Abstract

Objectives Our objectives were a) to identify univariate correlates of death in emergency department patients at risk for infection; b) to perform multivariate analyses and identify independent predictors of death; and c) to develop and internally validate a prediction rule that may be used in the emergency department to risk stratify patients into different risk groups to predict their mortality rate. Design Prospective cohort study. Setting Emergency department of an urban university referral center. Patients Consecutive emergency department patients, aged 18 or older, who were at risk for infection, as indicated by the emergency department physician ordering a blood culture between February 1, 2000, and February 1, 2001. Of 3,301 eligible patient visits, 3,179 (96%) were enrolled. Interventions None. Measurements and Main Results The primary outcome was 28-day in-hospital mortality rate. There were 2,070 visits in the derivation set, with 110 deaths (5.3%), and 1,109 visits in the validation set, with 63 deaths (5.7%). Independent multivariate predictors of death were terminal illness (odds ratio, 6.1; 95% confidence interval, 3.6–10.2), tachypnea or hypoxia (2.7, 1.6–4.3), septic shock (2.7, 1.2–5.7), platelet count 5% (2.3, 1.5–3.5), age >65 (2.2, 1.3–3.6), lower respiratory infection (1.9, 1.2–3.0), nursing home residence (1.9, 1.2–3.0), and altered mental status (1.6, 1.0–2.6). The clinical prediction rule stratified patients into mortality risk groups of very low, 0.9% (95% confidence interval, 0.2–1.5%); low, 2.0% (0.8–3.2%); moderate, 7.8% (5.6–10%); high, 20% (13–27%); and very high, 50% (36.1–64%) in the derivation set. Mortality rates for the corresponding risk groups in the validation set were 1.1%, 4.4%, 9.3%, 16%, and 39%, respectively. The receiver operating characteristic area for the rule was 0.82 in the derivation set and 0.78 in the validation set. Conclusions In patients with suspected infection, this model identifies significant correlates of death and allows stratification of patients according to mortality risk. As new therapies become available for patients with sepsis syndromes, the ability to predict mortality risk may be helpful in triage and treatment decisions.