Can We Predict Who Will Experience Adverse Events While Using Smoking Cessation Pharmacotherapy? A Secondary Analysis of the EAGLES Clinical Trial

Abstract

Introduction: Concerns about potential side effects remain a barrier to uptake of Food and Drug Administration-approved smoking cessation pharmacotherapy (ie, varenicline, bupropion, nicotine replacement therapy [NRT]). However, use of pharmacotherapy can double the odds of successful quitting. Knowledge of an individual's likelihood of side effects while taking smoking cessation pharmacotherapy could influence treatment planning discussions and monitoring.

Methods: We conducted a secondary, post hoc analysis to predict an individual's likelihood of adverse events (AEs) using the Evaluating Adverse Events in a Global Smoking Cessation Study data from 4209 adults in the United States who smoked. Participants were randomized to receive 12 weeks of treatment with varenicline, bupropion, NRT patch, or placebo. Our models predicted the likelihood of moderate to severe psychiatric and nonpsychiatric AEs during treatment.

Results: Using pretreatment demographic and clinical data, multivariable logistic regression models yielded acceptable areas under the receiver operating characteristic curve for an individual's likelihood of moderate to severe (1) psychiatric AEs for bupropion and NRT and (2) nonpsychiatric AEs for varenicline and bupropion. Once we adjusted for demographic and baseline characteristics, medication was not associated with psychiatric AEs. Varenicline differed from placebo with regards to nonpsychiatric AEs.

Conclusions: It is possible to predict person-specific likelihood of moderate to severe psychiatric and nonpsychiatric AEs during smoking cessation treatment, though the probability of psychiatric AEs did not differ by medication. Future work should consider factors related to implementation in clinical settings, including determining whether lower burden assessment protocols can be equally accurate for AE prediction.

Implications: Using data from a large dataset people who smoke in the United States, it is possible to predict an individual's likelihood of psychiatric and nonpsychiatric AEs during smoking cessation treatment prior to initiating treatment. These predictive models provide a starting point for future work addressing how best to modify and integrate such clinical decision support algorithms into treatment for smoking cessation.