A Machine-Learning Approach to Predicting Smoking Cessation Treatment Outcomes

Area of Science:

• Behavioral Science
• Addiction Medicine
• Machine Learning in Healthcare

Background:

• Most smokers attempt to quit annually, but relapse rates remain high (>90%) even with evidence-based treatments.
• Identifying predictors of treatment success is essential for improving smoking cessation outcomes and personalizing care.
• Current treatments, including cognitive-behavioral therapy (CBT), have limited long-term efficacy for many individuals.

Purpose of the Study:

• To identify baseline predictors of successful smoking cessation following group cognitive-behavioral therapy (CBT).
• To develop a predictive model using machine learning to forecast treatment outcomes.
• To inform personalized treatment strategies for tobacco dependence.

Main Methods:

• Two cohorts of smokers (N=90 training, N=71 validation) participating in group CBT were analyzed.
• Generalized estimating equations identified baseline predictors of cessation, measured by carbon monoxide and cotinine levels.
• Decision tree analysis was employed to predict quit status, with delay discounting identified as a key variable.

Main Results:

• Decision trees significantly improved prediction of smoking status post-treatment and at 6-month follow-up compared to chance.
• Delay discounting emerged as the single best predictor of group CBT treatment response.
• This predictor achieved 80% accuracy at post-treatment and 81% at follow-up in the training cohort.

Conclusions:

• This study represents a significant step towards personalized smoking cessation care.
• Delay discounting is a promising biomarker for predicting success in CBT for smoking cessation.
• Further research is needed to validate these findings and extend personalized approaches to other cessation interventions.