Validation of the DANCAMI, the Charlson Comorbidity Index, and the Elixhauser Comorbidity Index for predicting mortality in patients with infective endocarditis

Abstract

Abstract Introduction Comorbidity scores are often used to adjusted for confounding by comorbidity burden in population-based research as they increase statistical efficiency in smaller study populations. However, there is currently no comorbidity index available for predicting infective endocarditis prognosis. This gap in the literature hinders the ability to accurately adjust for confounding in studies investigating the prognosis of infective endocarditis, highlighting the need of validating existing comorbidity indices for this purpose. Purpose To validate the discriminatory ability of the Danish Comorbidity Index for Acute Myocardial Infarction (DANCAMI), the Charlson Comorbidity Index (CCI), and the Elixhauser Comorbidity Index (ECI) for predicting in-hospital, 1-year, and 10-year all-cause mortality after infective endocarditis. Methods We used nationwide Danish health registries to identify all patients with a first-time inpatient, primary or secondary diagnosis for infective endocarditis between 1995 and 2021 and their comorbidities. When identifying the comorbidities, we used all in and outpatient hospital information in the 10 years before the infective endocarditis diagnosis. Based on the presence of comorbidities, we estimated the DANCAMI, the CCI, and the ECI scores for each patient. Using a logistic regression model, we calculated area under the receiver operating curves (AUCs) for a baseline model including age and sex and for models including the scores from the comorbidity indices plus age and sex for in-hospital, 1-year, and 10-year all-cause mortality. The AUC describes the probability that for a pair of random patients, the model will assign a greater predicted risk to the patient dying. Results Overall, we identified 12,530 patients with infective endocarditis. The median age was 71 years and 35% were females. For the individual analyses, we had complete follow-up on 12,493 patients when predicting in-hospital mortality, 11,587 when predicting 1-year mortality, and 5,880 patients when predicting 10-year mortality. For in-hospital mortality, the AUCs were 0.62 for the baseline model, 0.65 for the DANCAMI, 0.64 for the CCI, and 0.65 for the ECI. For 1-year mortality, the AUCs were 0.66 for the baseline model, 0.71 for the DANCAMI, 0.70 for the CCI, and 0.70 for the ECI. For 10-year mortality, the AUCs were 0.76 for the baseline model, 0.82 for the DANCAMI, 0.81 for the CCI, and 0.81 for the ECI. Conclusions After accounting for sex and age, adding either the DANCAMI, the CCI, or the ECI did not increase the discriminatory ability for predicting in-hospital all-cause mortality after infective endocarditis but did increase the discriminatory ability for predicting 1-year and 10-year all-cause mortality. Consequently, comorbidity scores might be beneficial in reducing confounding by comorbidity burden when predicting 1-year or 10-year infective endocarditis mortality, depending on their association with the exposure of interest.