Dynamic Process of Health on Smoking-Cessation-Relapse Embedded with Transtheoretical Behavior Changes Using Markov Model

Background Numerous epidemiological studies have been conducted to study risk factors (including demographic features, betel quid chewing, alcohol drinking, co-morbidity, and health behavior) affecting the forces of being regular smoker (habitual use), quitting, and relapse separately, but very few studies have been proposed to elucidate three transitions (non-smoker regular smoker, regular smoker quitting, quitting relapse) with a dynamic process modelled by consolidating three processes as an unified framework that treats three processes as correlated outcomes to estimate the net force of being habitual use by each individual factor and a constellation of these factors. In addition to considering this dynamic process, what is the most lacking in such kind of epidemiological studies is the failure of considering those epidemiological correlates and smoking behavior particularly based on certain behavior theory (such as transtheoretical model, TTM) simultaneously. The application of statistical models to build up a dynamic process characterized by stage-dependent correlates that consist of both epidemiological and behavior change variables is worthy of being investigated. Aims The objectives of this thesis were in achieving statistical goal to (1) estimate the effects of demographic features, betel quid chewing, alcohol drinking, co-morbidity (diabetes mellitus and hypertension), and health behavior on three independent processes of smoking habits with conventional Cox proportional hazards regression models; (2) estimate the effects of all each of epidemiological factors indicated in the aim of (1) on a dynamic process (composed of three consecutive transitions, non-smoker regular smoker quitting; the reverse arrow represents the relapse) using a three-state Markov transition model ; (3) estimate the effects of all each of epidemiological factors indicated in the aim of (1) and relevant correlates (such as smoking within 30 minutes after wake-up and age at the commencement of smoking) on the stage of change of TTM (pre-contemplation contemplation preparation action) using a four-state Markov transition model; (4) estimate the effects of all each of epidemiological factors indicated in the aim of (1) on a dynamic process (composed of three consecutive transitions, non-smoker regular smoker quitting; the reverse arrow represents the relapse) using a three-state Markov transition model embedded with a four-state Markov model on cyclic four-state transition; and were in the province of smoking habit and behavior to (5) estimate the net force of being habitual use after the balance between the force of entering habitual use from non-smoker and the force of relapse from quitter by considering epidemiological factors indicated in the aim of (1) ; (6) estimate the net force of TTM-based stage of change (pre-contemplation contemplation preparation action) considering relevant correlates (such as smoking < 30 minutes after wake-up and age at the commencement of smoking); (7) estimate the net force of being habitual use after the balance between the force of entering habitual use from non-smoker and the force of relapse from quitter by considering the embedded probability of being TTM-based stage of change with respect to action or pre-contemplation epidemiological factors indicated in the aim of (1); Materials and Methods Study subjects of this thesis were derived from two Community-based Integrated Screening (CIS) programs from Keelung City and Changhua County. The two programs have been launched since 2000 and 2005 for Keelung and Changhua, respectively. A total of 228,258 subjects attended the CIS programs between 2000 and 2015, including 127,194 from Keelung and 101,064 from Changhua. Each participant completed a self-administered questionnaire to collect data on cigarette smoking status, betel-quid chewing, alcohol drinking habits, place of residence, physical activity, comorbidity, and health check-up experience. Data on comorbidity were ascertained by the on-site screening and self-reported status. The study population was categorized into three groups by smoking status: non- smokers, habitual use, and quitting. The habitual use was defined as those who regularly smoked at least 1 cigarette per week. We further recorded their age of smoking commencement, cigarettes per day and duration of being smoker. Those who have quit smoking longer than half year were defined as abstention. The period of having been a quitter was also collected. The questionnaire also collected data on smoker’s readiness to quit, which enables us to define the stage of change underpinning the transtheoretical model (TTM). We firstly estimated the effects of demographic features, betel quid chewing, alcohol drinking, comorbidity, and health behavior on three independent processes of smoking habits (nonsmoker habitual use, habitual use quitting, and quitting habitual use) with Cox proportional hazards regression models. A three-state Markov model was further proposed to estimate the state-specific effects of relevant risk factors on a dynamic process (composed of three consecutive transitions, non-smoker habitual use quitting). For the behavior change process, we used a four-state Markov transition model underpinning the TTM (pre-contemplation contemplation preparation action) to elucidate the state-specific effects of risk factors. The TTM-underpinned Markov model was then embedded in the proposed three-state Markov for smoking habit model to elucidate the net force of being habitual use after the balance between the force of quitting and the force of relapse by considering the embedded probability of being the stage of change of TTM with respect to action or pre-contemplation. Results The prevalence of habitual user and quitter were 17.7% and 6.9%, respectively. The smoking rates show significant gender differences in habitual use and quitting, 39.3% and 16.6% for males and 5.3% and 1.3% for females, respectively. Our results demonstrated the higher incidence rates of habitual use among the male, with lower education level, unmarried/divorced, quit/current betel-quid chewing or alcohol drinking. Using conventional Cox proportional hazards regression model, the aggravating factors responsible for habitual use were male, middle and low education, unmarried, divorced/widowed, quitting/current betel-quid chewer, alcohol drinking. The protective health-related behavior factors were consisting of regular exercise and experience of health check-up. For the factors responsible for quitting, the significant promoting factors included old age, male, regular exercise, and the uptake of health check-up experience. The significant factors preventing one from quitting were middle and lower education level and quitting/current betel-quid chewing. The significant factors responsible for relapse were lower education level, unmarried and current betel-quid chewing. Protective factors were regular exercise and male. The monthly rate of turning into habitual use, quitting, and relapse were estimated as 0.00027 (95% CI: 0.00025-0.00028), 0.0073 (95% CI: 0.0071-0.0042), and 0.0040 (95% CI: 0.0038-0.0042), respectively based on the dynamic Markov model. The five year probabilities of quitting for habitual users and relapse were estimated as 32% and 18%, respectively. In the end of statistical model, 35% of the population will be habitual uses and 65% be quitter. By looking at net force of the balance between three transitions, betel-quid chewing plays a major role toward habitual use with the drift estimated as 1.87 and 0.59 for current chewer and quitter, corresponding to the ORs of 6.5 and 1.8. Regular exercise revealed a negative drift of -0.87 toward habitual use with the corresponding risk reduction of 60%. The baseline TTM stage of action was significantly associated with a 73% increase of quitting and 68% reduction of relapse. The estimated net force toward habitual use of contemplation, preparation, and action were estimated as -0.41, -0.20, and -1.69, corresponding to the risk reductions of 34%, 18%, and 81%, respectively. As regards the factors associated with the stage of change of TTM, we found that smoking initiation before the age of 20 (OR=0.61, 95%CI:(0.30,1.23)) and first cigarette within 30 mins after waking up (OR=0.49, 95%CI:(0.26,0.94)) were less likely to go toward action stage from pre-contemplation stage in multivariate logistic regression analysis. The results based on the four-state Markov model were similar. First cigarette within 30 mins after waking up made significant contribution to the relapse from action to pre-contemplation of TTM stage. The results of adding the embedded four-state Markov model of the TTM to the dynamic three-state Markov model of smoking habit shows the probability of being in the action stage had a negative net force of being habitual use (net regression coefficient: -5.55, 95% CI: -9.83, 0.00) considering the balance between the force of departing from habitual use to quitting and of relapsing from quitting to smoking, whereas the probability of being in the pre-contemplation led to a positive net effect of being habitual use (net regression coefficient: 8.10, 95% CI: -5.30, 9.87). Conclusions This thesis has successfully demonstrated how to apply two Markov processes to modelling the corresponding two dynamic processes of smoking habits (non-smoker regular smoker abstention) and smoking behavior (pre-contemplation contemplation preparation action) by using two community-based integrated screening data with the conclusions drawn from this thesis on methodological improvements are three, including (1) dynamic process of smoking habit with the use of three-state Markov model is better than three independent processes with Cox proportional hazards regression model with respect to the precision of effect size on state-specific risk factors and dynamic transition between states; (2) the better use of three-state Markov process to estimate the net force of three transitions for each state-specific risk factor with adjustment for other confounding factors; (3) the better use of four-state Markov process to estimate the net force of four forward and backward transitions for two important correlates of smoking behavior; (4) the integration of the Markov process pertaining to the dynamic process of TMM behavior into the Markov process of smoking habits on regular smoker, quitting, and relapse; and those drawn from this thesis on empirical aspect of smoking habits and behavior include the respective findings as follows. (5) The largest net force of being habitual use (regular smoker) among these state-specific epidemiological factors was betel quid chewing; (6) significant state-specific factors for net force of being habitual use included young age, male, low education level, unmarried, current betel chewing, alcohol drinking, DM, no hypertension, lacking of regular exercise, and the failure of the uptake of health check-up; (7) epidemiological characteristics directing the change from pre-contemplation to action included old age, males, regular exercise, DM, hypertension, non-chewer, the quitter for drinking whereas only betel quid chewer play the most important role in the direction of the change from action to pre-contemplation; (8) Smoking behavior of first cigarette within 30 minutes when waking up made significant contribution to the resistance to the change from pre-contemplation to action and the drift toward pre-contemplation; (9) stage of change of TTM, particularly pre-contemplation, is still an independent predictor for the net force of being habitual use after controlling for all epidemiological risk factors using the three-state Markov model; smoking-behavior-adjusted transition probability of remaining in pre-contemplation or that from pre-contemplation to action derived from the embedded Markov process on the stage change of TMM behavior made independent significant contribution to the net force of being habitual use derived from three-state Markov process of smoking habits