Exacerbations of chronic obstructive pulmonary disease (COPD) are a major source of morbidity and mortality, and are the endpoint in many COPD clinical trials. The ability to predict a patient?s exacerbation risk is a key goal of improving clinical care, and would enable more efficient clinical development strategies for medications to manage COPD. While many computational models have been developed to predict exacerbations, assessment of their performance in different patient cohorts has been limited.

We evaluated the performance of the ACCEPT (v 2.0) model in NOVELTY, a real world, non-interventional cohort of patients with a physician-assigned diagnosis of asthma and/or COPD. We focused on performance in predicting COPD patients who subsequently had 0, 1 or ?2 exacerbations in the year following the baseline visit. The overall performance of ACCEPT in NOVELTY is moderate (AUC=0.66), with wide variation in performance between different exacerbation categories. Accuracy of ACCEPT is high when predicting 0 exacerbations, but low when predicting 1 or ?2 exacerbations. These patterns are consistent in patients with COPD only and those with COPD and features of asthma.

ACCEPT may be most useful in identifying COPD patients who are at low risk of experiencing an exacerbation in the following year. Importantly, while ACCEPT could be useful in stratifying COPD cohorts, performance may be too low to inform individual-level treatment decisions. Alternative clinical prediction models may be necessary to more accurately predict those most at risk of multiple exacerbations.