Background and Objectives
The heterogeneity of alcohol and tobacco co-use suggests that only a subset of individuals will respond to a given pharmacotherapy. Toward identifying treatment responders, statistical learning was applied to a clinical trial combining naltrexone and varenicline for smoking cessation and drinking reduction.
Method
Individuals (N = 165) who smoke cigarettes daily and drink alcohol heavily completed a Phase 2, double blind, randomized clinical trial comparing the efficacy of combination varenicline plus naltrexone versus varenicline plus placebo. Smoking cessation was defined by bio-verified nicotine abstinence. Drinking reduction was defined as a 2-level reduction in the World Health Organization (WHO) risk drinking level. Three statistical learning methods (ridge regression, LASSO regression, and random forest) were tested psychosocial and biological predictors of clinical response.
Results
For drinking reduction, the LASSO regression had the highest overall accuracy (86%) and AUC (0.88). Important predictors included baseline alcohol consumption, baseline smoking urge, age of first cigarette use, and years of education. For nicotine abstinence, LASSO regression had the highest overall accuracy AUC (0.69). Important predictors included medication condition, expired alveolar CO level, baseline alcohol consumption, depression symptoms, and years of education.
Conclusions
Baseline consumption patterns are a strong predictor of clinical outcome for both smoking cessation and drinking reduction. Results also underscore the important cross-relationship between drinking and smoking. Statistical learning models converged with previous hypothesis-driven studies and were well-suited for clinical trial datasets.
Scientific Significance
These findings highlight candidate variables that, with further validation, may support the development of personalized treatment strategies.
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号