PMU40 VALIDATION OF AN ICD-10 CODING ADAPTATION FOR THE CHARLSON COMORBIDITY INDEX IN UNITED STATES HEALTHCARE ADMINISTRATIVE DATA

Abstract

An ICD-10 coding adaptation of the Charlson comorbidity index (CCI) has not yet been validated for predicting one-year mortality in US healthcare data. We validated a previously developed coding adaptation in US administrative claims and compared its performance with the Canadian adaptation (Quan et al. Med Care 2005;43:1130-9) currently used in some US healthcare studies. Cohorts of patients with diverse medical conditions (rheumatoid arthritis, hip/knee replacement, lumbar spine surgery, AMI, stroke, pneumonia) in the IBM MarketScan Research Databases (a US commercial claims data source) were linked with the IBM MarketScan Mortality file. Predictive ability of both the US and Canadian coding adaptations of the CCI was measured using c-statistics, after adjusting for age and sex, and compared by the method of DeLong and colleagues (Biometrics; 1988;44:837-45). Full code lists are available at https://doi.org/10.5281/zenodo.3604394. C-statistics were generally high (∼0.8 or greater) for five of seven patient cohorts. Results were similar between the US and Canadian adaptations, with absolute differences <0.001. For lumbar spine surgery, CCI contributed a more substantial role in predicting 12-month mortality than for other conditions. For hip/knee replacement and pneumonia, age and sex played a more substantial role, while for AMI and stroke, other factors were likely more important (e.g., distribution of vascular disease, severity, health delivery factors, etc) although age, sex, and CCI play some role for predicting mortality. This US ICD-10 coding adaptation of the CCI performed well in US commercial claims for predicting one-year mortality and similarly to the Canadian adaptation. Using the Canadian adaptation had minimal impact on predictive ability but could result in erroneous assignment of comorbidities (i.e., construct validity). We recommend using adaptations specific to the country of origin of the data based on good research practice.