367 Validation of a Prediction Instrument for Endocarditis in Injection Drug Users With Fever

Abstract

To validate the diagnostic performance of a previously derived endocarditis prediction instrument to rule out endocarditis in febrile injection drug users. From June 2007 to March 2001, we collected clinical prediction data from the first 6 hours of emergency department (ED) presentation of febrile injection drug users patients admitted from 2 urban EDs. Outcome data from inpatient records (echocardiogram, blood culture results, and discharge diagnoses) were independently abstracted by 2 investigators who were blinded to the ED prediction data. Diagnosis of endocarditis was based on Modified Duke criteria and documentation of endocarditis upon discharge. We determined the screening performance of the individual criteria and the endocarditis prediction instrument (combination of tachycardia, cardiac murmur, and absence of skin infection) Table 1.Tabled 1Screening performance of clinical criteriaTachycardiaCardiac MurmurAbsence of skin infectionDecision instrument (All 3)CXR infiltratesHistory of endocarditisSensitivity (95% CI)0.889 (0.653–0.986)0.722 (0.465–0.903)0.944 (0.727–0.999)1.000 (0.815–1.000)0.444 (0.215–0.692)0.556 (0.308–0.785)Specificity (95% CI)0.433 (0.368–0.500)0.706 (0.642–0.764)0.433 (0.368–0.500)0.134 (0.093–0.185)0.840 (0.786–0.885)0.879 (0.830–0.918)Positive predictive value (95% CI)0.109 (0.064–0.171)0.160 (0.088–0.259)0.115 (0.068–0.178)0.083 (0.050–0.127)0.178 (0.080–0.321)0.263 (0.134–0.431)Negative predictive value (95% CI)0.980 (0.931–0.998)0.970 (0.932–0.990)0.990 (0.946–1.000)1.000 (0.888–1.000)0.951 (0.912–0.976)0.962 (0.927–0.984)Odds ratio (95% CI)6.107 (1.522–24.320)6.232 (2.219–17.434)12.977 (2.155–77.483)Infinite4.195 (1.597–11.050)9.063 (3.386–24.263) Open table in a new tab Of the 249 subjects enrolled, 18 (7.3%) were diagnosed with endocarditis. Absence of skin infection had the highest sensitivity and diagnostic odds ratio. Other useful screening criteria were infiltrates on chest x-ray and history of endocarditis. Hyponatremia, thrombocytopenia, leukocytosis, and HIV status were poorly predictive of endocarditis. Endocarditis prediction instrument had 100% sensitivity (95% CI 81.5-100), 100% negative predictive value (95% CI 88.8-100), and 13.4% specificity (95% CI 9.3-18.5). In this internal validation study, the endocarditis prediction instrument consisting of tachycardia, murmur, and absence of skin infection effectively ruled out endocarditis with 100% sensitivity and 100% negative predictive value, consistent with the derivation study results. If external validation demonstrates similar screening performance, the endocarditis prediction instrument may safely decrease admissions for “rule out endocarditis” in febrile injection drug users.