrative data alone is useful for comparing quality of care. Methods:  Outcome was defined as death within 30 days of discharge. For administrative data, the risk factors were; age, sex, and 11 past histories and two past major procedures, which were retrospectively chased in National Health Insurance database using patient Identification Number. For clinical data, the severity score of the three risk-adjustment measures [MedisGroups, Disease Staging (DS) and Computerized Severity Index (CSI)] was used as the independent predictors of 30-day mortality. Risk-adjustment models were developed by logistic regression analysis for 13,885 Acute Myocardial Infarction (AMI) and 2115 Coronary Artery Bypass Graft (CABG) patients based on administrative data and for 208 AMI patients and 478 CABG patients using clinical data. Performances of models were examined using c-statistic and Hosmer-Lemeshow statistic. Results:  The results obtained showed the superiority of the clinical model. For AMI, the c-statistic of the administrative model was 0.696, and those of the CSI, DS and MedisGroups models were 0.772, 0.861 and 0.988 respectively. For CABG, the c-statistic of the administrative model was 0.568, and those of the CSI, DS and MedisGroups models were 0.665, 0.731 and 0.816 respectively. Conclusion:  Our results indicate that risk-adjustment model only using administrative data are probably not useful for assessing quality of care in Korea.