Objectives: To identify independent risk factors for death in elderly emergency department (ED) patients admitted for infection and to derive and validate a mortality-prediction rule for such patients.
Design: Prospective cohort study.
Setting: Tertiary hospital ED with 55,000 annual visits.
Participants: ED patients aged 65 and older admitted for infection between December 2003 and September 2004 in the derivation cohort and October 2005 and October 2006 in the validation cohort.
Primary outcome: 28-day in-hospital mortality. Data were extracted from charts, and multivariate logistic regression were performed to identify independent mortality predictors. A prediction model was constructed and then validated in a second cohort.
Results: Nine hundred thirty-five patients were included in the derivation cohort and 2,015 in the validation cohort. Mortality was 6% in the derivation cohort and 7% in the validation cohort. In the derivation cohort, logistic regression revealed five independent mortality predictors: respiratory compromise (respiratory rate >20 breaths per minute or hypoxemia) (odds ratio (OR)=4.0, 95% confidence interval (CI)=1.7-9.4), tachycardia (heart rate > or = 120 betas per minute; OR=3.2, 95% CI=1.6-6.3), cardiovascular failure (systolic blood pressure <90 mmHg despite fluid challenge or lactate > or = 4.0; OR=9.0, 95% CI=4.7-17), preexisting terminal illness (OR=5.7, 95% CI=2.2-15), and platelet count less than 150,000/mm3 (OR=2.7, 95% CI=1.3-5.6). Mortality increased with the number of factors: 0.51% for no factors, 3.1% for one factor, 14% for two factors, 47% for three or more risk factors. The c-statistic was 0.87 for the derivation model and 0.74 for the validation model. Almost 80% of patients in both cohorts were in low-risk groups (0 or 1 factor).
Conclusion: A rule derived from five readily available variables predicts mortality in infected elderly ED patients and allows identification of a large low-risk subgroup.