Abstract
Rationale: Chronic obstructive pulmonary disease (COPD) hospitalizations are a major burden on patients. Diffusing capacity of the lung for carbon monoxide (Dl(CO)) is a potential predictor that has not been studied in large cohorts. Objectives: This study used electronic health record data to evaluate whether clinically obtained Dl(CO) predicts COPD hospitalizations. Methods: We performed time-to-event analyses of individuals with COPD and Dl(CO) measurements from the Johns Hopkins COPD Precision Medicine Center of Excellence. Cox proportional hazard methods were used to model time from Dl(CO) measurement to first COPD hospitalization and composite first hospitalization or death, adjusting for age, sex, race, body mass index, smoking status, forced expiratory volume in 1 second (FEV(1)), history of prior COPD hospitalization, and comorbidities. To identify the utility of including Dl(CO) in risk models, area under the receiver operating curve (AUC) values were calculated for models with and without Dl(CO). Results were externally validated in a separate analogous cohort. Results: Of 2,793 participants, 368 (13%) had a COPD hospitalization within 3 years. In adjusted analyses, for every 10% decrease in Dl(CO)% predicted, risk of COPD hospitalization increased by 10% (hazard ratio, 1.1; 95% confidence interval, 1.1-1.2; P < 0.001). Similar associations were observed for COPD hospitalizations or death. The model including demographics, comorbidities, FEV(1), Dl(CO), and prior COPD hospitalizations performed well, with an AUC of 0.85 and an AUC of 0.84 in an external validation cohort. Conclusions: Diffusing capacity is a strong predictor of COPD hospitalizations in a clinical cohort of individuals with COPD, independent of airflow obstruction and prior hospitalizations. These findings support incorporation of Dl(CO) in risk assessment of patients with COPD.