Analysis of Smoking Cessation Patterns Using a Stochastic Mixed Effects Model with a Latent Cured State

Smoking is a major cause of a large number of diseases, e.g. cancers of the lung, larynx, and pharynx, etc. The slow reduction of adult smoking prevalence is partly due to high rates of relapse following quit attempts among smokers. A major problem when studying smoking addiction behavior is that participants make several quit attempts before successfully quit. For efficient development, targeting and evaluation of interventions, it is necessary to distinguish transient quitting (temporarily smoking-free but relapse later) from permanent quitting (lifelong smoking-free). We identified and quantified baseline factors associated with permanent quitting using the Alpha-Tocopherol Beta-Carotene (ATBC) Lung Cancer Prevention study dataset (a longitudinal cohort study with 29133 subjects). We modeled the smoking cessation patterns using the discrete-time stochastic mixed-effect model with three states: smoking, transient quitting and permanent quitting. We also designed computationally practical methods for dealing with the size of the data set and complexity of the models. We found that age was positively associated with probability of making quit attempts (p<0.001). However, years of smoking, cigarette and alcohol consumption had inverse association with probability of making quit attempt (p<0.001). If the quit attempt was made, more alcohol consumption per day was associated with higher probability of relapsing (p=0.005). Moreover, 5 years in age increased the odds of permanent quitting by 10.2% (CI -0.40%~21.8%; p=0.06). Individuals with psychological symptoms were significantly less likely to be successful permanent quitters (p=0.03). Thus, baseline risk factors had different effects on different transition probabilities.