Search

Background Few research studies directly compare the predictive and discriminative performance of the commonly used comorbidity measures: the Charlson Comorbidity Index, Elixhauser Index (EI), Chronic Disease Score (CDS), and Health related Quality of Life Comorbidity Index (HRQL-CI). The commonly used approach of a single summative comorbidity score provides a limited view of the differential impact of individual comorbidities on healthcare outcomes. Objectives To assess and compare the predictive and discriminative performances of comorbidity measures in healthcare outcomes and evaluate the dimensionality of comorbidity using psychometric techniques. Methods The study conceptual framework drew upon the behavioral model of health services primarily. The study sample was type 2 diabetes patients from the MarketScanTM Medicaid database (2003 to 2007). Multiple regression analyses assessed the predictive performance of comorbidity measures. The c statistic assessed discriminative performance of the comorbidity indices. Confirmatory factor analyses identified dimensionality of the comorbidity indices. Three outcomes of interest were healthcare behaviors, including physician treatment adherence and patient medication adherence, utilization and expenditures. The SAS™, STATA™, and LISREL™ software were utilized for data analyses. Results The final study sample was 9,832 type 2 diabetes patients with a mean age of approximately 45 years. The CDS had the best performance in predicting physician treatment adherence and discriminating medication adherence behavior. The CDS and HRQL-CI mental aspect index had better predictive validity for medication adherence and similar discrimination of physician treatment adherence compared to the other two indices. Diagnosis-driven indexes (e.g., EI) had better predictive and discriminative performance in healthcare utilization and expenditures outcomes. A 7-factor pattern was noticeable in the correlations related to comorbidity and it provided best model fit and predictive performance across different healthcare outcomes. Individual comorbidity dimensions demonstrated differential impact for a given outcome. Conclusion For studying healthcare behaviors, the CDS and HRQL-CI mental aspect index seem to serve as better risk adjustment tools. Diagnosis-driven indexes such as the EI still need to remain the first choice for healthcare utilization and expenditures data. Accounting for comorbidity dimensionality provides better risk adjustment and insightful information regarding the differential impact of different features of comorbidities in predicting patient outcome.