Positive Affect as a Predictor of Smoking Cessation and Relapse: Does It Offer Unique Predictive Value Among Depressive Symptom Domains?

Abstract

Background:

Prior studies have suggested that, among the domains of depressive symptoms, low positive affect (PA) may have a distinct relationship with smoking cessation and relapse. However, the empirical basis for PA-focused interventions cessation is limited, with some mixed findings.

Objectives:

Using a large, diverse sample of treatment-seeking smokers, this study tested the hypothesis that PA adds unique predictive value beyond the effects of the other symptom domains in models of cessation and relapse.

Methods:

Adult smokers participating in a smoking cessation trial (n=450) were included in this post hoc analysis. Cessation outcomes included smoking abstinence at end of treatment and at 6-month follow-up. Relapse was defined as recurrence of smoking at 6-month follow-up among the end-of-treatment abstainers. Depressive symptoms were assessed at baseline using the Center for Epidemiologic Studies-Depression (CES-D) scale.

Results:

With the exception of PA, all of the CES-D domains predicted reduced likelihood of smoking abstinence at end of treatment and cotinine-confirmed (but not self-reported) abstinence at 6 months, as did total CES-D score (all p-values < .05). None of the symptom domains predicted relapse.

Conclusions/Importance:

Our results provide further evidence that current depressive symptoms predict worse cessation outcomes, but they fail to support recent work suggesting that low PA has incremental predictive value for cessation or relapse beyond the other depressive symptom domains. To improve quit rates for smokers with depressive symptoms, evidence-based mood management interventions should be included in treatment planning.

1. INTRODUCTION

Compared with the general population, the prevalence of cigarette smoking is two to three times higher among people with mood disorders (Goodwin, Zvolensky, Keyes, & Hasin, 2012; Lasser et al., 2000), the most common category of mental health conditions among US adults (Kessler, Chiu, Demler, Merikangas, & Walters, 2005). Further, the presence of depressive symptoms at the time of a smoking cessation attempt, even at levels below the diagnostic threshold for mood disorders, has been found to predict difficulty quitting smoking or relapse in a multitude of studies (Brodbeck, Bachmann, Brown, & Znoj, 2014; Cooper, Borland, McKee, Yong, & Dugue, 2016; Hitsman et al., 2013; Piper et al., 2010). However, as reported in a recent systematic review (Weinberger, Mazure, Morlett, & McKee, 2013), other studies have found no effect of depressive symptoms on cessation outcomes or have reported better outcomes among smokers with higher levels of depression, and these contradictory findings are not explained by methodological differences such as treatment approach (i.e., pharmacotherapy vs. behavioral intervention) or sample size. To better understand the mixed findings, recent research has begun to examine the heterogeneity of depressive symptom phenomenology as a possible explanation.

Several lines of evidence suggest that the affective symptoms of depression reflect distinct negative and positive affective components. Negative affect (NA) is a general distress factor describing aversive internal states such as anxiety, depression, or anger; positive affect (PA) refers to pleasurable internal states such as enthusiasm or energy (Watson, 1988). Although correlated, these components are independent and have unique patterns of relationships with other dimensions of mental health symptoms (Clark & Watson, 1991). Consequently, it is reasonable to hypothesize that NA and PA would also have distinct relationships with smoking cessation.

NA, including depressed mood, has been studied extensively in relation to smoking cessation and has demonstrated a robust relationship with initiating and sustaining abstinence (Berlin & Covey, 2006; Kenford et al., 2002; Shiffman & Waters, 2004). There has been substantially less focus on PA as a predictor of smoking cessation and relapse (Castro et al., 2014), but existing evidence suggests that low PA and the related construct of anhedonia—the inability to experience positive affect in anticipation of or in response to pleasurable stimuli—also predicts lower likelihood of quitting. Specifically, the occurrence of low PA or anhedonia during one’s lifetime (Leventhal, Piper, Japuntich, Baker, & Cook, 2014), at the time of a cessation attempt (Doran et al., 2006; Leventhal, Ramsey, Brown, LaChance, & Kahler, 2008), or during the immediate post-cessation period (J. W. Cook et al., 2015) has been found to predict poorer smoking cessation outcomes, including failure to initiate smoking abstinence and increased likelihood of relapse. Additionally, among all of the depressive symptom domains measured in the Center for Epidemiologic Studies Depression (CES-D; (Radloff 1977)) scale—one of the most commonly used self-report measures of depressive symptoms—low PA is the only domain that has incrementally predicted smoking cessation outcomes (Leventhal et al., 2008). That is, while the domains of low PA, depressed affect (e.g., feelings of sadness or hopelessness), somatic activity (e.g., problems with sleep and appetite), and interpersonal difficulties (e.g., feeling disliked or that others are unfriendly) all predicted cessation when considered independently, only low PA emerged as a predictor of cessation when the effects of the other domains were held constant (Leventhal et al., 2008). Findings from this emerging area of research have led to the conclusion that low PA may be a critical factor in the relationship between depression and smoking and that targeted cessation interventions could improve outcomes by increasing PA (Kahler et al., 2015; Leventhal et al., 2014; Leventhal et al., 2008).

The earliest studies supporting the critical role of low PA as a predictor of smoking cessation, either globally or in the context of anhedonia, were limited by their exclusive focus on: (a) special population samples, including heavy drinkers (Leventhal et al., 2008) or major depressive disorder (MDD) history-positive (J. Cook, Spring, McChargue, & Doran, 2010) smokers, or (b) smokers receiving antidepressant pharmacotherapy (Doran et al., 2006), limiting generalizability. Recent studies that do not have these characteristics have been more mixed, with some studies supporting the relationship between low PA or anhedonia and smoking cessation (Castro et al., 2014; Leventhal et al., 2014) and others failing to demonstrate a relationship (Schnoll, Leone, & Hitsman, 2013) or showing that anhedonia predicted better cessation outcomes (Powers et al., 2016). Consequently, additional evaluation of the predictive value of PA or smoking cessation is warranted, including a replication of the finding that low PA offers unique predictive value beyond the broader construct of depression. This work is necessary to determine whether the relationship between PA and cessation outcomes is robust to differences in study methodology, characteristics of the sample, and the method(s) of cessation.

To address this need, we conducted a post hoc analysis of data from a clinical trial of two group-delivered smoking cessation interventions for adult daily smokers (n=450). Based on prior results linking low PA or anhedonia to difficulty both initiating and maintaining smoking abstinence, our specific hypothesis was that low PA, as well as the other domains of depressive symptoms, would predict, in a univariate model, (a) lower likelihood of abstinence at the end of smoking cessation treatment, (b) lower likelihood of abstinence at 6-month follow-up, and (c) post-treatment relapse among those who are able to achieve abstinence during treatment. Moreover, given previous findings of the unique predictive value of low PA among all of the domains of depression (Leventhal et al., 2008), we hypothesized that only low PA would remain a significant predictor in multivariable models that included all of the other depressive symptom domains.

2. MATERIALS AND METHODS

2.1. Participants

Participants in these analyses (n=450) are the complete sample of participants enrolled in a clinical trial (NCT# 01533974) comparing the effectiveness of two group-delivered smoking cessation interventions provided in combination with nicotine replacement therapy (NRT). Neither treatment specifically targeted depressive symptoms. Eligibility criteria for the trial were: (1) age 18 and older; (2) smoke at least 10 cigarettes per day for every day in the past 12 months; (3) want to quit smoking in the next 30 days; (4) able to speak and read in English; (5) enrolled as a member of Group Health (GH), a large Washington-based nonprofit healthcare organization, and living in Western Washington; (6) not currently participating in other smoking cessation interventions, including behavioral treatment or pharmacotherapy; (7) not currently using other tobacco products (e.g., smokeless tobacco); (8) willing to attend five 90-minute group intervention sessions and to receive NRT over 3 months; (9) no medical contraindications for NRT use (i.e., pregnant, breastfeeding, recent heart attack, or skin allergy preventing use of the patch, as assessed using the standard GH contraindication screening for NRT patch); (10) no significant cognitive or physical impairment that would preclude full participation in the counseling sessions.

The average age of participants was 51.3 years (SD=12.1), and 53% were female. The majority were Caucasian (83%), with smaller proportions identifying as African American (6%) or more than one race (7%) and Hispanic (4%). Slightly more than half (56%) were married, and most were employed (72%) and had greater than a high school level of education (78%). In terms of smoking behavior, the majority of participants (75%) smoked a pack a day or less and had been smoking for at least ten years (88%). The average number of quit attempts in the past 12 months was 1.9 (SD=6.4). The average CES-D score at baseline was 11.9 (9.2), which is below the threshold of ≥ 16 for elevated depressive symptoms (Radloff 1977).

2.2. Measures

2.2.1. Depressive symptom domains.

The Center for Epidemiologic Studies-Depression scale (CES-D; (Radloff 1977)) was administered as part of the baseline survey. For each item, responses range from 0 to 3, where 0=“Rarely or none of the time” and 3=“Most or all of the time,” with higher scores (out of a total score of 60) indicating higher levels of depression symptoms. Previous factor analyses of the CES-D support a 4-factor solution consisting of the following domains: Depressed Affect (7 items), Positive Affect (4 items), Somatic Activity (7 items), and Interpersonal Difficulties (2 items) (Radloff 1977). Consistent with prior investigations (Leventhal et al., 2008), we used the mean item scores for the total CES-D and the factor scores in the analyses.

2.2.2. Smoking cessation outcomes.

The three smoking outcomes were (1) smoking abstinence at end of treatment (n=450 included in analyses), (2) smoking abstinence at 6-month follow-up (n=450), and (2) relapse by 6 months among those who were able to achieve end-of-treatment abstinence (n=162). Abstinence at end of treatment (assessed at 1-week post-treatment) and at 6-month follow-up was defined as 7-day point prevalence abstinence (PPA). Point prevalence abstinence (PPA) was assessed using the following question: “When did you last smoke?” (where 7-day PPA= response of “8–30 days ago” or “Over 31 days ago”). Relapse amongst those who indicated abstinence at end of treatment was defined as a report of any smoking in the 30 days prior to the 6-month follow-up evaluation. Because self-reported smoking status was the primary method of assessing efficacy in the parent trial and there was no biochemical verification of abstinence at end of treatment, our primary abstinence outcomes were based on self-report. However, as a sensitivity analysis, we also report cotinine-confirmed smoking outcomes for 7-day PPA at 6 months as well as for relapse between end of treatment and 6-month follow-up.

2.3. Procedures

Electronic medical record data (e.g., based on age, smoking status, primary care clinic location) were used to identify potentially eligible smokers, all of whom lived in Western Washington State and were members of Group Health (GH). Study staff mailed these potential participants a study invitation letter and followed up by phone to invite them to be screened for study eligibility. Persons screened as potentially eligible by phone were invited to attend an in-person orientation session at a local GH clinic. At the orientation, a research specialist reviewed the study objectives and activities, answered questions, obtained written informed consent, asked participants to complete an eligibility survey, and asked eligible participants to complete a baseline survey. Eligible individuals who decided to participate were then randomized to treatment, which consisted of 5 weekly sessions of group-delivered smoking cessation counseling along with eight weeks of nicotine patch. The quit date was not set until the first group session in both treatment arms. Consistent with the differing approaches in the two arms, the CBT group was encouraged to set a date that occurred between Sessions 2 and 3 (Weeks 2–3) and the ACT group was encouraged to set a quit date before the end of the 5-week treatment period, but with ACT-consistent flexibility within that time. Follow-up surveys of treatment response were conducted at one week post-treatment and six months post-randomization. At the 6-month follow-up, participants who reported abstinence from smoking were mailed a saliva cotinine kit to return to the researchers.

The purpose of the parent study was to compare the efficacy of a counseling protocol based on acceptance and commitment therapy (ACT) to a protocol based on traditional cognitive-behavioral therapy (CBT). Long-term follow-up data are currently being analyzed, and the main study outcomes will be presented in a future manuscript. All study procedures were reviewed and approved by the Institutional Review Boards of the Fred Hutchinson Cancer Research Center and Group Health.

2.4. Statistical analyses

We used confirmatory factor analysis to investigate how well the 4-factor model for the CES-D (Radloff 1977) fit the sample of study participants. Support for distinguishing between the four subscales--Depressed Affect, Somatic Activity, Positive Affect, and Interpersonal Difficulties--would be supported if an oblique 4-factor model outperformed the 1-factor model defined by all 20 CES-D items. We investigated additional measures of model fit using cutoff criteria proposed by Hu and Bentler (Hu & Bentler, 1999), including the Comparative Fit Index (CFI), Tucker-Lewis Index (TLI), root mean squared error of approximation (RMSEA), and the maximum likelihood-based standardized root mean squared residual (SRMR). Internal consistency of the 20-item CES-D and each of the four subscales was estimated with Cronbach’s alpha.

Factor scores used in subsequent analyses were calculated using the average of the items within each factor. Models using smoking abstinence and relapse as the outcomes were fit using the generalized linear mixed effect model with a binomial error distribution and logit link function. These models were adjusted for treatment group assignment as well as any baseline demographic covariates that could be confounders. Age, gender, education, marital status, and employment status are associated with both depression (Akhtar-Danesh & Landeen, 2007; Pratt & Brody, 2014) and smoking cessation (Kale, Gilbert, & Sutton, 2015; Meamar et al., 2013; Schuck, Otten, Kleinjan, Bricker, & Engels, 2014) and demonstrated univariate relationships with depressive symptoms scores at the p<.10 level, so they were included in the models as covariates... The models were also fit without these demographic covariates as a sensitivity analysis. Although neither intervention targeted depressive symptoms directly, ACT is an efficacious treatment for depression (Pots et al., 2016; Walser et al., 2015) and therefore could affect the relationship between baseline depressive symptoms and cessation or relapse outcomes. Consequently, we preliminarily examined interactions between each CES-D subscale and treatment group assignment for each smoking outcome to determine whether results should be stratified by treatment group. Since none of these interactions were statistically significant (all p-values >.10), the two treatment groups were combined for analysis. A random effects term was included to account for intra-group nested within-facilitator variability. Analyses were intent-to-treat, and participants with missing follow-up data were considered to be smokers. For the analyses of cotinine-confirmed abstinence, any participants who self-reported abstinence but who did not return cotinine samples were coded as smokers.

All statistical analyses were done using the R programming language (Team, 2013). Confirmatory factor analysis was done using the R extension lavaan for structural equation models (Rosseel, 2012). Generalized linear models were fit using the R extension lme4 (Bates, 2015, e-print, in press). All statistical tests were two-sided, and significance was determined at α = 0.05.

3. RESULTS

3.1. Concordance of self-reported and cotinine-verified abstinence at 6-month follow-up

Sixty-five of 85 (76%) self-reported quitters returned a saliva cotinine sample. Of the returned samples, 12 (18%) were flagged as indicative of smoking.

3.2. CES-D confirmatory factor analysis

The confirmatory factor analysis indicated that the 4-factor model fit the data significantly better than the undifferentiated 1-factor model ${{\chi }^2}_{difference}$ (6)=300.1, p < 0.0001. The measures of model fit for the 4-factor model were as follows: CFI = 0.92, TLI = 0.91, RMSEA = 0.06, SRMR = 0.05, which provided reasonable support for the use of the four subscales in our study. The estimates of internal consistency for each of the four subscales were as follows: Depressed Affect (α = 0.88), Somatic Activity (α = 0.71), Positive Affect (α = 0.76), and Interpersonal Difficulties (α = 0.63). The internal consistency of the total CES-D scale was very good (α = 0.90). Table 1 contains the means and standard deviations of the total and subscale scores of the CES-D as well as the correlations among them, which were in the moderate to large range.

Table 1.

Descriptive and correlational data on CES-D symptom total and domain scores.

	Mean (SD)	1	2	3	4	5
1. Total Score	0.60 (0.46)	--	0.755	0.911	0.832	0.635
2. Positive Affect	0.83 (0.73)	0.755	--	0.566	0.429	0.362
3. Depressed Affect	0.46 (0.55)	0.911	0.566	--	0.667	0.587
4. Somatic Activity	0.69 (0.48)	0.832	0.429	0.667	--	0.434
5. Interpersonal Difficulties	0.29 (0.52)	0.635	0.362	0.587	0.434	--

Note: CES-D=Center for Epidemiologic Studies Depression Scale. Mean (SD) scores are an average of item scores for which response options are scored on a 0–3 scale.

3.3. Depressive symptoms and cessation outcomes

3.3.1. End-of-treatment abstinence.

At 1-week post-treatment, 36% of the sample met criteria for 7-day PPA. Table 2 shows the results of univariate predictive models for the depressive symptom domains, including both self-reported (primary) and cotinine-confirmed (secondary, when available) abstinence as outcomes. After adjusting for treatment group assignment and baseline covariates that were related to depressive symptoms (i.e., age, gender, education, marital status, and employment status), higher total baseline CES-D score predicted lower success at quitting (p=.002). Examining the four factors separately in univariate models, Depressed Affect (p=.002), Somatic Activity (p=.013), and Interpersonal Difficulties (p=.012) were all associated with lower odds of abstinence, whereas Positive Affect (p=.057) did not reach the level of statistical significance. Point estimates of the effects of the three factors associated with quitting suggest that one-point increases in the average scores of these subscales were associated with 45–51% decreases in the odds of quitting. In unadjusted models, the CES-D total score and all four factors were significantly associated with quitting (see Table 2).

Table 2.

Depressive symptom domains as predictors of smoking cessation and relapse.

	Self-reported abstinence, unadjusted1 models		Self-reported abstinence, adjusted2 models		Cotinine-confirmed abstinence,3 unadjusted1 models		Cotinine-confirmed abstinence,3 adjusted2 models
	p	OR (95% CI)	p	OR (95% CI)	p	OR (95% CI)	p	OR (95% CI)
7-day PPA at end of treatment (n=450)
CES-D Total	.0002	0.95 (0.93,0.98)	.002	0.43 (0.25, 0.73)
Positive Affect	.020	0.72 (0.54, 0.95)	.057	0.76 (0.57, 1.01)
Depressed Affect	.0003	0.45 (0.30, 0.69)	.002	0.49 (0.31, 0.77)
Somatic Activity	.001	0.49 (0.32, 0.76)	.013	0.55 (0.34, 0.88)
Interpersonal Difficulties	.003	0.50 (0.32, 0.79)	.012	0.54 (0.34, 0.87)
7-day PPA at 6-month follow-up (n=450)
CES-D Total	.052	0.97 (0.95, 1.00)	.283	0.73 (0.42, 1.29)	.004	0.94 (0.91,0.98)	.025	0.40 (0.18, 0.89)
Positive Affect	.205	0.82 (0.60, 1.12)	.413	0.87 (0.63, 1.21)	.226	0.78 (0.53, 1.16)	.455	0.86 (0.57, 1.28)
Depressed Affect	.129	0.71 (0.46, 1.10)	.524	0.86 (0.54, 1.37)	.011	0.41 (0.21, 0.82)	.054	0.50 (0.25, 1.01)
Somatic Activity	.044	0.60 (0.37, 0.99)	.232	0.73 (0.43, 1.23)	.004	0.38 (0.19, 0.74)	.026	0.45 (0.22, 0.91)
Interpersonal Difficulties	.161	0.71 (0.44, 1.15)	.470	0.83 (0.51, 1.37)	.002	0.19 (0.07, 0.55)	.005	0.21 (0.07, 0.62)
Relapse by 6-month follow-up (n=162)
CES-D Total	.705	0.99 (0.95,1.03)	.255	0.58 (0.23, 1.49)	.168	1.04 (0.98, 1.09)	.385	1.65 (0.53, 5.15)
Positive Affect	.427	0.83 (0.53, 1.30)	.288	0.77 (0.48, 1.24)	.960	0.99 (0.59, 1.64)	.780	0.93 (0.54, 1.58)
Depressed Affect	.437	0.75 (0.35, 1.56)	.117	0.49 (0.20, 1.19)	.267	1.73 (0.66, 4.53)	.581	1.34 (0.48, 3.74)
Somatic Activity	.662	1.17 (0.58, 2.36)	.822	0.91 (0.42, 2.01)	.100	2.08 (0.87, 4.97)	.230	1.80 (0.69, 4.69)
Interpersonal Difficulties	.767	1.12 (0.52, 2.42)	.888	0.94 (0.41, 2.17)	.019	9.07 (1.45, 56.90)	.029	8.66 (1.25, 59.80)

Note: PPA=point prevalence abstinence. CES-D=Center for Epidemiologic Studies Depression Scale.

¹Unadjusted models controlled for treatment group assignment only.

²Adjusted models controlled for treatment group assignment plus age, gender, education, marital status, and employment status.

³Biochemical verification of abstinence was not conducted at end of treatment.

Because Positive Affect was close to the level of statistical significance (p=.057) as a univariate predictor of end-of-treatment abstinence in the adjusted model, we tested its incremental predictive value using a regression model containing all four of the CES-D factors. The p-value for the likelihood ratio test comparing the full multivariable model to a null model was 0.002, indicating that the full model explained a significant amount of the variability in the data. However, none of the individual depressive symptom domains—including PA—were significant predictors of cessation in this full model.

3.3.2. Abstinence at 6-month follow-up.

At 6-month follow-up, 27% of the sample met criteria for 7-day point prevalence abstinence. Using the primary outcome measure of self-reported smoking abstinence, none of the CES-D factors were significant predictors in the adjusted model. In the unadjusted model, the CES-D total (p=.052) and Somatic Activity factor (p=.044) were significant predictors. Using the secondary outcome of cotinine-confirmed abstinence, the CES-D Total score (p=.025) as well as the domains of Somatic Activity (p=.026), Interpersonal Difficulties (p=.005), and Depressed Affect (p=.054) predicted cessation in the adjusted models, whereas Positive Affect (p=.455) did not. The unadjusted models showed the same pattern of results. Because PA was not associated with 6-month cessation outcomes in the adjusted analyses using either the self-reported or cotinine-confirmed abstinence outcomes, no multivariable modeling was done to test its incremental value as a predictor.

3.3.3. Relapse.

Of those who achieved 7-day PPA at 1-week post treatment, 61% had relapsed at the 6-month follow-up. Baseline CES-D total score was not a significant predictor (p=.255) of relapse based on self-reported smoking status, nor were any of the individual depressive symptom domains (all p-values > .05; see Table 2). On the secondary definition of relapse (using cotinine-confirmed abstinence), the Interpersonal Difficulties domain was the only significant predictor (p=.029). Results of the unadjusted models yielded the same conclusions. Consequently, regression modeling with multiple predictors was not conducted as a test of PA’s incremental predictive value.

3.3.4. Post hoc analyses: Depressive symptoms and discordance in self-reported vs. cotinine confirmed abstinence.

Because we observed differences in the relationship between depressive symptoms and smoking outcomes by self-report vs. cotinine-confirmed abstinence, we conducted additional analyses to evaluate possible causes for the discrepancy. Specifically, we tested two hypotheses regarding causes for discordant findings that would have resulted in a subset of self-reported abstainers being reclassified as smokers: (a) depressive symptoms predict failure to return a cotinine sample, and (b) depressive symptoms predict elevated cotinine results indicative of smoking. We were unable to evaluate the hypothesis that failure to return a cotinine sample was associated with depressive symptoms due to small sample size and resultant failure of predictive models to converge. However, increased scores on CES-D Total (p=.029), Depressed Affect (p=.034), and Interpersonal Difficulties (p=.020) were all associated with reclassification from a self-reported abstainer to a smoker as a result of elevated cotinine levels.

4. DISCUSSION

With the exception of PA, all of the depressive symptom domains, as well as the total symptom score, were associated with lower odds of stopping smoking by the end of treatment in a large generalizable sample of treatment-seeking smokers. These same domains were also predictive of cotinine-confirmed abstinence, but not self-reported abstinence, at 6-month follow-up. This finding is consistent with those of a multitude of studies demonstrating that current total depressive symptoms are a risk factor for treatment failure (Brodbeck et al., 2014; Hitsman et al., 2013; Piper et al., 2010). Thus, screening smokers for elevated depressive symptoms at the time of the quit attempt will identify individuals who may have less favorable outcomes. For these individuals, treatment that includes a mood management component has been demonstrated to improve quit rates (van der Meer, Willemsen, Smit, & Cuijpers, 2013; van der Meer, Willemsen, Smit, Cuijpers, & Schippers, 2010).

Contrary to our hypotheses, PA did not predict cessation at any time point after controlling for the potential confounds of age, gender, education, marital status, and employment status. These findings are consistent with those of Schnoll and colleagues (Schnoll et al., 2013), who did not find a relationship between PA and cessation. However, that study was small (n=87) and likely underpowered, with none of the CES-D factors demonstrating an association with quitting. Using a larger sample with greater power to detect associations with quitting and relapse, we obtained results that diverge from those reported in the prior studies that have examined the relationship between low PA and smoking cessation (Castro et al., 2014; Leventhal et al., 2008) or the related construct of anhedonia and cessation (J. Cook et al., 2010; Leventhal et al., 2014). These divergent findings highlight the importance of context, as they may stem from differences in study design or population characteristics (e.g., age, recency or severity of depressive symptoms, co-occurring heavy drinking) or from mood context (e.g., global low PA vs. low PA in response to actual or anticipated reward, consistent with anhedonia). It may be that PA in the context of anhedonia, rather than global PA, is most relevant to cessation. Given the early state of this literature, it is not possible to detect how contextual variables might impact the relationship between low PA and cessation outcomes. Regardless, discordant findings may point to generalizability limitations.

The present findings, along with those of prior studies that show either no relationship between low PA and quitting (Schnoll et al., 2013) or a positive relationship between anhedonia and quitting (Powers et al., 2016), suggest caution regarding the proposal that increased PA should be a central focus of targeted treatments in order to improve cessation outcomes (Audrain-McGovern, Wileyto, Ashare, Cuevas, & Strasser, 2014; Hoeppner, Hoeppner, Kelly, Schick, & Kelly, 2017; Leventhal et al., 2014; Leventhal et al., 2008). At the same time, the results of three pilot studies do suggest some early potential for interventions that focus primarily on boosting PA to assist smokers with low PA to quit successfully (Kahler et al., 2014; Kahler et al., 2015; MacPherson et al., 2010). These pilot results are not conclusive, however, and the studies conducted by Kahler and colleagues (Kahler et al., 2014; Kahler et al., 2015) suggested that a positive psychotherapy-based intervention may have lower acceptability and/or efficacy among the smokers who theoretically stand to benefit most from it—those with the lowest levels of PA at baseline. Additional evidence is needed to determine whether PA-focused interventions are as effective as a broader CBT-based mood management approach.

None of the symptom dimensions or total CES-D score predicted relapse among those who successfully initiated abstinence. There are several potential explanations for this finding. It may be that those smokers who are able to successfully quit in spite of experiencing elevated depressive symptoms are less susceptible to relapse. It may also be due to the exclusion of more depressed participants from the relapse analysis because they were unable to achieve initial abstinence. Alternatively, depressive symptoms may improve over time, with or without quitting smoking (Prochaska et al., 2008), reducing the predictive value of baseline symptom measures. Because the parent trial was not designed to examine changes in depression over the course of a quit attempt and the CES-D was not administered at end of treatment, we were unable to test this hypothesis. Future studies should incorporate an assessment of depression that is more proximal to the time period over which relapse potential is to be evaluated.

Our finding that, for some 6-month outcomes, depressive symptom total scores or domains predicted cotinine-confirmed, but not self-reported, abstinence outcomes also warrants comment. To gain a greater understanding of the reason for the discrepancy, we evaluated the relationship between depressive symptoms and reclassification of self-reported abstainers as smokers and found that depressive symptoms predicted elevated cotinine levels among self-reported abstainers. Although some prior work has evaluated smoker characteristics as predictors of discordant self-reported vs. biochemically-confirmed abstinence (Morales et al., 2013), this is the first study, to our knowledge, to identify depression as a predictor of discordant findings. We consider these findings preliminary, however, because of the small number of self-reported quitters in which cotinine levels were elevated (n=12) and because elevated cotinine levels due to use of alternative sources of nicotine (e.g., nicotine replacement products, electronic cigarettes) could not be ruled out. If these findings are replicated and not due to alternative sources of nicotine, cessation studies focusing on smokers with depressed mood may require biochemical verification of abstinence for an accurate assessment of outcome.

4.1. Limitations

All participants in the trial received nicotine replacement as well as one of two behavioral interventions for cessation. Results may not generalize to smokers who use other forms of treatment or do not seek treatment. Second, our definition of relapse is limited by the time points at which follow-up data were collected and the maximum time window over which smoking was assessed (i.e., 30 days prior to each follow-up). Thus, it is possible that some episodes of relapse between 1-week post-treatment and 6-month follow-up were not detected. Additionally, PA and other depressive symptoms were only assessed at baseline. Therefore, we were unable to evaluate the hypothesis that changes in PA during the quit attempt may predict cessation or relapse. Finally, although moderate to large correlations have been observed between the global assessment of PA in the CES-D and self-report measures of anhedonia (Ameringer, Chou, Sussman, Unger, & Leventhal, 2015), these results may not generalize to anhedonia-specific low PA (i.e., the inability to experience PA in anticipation of, or in response to, pleasurable events).

4.2. Conclusions

The present study is the first to use a large, diverse sample of treatment-seeking smokers to examine the relationship between current depressive symptom domains and three critical cessation outcomes: early cessation, long-term cessation, and relapse. Results indicated that the CES-D total score as well as three out of four symptom domains predicted abstinence at end of treatment and biochemically-confirmed abstinence at 6-month follow-up, and there was no evidence that PA had incremental predictive value beyond that of the other domains. Findings from the present study have two key treatment implications for the broader population of treatment-seeking smokers: (1) pre-quit depressive symptoms should be targeted as a risk factor for treatment failure, and (2) the role of PA and PA-focused treatments in smoking cessation, a nascent area of research, requires additional evaluation before such treatments can be recommended over more comprehensive CBT-based mood management.