Association between diagnosis code expansion and changes in 30-day risk-adjusted outcomes for cardiovascular diseases

BACKGROUND: In January 2011, Centers for Medicare and Medicaid Services expanded the number of inpatient diagnosis codes from 9 to 25, which may influence comorbidity counts and risk-adjusted outcome rates for studies spanning January 2011. This study examines the association between (1) limiting versus not limiting diagnosis codes after 2011, (2) using inpatient-only versus inpatient and outpatient data, and (3) using logistic regression versus the Centers for Medicare and Medicaid Services risk-standardized methodology and changes in risk-adjusted outcomes. METHODS AND RESULTS: Using 100% Medicare inpatient and outpatient files between January 2009 and December 2013, we created 2 cohorts of fee-for-service beneficiaries aged ≥65 years. The acute myocardial infarction cohort and the heart failure cohort had 578 728 and 1 595 069 hospitalizations, respectively. We calculate comorbidities using (1) inpatient-only limited diagnoses, (2) inpatient-only unlimited diagnoses, (3) inpatient and outpatient limited diagnoses, and (4) inpatient and outpatient unlimited diagnoses. Across both cohorts, International Classification of Diseases, Ninth Revision (ICD-9) diagnoses and hierarchical condition categories increased after 2011. When outpatient data were included, there were no significant differences in risk-adjusted readmission rates using logistic regression or the Centers for Medicare and Medicaid Services risk standardization. A difference-in-differences analysis of risk-adjusted readmission trends before versus after 2011 found that no significant differences between limited and unlimited models for either cohort. CONCLUSIONS: For studies that span 2011, researchers should consider limiting the number of inpatient diagnosis codes to 9 and/or including outpatient data to minimize the impact of the code expansion on comorbidity counts. However, the 2011 code expansion does not appear to significantly affect risk-adjusted readmission rate estimates using either logistic or risk-standardization models or when using or excluding outpatient data.