Inability to validate a predictive score for infective endocarditis in intravenous drug users

Abstract

We sought to identify presenting features that should increase the accuracy of diagnosis of endocarditis in symptomatic intravenous drug use (IVDU) patients. We used two data sets of IVDU patients to develop and validate a score for endocarditis. A retrospective analysis of admitted IVDU patients was conducted with subsequent validation of endocarditis score on consecutive admitted IVDU patients at the same urban teaching hospital emergency department. Statistical comparison was made of presenting clinical features in patients with and without positive blood cultures or vegetation on their echocardiogram. Multivariable analysis of significant features was carried out. A clinical score was developed using a likelihood ratio to determine weighting. Validation of the score was tested by application to a second set of admitted symptomatic IVDU patients. The performance of the score was assessed by comparison of receiver operating characteristic curves for the two data sets. Factors associated with endocarditis on multivariable analysis were past history of endocarditis (negative correlation), total white blood cell (WBC) count, percentage neutrophils and bands on differential WBC count, infiltrate on chest x-ray, and arterial oxygenation (negative correlation). Prospective validation showed poor predictive performance of the endocarditis score. Presenting clinical, radiographic, and laboratory features do not appear helpful in estimating the clinical likelihood of endocarditis in symptomatic IVDU patients.