A Head-to-Head Comparison of the Validity and Predictive Ability of Diagnosis Versus Medication-Based Comorbidity Indices

OBJECTIVES: To compare the performance of the diagnosis-based Charlson comorbidity index (CCI) and medication-based Rx-Risk comorbidity index (Rx-Risk).

METHODS: This analysis used data from n=221 patients from the InDePendent Study. CCI and Rx-Risk scores were calculated using documented ICD-10 diagnoses and prescribed medications from general practitioner practices. Health-related quality of life (HRQoL, EQ-5D-5L), functional impairment (B-ADL), cognitive decline (MMSE), and healthcare utilization (physician visits, hospitalizations) were assessed via interviews. Indices performance was evaluated in terms of agreement (Cohens Kappa (k)), known-group validity (ANOVA, t-test), convergent validity (correlation coefficient (r_s)) and predictive ability (R², Akaike information criterion (AIC)).

RESULTS: Patients averaged 80 years old, were more likely female (55%), and had, on average, 12 diagnoses and seven medications. The agreement between both indices was poor for all conditions except diabetes (k =0.645) and chronic airway diseases (k =0.486). Rx-Risk differed more than CCI in known groups, especially for HRQoL and hospitalization. Both indices yielded poor correlations. However, Rx-Risk had a better convergent validity than CCI for the EQ-5D index (r_s, -0.215 vs. -0.134), and risk of hospitalization (r_s, 0.145 vs. 0.128). The predictive ability of both indices was good. Rx-Risk performed better than the CCI in predicting the change of the EQ-5D index (R², 30% vs. 28%), functional (R², 55 vs. 52%) and cognitive decline (R², 46 vs. 47%), and all EQ-5D dimensions (mobility (AIC, 489.4 vs. 490.2); self-care (AIC, 552.5 vs. 557.5); usual activities (AIC, 580.6 vs. 584.4); pain/discomfort (AIC, 467.2 vs. 470.0); anxiety/depression (AIC, 446.9 vs. 447.3)) and physician consultations (AIC, 649.2 vs. 551.0), except for hospitalization (AIC, 149.2 vs. 147.1).

CONCLUSIONS: The Rx-Risk's was slightly superior in validity and predictive ability, especially regarding HRQoL and healthcare utilization. However, caution is warranted due to limitations in assessing functional and cognitive impairment, suggesting that alternative instruments should be considered to predict such outcomes better.