Predicting Chronic Obstructive Pulmonary Disease Exacerbations: When the Past Does Not Inform the Future

Exacerbations are important events in the course of chronic obstructive pulmonary disease (COPD), resulting in prolonged elevations in respiratory symptom burden. They are associated with accelerated decline in lung function, faster progression of emphysema, and high risk of mortality (1, 2). The most recent Global Initiative for Obstructive Lung Disease treatment paradigm recommends categorizing patients with COPD on initial evaluation as having either a low or a high (group E) exacerbation risk (3). The risk stratification is based on the occurrence of at least two moderate or one severe exacerbation in the prior year and is founded on the recognition that the strongest predictor of future exacerbations is a history of exacerbations within the previous 12 months (4). Several attempts have been made to develop risk assessment calculators to improve the prediction of exacerbations (5). These tools also include exacerbation history; therefore, for the goal of potentially predicting exacerbations in those without such history, they are not helpful.

In the current issue of AnnalsATS, Ferrera and colleagues (pp. 421–427) aim to identify risk factors associated with the occurrence of exacerbations in the absence of a history of recent exacerbations (6). They analyzed data from the multicenter Genetic Epidemiology of COPD (COPDGene) study and included individuals who participated in the second in-person visit of the study, had airflow obstruction on spirometry, and had no exacerbations documented in the prior 12 months. Participants were prospectively followed with surveys administered every 6 months. Using multivariable zero-inflated negative binomial regression models to account for exacerbation frequency dispersion, risk factors associated with exacerbation within the next 36 months were identified. Risk factors studied included demographic information, smoking status, lung function and symptom burden including the presence of chronic bronchitis, gastroesophageal reflux disease (GERD), cardiac disease, and blood cell counts. Models were created first using more easily available parameters and subsequently using spirometry and surveys, which are less readily available in clinical practice. The risk factors identified as significant were not surprising and included chronic bronchitis, higher symptom burden, GERD, and lower lung function. The regression coefficients from the multivariable models were summated to create simple risk scores to predict future exacerbations. The basic model, which included dyspnea, chronic bronchitis, and GERD, had a concordance index of 0.64. The addition of the COPD Assessment Test and lung function did not substantially improve the concordance index.

The identification of risk factors for future exacerbations in those without a history of recent exacerbations, and the development of a simple score to compute the risk of exacerbation, is an important addition to the literature. Nonetheless, there are several caveats. First, a distinction should be made between the prediction of exacerbations in those who have never had an exacerbation in the past versus those who have had an exacerbation-free previous 12 months. This study focuses on the latter. Identifying risk factors for predicting the first exacerbation may shed insight on how to prevent an exacerbation in the first place. Although the total absence of exacerbations is often difficult to determine, the investigators could have taken advantage of information about exacerbations from the 5 years preceding the second visit in COPDGene.

Second, risk scores were developed for predicting future exacerbations over the next 36 months. This may be a time span that is too long to be of immediate interest to both clinicians and patients. Third, a significant limitation is the heterogeneity in the use of controller medications. Admixing those on minimal or no therapy with those on triple therapy with inhaled corticosteroids, long-acting β agonists, and long-acting muscarinic antagonists introduces bias. This is somewhat attenuated by the consideration of treatment history in the secondary risk models. However, the variability in background controller medication use may limit translation of these risk scores into clinical practice. Fourth, the entry criterion mandated that all individuals should not have had any exacerbations in the prior 12 months. The fact that several participants were on significant controller therapy, including inhaled corticosteroids, suggests that these individuals may have had exacerbations in the past that were unaccounted for in these models. Indeed, those receiving dual- or triple-controller therapy had a higher proportion of individuals with any exacerbation over 36 months. The inclusion of information about remote exacerbations in these models could potentially attenuate the signals from the other risk factors identified. Fifth, this potential treatment bias also raises questions about the stability of this nonexacerbator phenotype. Even the frequent exacerbator phenotype is not a stable phenotype, and individuals who suffer from frequent exacerbations in one year may go another year with no exacerbations at all (7, 8).

Sixth, the discriminative accuracy of the risk score is modest. Consideration of a lengthier time period to include remote exacerbations may perhaps have improved the concordance index. Of note, the addition of the COPD Assessment Test and lung function information did not significantly add to the concordance index, although they do seem to add to the prediction of those who experience much higher exacerbation rates. Seventh, the reproducibility of the discrimination and calibration of the risk score should be evaluated in a replication cohort. Despite these limitations, the development of risk scores to predict exacerbations after periods of quiescence is desirable and may help target modifiable risk factors.

Several questions remain unanswered in the prevention of exacerbations. Will addressing the identified risk factors prevent a first exacerbation? In those already on controller therapies, what is the period of activity or quiescence required for either escalation or deescalation of therapy, respectively? Should the period of follow-up necessarily account for seasonality of exacerbations? What is the period of quiescence that confers a lower risk for future exacerbations? More research is needed to answer these questions. This study by Ferrera and colleagues is a step toward answering these questions.