Predicting the need for ICU admission in community-acquired pneumonia

Abstract

Background: Multiple criteria have been proposed to define community-acquired pneumonia (CAP) severity and predict ICU admission. Validity studies have found differing results. We tested four models to assess severe CAP built upon the criteria included in the 2007 IDSA/ATS guidelines, hypothesizing that a model providing different weights for each individual criterion may be of better predictability.

Methods: Retrospective analysis of a prospective cohort study of adult hospitalizations for CAP at nine hospitals in Louisville, KY from June 2014 to May 2016. Four models were tested. Model 1: original 2007 IDSA/ATS criteria. Model 2: modified IDSA/ATS criteria by removing multilobar infiltrates and changing BUN threshold to ≥30 mg/dL; adding lactate level >2 mmol/L and requirement of non-invasive mechanical ventilation (NIMV). CAP was severe with 1 major criterion or 3 minor criteria. Model 3: same criteria as model 2, CAP was severe with 1 major criterion or 4 minor criteria. Model 4: multiple regression analysis including the modified criteria as described in models 2 and 3 with a score assigned to each variable according to the magnitude of association between variable and need for ICU.

Results: 8284 CAP hospitalizations were included. 1458 (18%) required ICU. Model 4 showed highest prediction of need for ICU with an area under the curve of 0.91, highest accuracy, specificity, positive predictive value, and agreement among models.

Conclusion: Assigning differential weights to clinical predictive variables generated a score with accuracy that outperformed the original 2007 IDSA/ATS criteria for severe CAP and ICU admission.

Keywords: Cohort; ICU; Infection; Pneumonia; Prediction; Prognosis.