External validation of risk scores and multivariate models for the diagnosis of community-acquired pneumonia in outpatients

Abstract

Background: While several risk scores for the diagnosis of community-acquired pneumonia (CAP) have been developed, they require prospective external validation. Objectives: To externally validate existing prediction models, risk scores, and heuristics for the diagnosis of CAP in adults. Methods: The Enhancing Antibiotic Stewardship in Primary Care (EAST-PC) study recorded signs, symptoms, demographics, and vitals in 718 adults presenting to primary or urgent care clinics with acute lower respiratory tract infection between 2019 and 2023. C-reactive protein (CRP) was available for 575. The diagnosis of CAP was based on the clinician diagnosis and/or chest radiograph. Literature was searched for previous risk scores. Using the EAST-PC population, the area under the receiver operating characteristic curve (AUROCC), calibration curves, and percentage with CAP in each risk group were calculated for each risk score. Results: We identified 11 studies describing 4 risk scores, 9 multivariate models, and 5 simple heuristics. The Genomics to Combat Resistance Against Antibiotics in Community-acquired LRTI in Europe (GRACE) risk score using the absence of a runny nose, the presence of breathlessness, crackles, diminished vesicular breathing, heart rate > 100/min, temperature >37.8 °C, and CRP > 30 mg/L was the most accurate (AUROCC 0.81). It classified 280 patients as low (0.7% CAP), 265 as moderate (5.7%) and 30 as high risk (33.3%) for CAP. The GRACE score without CRP performed similarly. Other risk scores had poor calibration or failed to accurately classify patients as low or high risk. Conclusions: The previously derived GRACE risk scores were successfully externally validated in a contemporary US outpatient population.