Derivation and validation of a risk assessment model for drug-resistant pathogens in hospitalized patients with community-acquired pneumonia

Objective: To derive and validate a model for risk of resistance to first-line community-acquired pneumonia (CAP) therapy. Design: We developed a logistic regression prediction model from a large multihospital discharge database and validated it versus the Drug Resistance in Pneumonia (DRIP) score in a holdout sample and another hospital system outside that database. Resistance to first-line CAP therapy (quinolone or third generation cephalosporin plus macrolide) was based on blood or respiratory cultures. Setting: This study was conducted using data from 177 Premier Healthcare database hospitals and 11 Cleveland Clinic hospitals. Participants: Adults hospitalized for CAP. Exposure: Risk factors for resistant infection. Results: Among 138,762 eligible patients in the Premier database, 12,181 (8.8%) had positive cultures and 5,200 (3.8%) had organisms resistant to CAP therapy. Infection with a resistant organism in the previous year was the strongest predictor of resistance; markers of acute illness (eg, receipt of mechanical ventilation or vasopressors) and chronic illness (eg, pressure ulcer, paralysis) were also associated with resistant infections. Our model outperformed the DRIP score with a C-statistic of 0.71 versus 0.63 for the DRIP score (P < .001) in the Premier holdout sample, and 0.65 versus 0.58 (P < .001) in Cleveland Clinic hospitals. Clinicians at Premier facilities used broad-spectrum antibiotics for 20%–30% of patients. In discriminating between patients with and without resistant infections, physician judgment slightly outperformed the DRIP instrument but not our model. Conclusions: Our model predicting infection with a resistant pathogen outperformed both the DRIP score and physician practice in an external validation set. Its integration into practice could reduce unnecessary use of broad-spectrum antibiotics.