EVALUATING THE USE OF ADMINISTRATIVE COMORBIDITY DEFINITIONS IN A REGISTRY OF STEMI PATIENTS

Abstract

Currently, International Classification of Disease, version 10 (ICD-10) codes are used to identify comorbidities in administrative healthcare databases. Although not readily available in all jurisdictions, case definitions typically use physician billing claims data. We linked administrative data to a regional ST-elevation myocardial infarction (STEMI) registry to determine the accuracy of several alternative case definitions to identify the presence of cardiovascular comorbidities, using the registry as gold standard. Vital Heart Response (VHR) is a comprehensive registry of 3049 consecutive STEMI patients (2006 and 2011) within a defined health region (Edmonton, Alberta). Clinical information in VHR was acquired via chart review by a dedicated research organization (EPICORE). Using unique patient identifiers, the registry data were linked to the inpatient discharge abstract database (DAD) and the outpatient (ACCR – 2010; NACRS >2010) databases. The DAD includes most responsible diagnosis and up to 24 other diagnoses (all acute care hospitalizations) while the outpatient databases include up to 10 diagnoses for all visit to ERs or hospital-based clinics. Both DAD and ACCS used the Canadian enhancement of ICD-10 (ICD-10-CA) during the study time period. Four ICD-10-CA based case definitions were evaluated for each comorbidity based on: a) the Index (STEMI) hospitalization only; b) the Index hospitalization or any DAD or ACCS records in the previous year; c) the Index hospitalization or any DAD or ACCS records in the previous 2 years; and d) at least two occurrences in any DAD or ACCS records in the previous 2 years. The comorbidities of interest were: Prior Angina, Previous MI, Hypercholesterolemia, Hypertension, Diabetes, Prior Heart Failure, and Atrial Fibrillation. Concordance between VHR and administrative algorithms was compared using positive predictive value (PPV), negative predictive value (NPV), sensitivity and specificity. Results varied considerably by case definition for each comorbidity (Table), with PPV ranging from 20% for Prior Heart Failure (NPV 98.9%) to 88.2% for Previous MI (NPV 92.4%) using the algorithm based only on index hospitalizations. Similar results were obtained by incorporating single code occurrences in any DAD or ACCS visit in the prior 1 and 2 years. The algorithm that required 2 occurrences in the prior 2 years yielded the highest specificity. The capture of comorbidity data in administrative databases is reasonably accurate, although it varied depending on the condition of interest and the case definition. Different case definitions may be appropriate for cohort generation depending on whether investigators want to maximize PPV or NPV.