The performance of five comorbidity measures, including the Charlson and Elixhauser indices, was investigated for predicting mortality, hospitalization, and fracture outcomes in two osteoporosis cohorts defined from administrative databases. The optimal comorbidity measure depended on the outcome of interest, although overall the Elixhauser index performed well.
Introduction: Studies that use administrative data to investigate population-based health outcomes often adopt risk-adjustment models that include comorbidities, conditions that coexist with the index disease. There has been limited research about the measurement of comorbidity in osteoporotic populations. The study purpose was to compare the performance of comorbidity measures for predicting mortality, fracture, and health service utilization outcomes in two cohorts with diagnosed or treated osteoporosis.
Methods: Administrative data were from the province of Saskatchewan, Canada. Osteoporosis cohorts were identified from diagnoses in hospital and physician data and prescriptions for osteo-protective medications using case definitions with high sensitivity or high specificity. Five diagnosis- and medication-based comorbidity measures and five 1-year outcomes, including mortality, hospitalization (two measures), osteoporotic-related fracture, and hip fracture, were defined. Performance of the comorbidity measures was assessed using the c-statistic (discrimination) and Brier score (prediction error) for multiple logistic regression models.
Results: In the specific cohort (n = 9,849) for the mortality outcome, the Elixhauser index resulted in the largest improvement (8.96%) in the c-statistic and lowest Brier score compared to a model that contained demographic and socioeconomic variables, followed by the Charlson index (6.06%). For hospitalization, the number of different diagnoses resulted in the largest improvement (14.01%) in the c-statistic. The Elixhauser index resulted in significant improvements in the c-statistic for osteoporosis-related and hip fractures. Similar results were observed for the sensitive cohort (n = 28,068).
Conclusions: Recommendations about the optimal comorbidity measure will vary with the outcome under investigation. Overall, the Elixhauser index performed well.