Difficulty in predicting bacteremia in elderly emergency patients

Abstract

To characterize the clinical presentation and identify factors predictive of bacteremia in elderly patients. Retrospective review of emergency department charts, hospital records, and microbiology reports. Community teaching hospital with annual ED census of 65,000 adults. Seven hundred fifty elderly patients (aged 65 to 99 years) who were evaluated by the emergency physician, had blood cultures obtained in the ED, and were hospitalized with a suspected infectious process during a 12-month period. Records were analyzed for demographic information, underlying diseases, clinical presentation, laboratory findings, sources of infection, and causative organisms. Using contingency tables, 79 patients with positive blood cultures were compared with a random sample of 136 patients with sterile blood cultures to identify clinical variables significantly (P less than .05) associated with bacteremia. Logistic regression analysis was performed with significant factors to develop a model to predict bacteremia. Sensitivity, specificity, and predictive values were calculated for the model. The prevalence of bacteremia was 10.6%. Escherichia coli was the most commonly isolated pathogen (29% of cases), and the urinary tract was the most common source of infection (44.3% of cases). Logistic regression analysis showed that altered mental status (odds ratio, 2.88; 95% confidence interval [Cl], 1.52 to 5.50), vomiting (odds ratio, 2.63; 95% Cl, 1.16 to 6.15), and WBC band forms of more than 6% (0.06) (odds ratio, 3.50; 95% Cl, 1.58 to 5.27) were independent predictors of bacteremia. The presence of at least one of these three factors had a sensitivity of 0.85 (95% Cl, 0.75 to 0.92) and a specificity of 0.46 (95% Cl, 0.38 to 0.55) for predicting bacteremia in the study group. The positive predictive value was 0.16 (95% Cl, 0.12 to 0.19) and the negative predictive value was 0.96 (95% Cl, 0.94 to 0.98) for the ED patient group that met inclusion criteria. Elderly patients fail to manifest identifiable clinical features indicative of bloodstream infection. The sensitivity and specificity of the best statistical model for identifying bacteremic elderly patients suggest that clinical indicators alone are unreliable predictors of bacteremia in the geriatric ED population studied.