Identifying Distinct Quitting Trajectories after an Unassisted Smoking Cessation Attempt: An Ecological Momentary Assessment Study ()
1. Introduction
The definition of relapse depends on the underlying disorder model [1]. The disease model produces a binary restriction on the possible range of outcomes. A person can either be abstinent or relapsed. In the traditional outcome literature of addiction the definition of relapse is based on the disease model (outcome-related view) [2]. This approach aims to assess who relapses or what predicts relapse [3] e.g. [4,5]. A lapse is defined as a relapse [6] and individuals are either abstinent or relapsed [7-9]. However, as seen in the relapse model, lapses do not lead to a relapse in every case [2].
The relapse model of Marlatt and Gordon has a different approach. In this model, relapse is seen as a transitional process (process-oriented view). The appearance of a lapse is viewed as a fork in the road and defined as the initial use of substance or a violation of a self-imposed rule. One of the two paths leads to the former problem level (relapse or total collapse), the other path is a continuation in the positive change [2]. Shiffman defines lapse as a limited episode of smoking [3].
To assess the relapse process multiple time points are needed. Several research groups published studies using the Ecological Momentary Assessment (EMA) to study the relapse process [10-13].
Distinct patterns of the relapse process were identified in two studies [14,15]. Furthermore, another study, which focused on the cigarette use reduction process found three distinct trajectories [16]. The goal of the first two studies was to identify subgroups in the relapse process. These studies report evidence for three [15] and five [14] quitting trajectories. Quitters, reducers and persistent smokers were identified in a sample of Chinese smokers over 6 months time. This sample (N = 402) was aged 12 to 25 years, received quit line service and smoked at least 5 cigarettes per day. On average the study participants had a mild to moderate level of nicotine dependency (M = 3.3, SD = 2.2). The reducers were the largest group (56.2%), followed by 28.9% persistent smokers and 14.9% quitters. Five distinct trajectories were identified in a sample of daily female smokers, mean age 45.5 who attended an intensive nonmedication cognitive-behavioral therapy (N = 108) over one year. On average those participants had a nicotine dependency score of 5 (SD = 1.2). 27% of the sample maintained abstinent, 8% were low-level users, 17% moderate users, 15% slow-returners, and 33% quick-returners after one year. As this approach is explorative, the examination of how the classes related to a distal outcome variable i.e. frequency of smoking in a follow-up is a good way to validate the classes. Both studies did not validate their classes further.
Based on the relapse model and prior evidence, we assume that more than two classes (abstinent/relapsed) exist among self-quitters in their relapse process. Our primary aim of this study is to identify trajectories of the relapse process in self-quitters. Furthermore we want to assess if these trajectories predict smoking frequency one, three and six months after the quit attempt.
2. Method
2.1. Participants and Procedure
Participants aged 20 - 40 years were recruited through newspapers, magazines, radio, the Internet, Facebook, our study homepage, mailing lists, and flyers. The smoking prevalence among 20 - 24-year-olds is the highest and declines afterwards [17]. All pertinent study information was available via our homepage. Individuals provided informed consent prior to participating. Inclusion criteria were: 1) smoking at least 10 cigarettes per day over a minimum of one year; and 2) intending to make an unassisted quit attempt in the next 30 days. After enrolling into the study, participants completed online questionnaires. The first questionnaire was filled in at baseline, the second one month after the quit attempt, followed by 3 and 6 months after the quit attempt. All participants received 100 Swiss Francs if they participated to the end of the study.
The data collection for the EMA combined timeand event sampling strategies. Participants used their mobile phones and were prompted to complete a short questionnaire at three random times a day. In situations where participants experienced an intense urge to smoke or smoked they were directed to download a short questionnaire (event sampling). The ambulant measurement took place over 29 days.
A total of N = 269 study participants who made a quit attempt were included. Of these 269 self-quitters, only those 230 self-quitters who had answered text messages at least for 3 days during 29 days were included into the current analysis. This is the minimum number of measurement points to estimate linear growth curves. The average number of answered text messages per participant was 20.92 (range 3 to 29).
The majority of the participants were men (76%), with a mean age of 28.0 years (SD = 5.57). Most had a Swiss citizenship (91.7%), were not married (83.9%), had experienced a previous quit attempt (83.5%) and had a low to very low nicotine dependency score (M = 3.7, SD = 1.8) (67.9%).
2.2. Measures
At baseline (T1), study participants completed an online questionnaire containing sociodemographic variables including gender, age, nationality, marital status, and smoking history such as previous quit attempt, and nicotine dependency.
In the ecological momentary assessment after the quit attempt one variable assessed the outcome variable. Have you smoked since the last text message? The response was either yes or no. If participants smoked in one of the three measurement points, then participants were given the value smoking (yes) at that day.
One, three and six months after the smoking cessation attempt, participants reported the frequency of smoking in the last month. Response categories were 1 = “never”, 2 = “one to three times a month”, 3 = “one to six times a week”, and 4 = “daily”.
2.3. Data Analysis
To identify discrete classes with similar patterns in smoking over 29 days, Latent Class Growth Model with binary outcome (LCGM) was used (Mplus 6.1.1.) [18]. Firstly, the typical trajectories of smoking cessation process were identified based on the assumption of linear relationship. Linear, quadratic and cubic models were estimated. As the quadratic and cubic term was not significant in any of the four trajectories, the linear model was applied.
To choose the number of classes, every model fit was evaluated using different criteria. The BootstrappedLikelihood Ratio Test (BLRT), the Bayesian Information Criterion (BIC), the Entropy, the average posterior probabilities for class membership, the greater drop from one to the other model in the model fit indices, and the number of participants in each class were taken into account.
Regression analysis was used to test whether class membership of the quitting trajectories predicts the frequency of smoking one, three and six months after the quit attempt.
3. Results
3.1. Descriptive Results
Of the 230 participants 18.7% (n = 43) did not smoke over 29 days of EMA. More males (14.4%) than females (4.3%) did not smoke. The majority (61.3%) of the 230 participants did not smoke on the first day after quitting; these were 18.3% female and 43% male compared to 14.8% female and 23.9% male who smoked.
3.2. Identification of Distinct Quitting Trajectories
A two-, three-, fourand five-class model was tested. Based on the goodness-of-fit indices, the theoretical background, the interpretability and the size of the classes, we preferred the four-class model as the best fitting model. As presented in Table 1 the BIC steadily decreased from twothrough five-class-model as well as the Entropy. The BLRT also slightly favored a five-class solution (four vs. five: p < 0.001). However a larger drop in all goodness-of-fit indices of the LCGM occurred from the threeto the four-class model compared to the drop from the fourto the five-class model. Furthermore in the five-class solution, the additional small class (8.1% of the participants) was characterized by the same course, but a slightly higher intercept than the quitter class. Therefore we preferred the more parsimonious four-class solution which identified distinct and meaningful trajectories that had high average probabilities for class membership (0.83 to 0.93) and also consisted of enough large grouping sizes (12.5% to 42%).
Figure 1 illustrates the four quitting trajectories. Quitters (n = 101, 43.9%) included individuals with persistent abstinence over 29 days. They had consistently low probabilities of smoking over 29 days. Participants in the late quitters class (n = 26, 11.3%) had a later onset of not smoking. This group had initially a high probability of smoking, however, the probability decreased gradually. Those in the returners class (n = 39, 17%) included individuals with increasing smoking. Those participants had an initially modest probability of smoking that increased over 29 days after the quit attempt. Finally, persistent smokers (n = 64, 27.8%) smoked over 29 days with a consistently high probability of smoking.
The quitting trajectories did not differ in terms of gender, marital status, nationality, previous quit attempt, and nicotine dependency (all p > 0.10). Only age resulted in a significant effect on quitting trajectories, F(3, 226) = 5.57, p < 0.01. Quitters (M = 26.4, SD = 4.7) were younger than persistent smokers (M = 29.5, SD = 6.0) and late quitters (M = 29.9, SD = 6.2).
3.3. Predictive Validity of Quitting Trajectories
Regression analysis1 showed that the trajectories significantly predicted smoking frequency one month, (β = .83, p < 0.001) three (β = .64, p < 0.001) and six months (β = .48, p < 0.001) after the quit attempt. The returners smoked less frequently than persistent smokers (one month: β = –.57, p < 0.001; three month: β = –.25, p < 0.05; six month: β = –.21, p = 0.09). Late quitters smoked more than quitters (one month: t1: β = .56, p < 0.001; three month: β = .21, p < 0.05; six month: β = .16, p = 0.14).
Table 2 specifies the smoking frequency in the distinct trajectory groups. Quitters have a probability of 55% of being abstinent at one month, 47% at three months and 40% at six months after the quit attempt. In persistent smokers the probability of daily smoking increased over time (66% to 80%). In the returner group most participants (96%) were occasional smokers and smoked one to six times a week one month after the quit attempt. At three and six months after the quit attempt the probability of daily smoking increased from 41.6% to 51.1% and decreased in smoking one to six times a week from 50.4% to 34.4%. Most of the late quitter (66.3%) smoked one to three times in the last month prior to the quit attempt. Three and six months after the quit attempt late quitters either smoked on a daily basis or did not smoke.