Effects of tobacco on affect and craving during opioid addiction recovery: An ecological momentary assessment study

1. Introduction

Smoking prevalence varies by subgroup in the United States, and individuals with substance use disorders (SUD) are a particularly high-risk group. While elevated smoking prevalence is reported for individuals in all types of drug treatment, the highest prevalence (73–98%) has been reported in patients with opioid use disorders (OUD; Chun et al., 2009; Clemmey et al., 1997; Guydish et al., 2011; Kalman et al., 2005; Zirakzadeh et al., 2013).

Promoting tobacco cessation for individuals with OUD may reduce future morbidity and mortality, and opioid-dependent individuals in treatment tend to have favorable attitudes towards cessation (McClure, Acquavita, Dunn, Stoller, & Stitzer, 2014; Stein, Anderson, Thurmond, & Bailey, 2015). Despite these favorable attitudes, intervention efforts - including behavioral, nicotine replacement, and nicotine agonist approaches - have had little success in producing long term cessation (Okoli et al., 2010; Zirakzadeh et al., 2013). We review the psychological explanations thought to contribute to this lack of success below to provide background for the negative reinforcement regulation hypotheses tested in this manuscript.

1.1. Hypothesized barriers to tobacco cessation during OUD treatment

Evidence suggests that reinforcement for tobacco use during OUD recovery is partly driven by user expectancies regarding affect regulation. Expectancies may arise from past experiences with tobacco, or they may develop for other reasons (e.g., social learning; Bandura, 1978). Individuals with SUD have reported using tobacco to manage affective states and stress levels, as well as to distract them from drug cravings (Asher et al., 2003; Bhuiyan, Jonkman, Connor, & Giannetti, 2017; Kathuria et al., 2019). Similarly, polysubstance-using smokers reported higher expectancies for poor recovery outcomes, including increases in other drug use, if they were to quit smoking (Hendricks, Peters, Thorne, Delucchi, & Hall, 2014). Kraemer et al. (2013) found that low emotional distress tolerance was a significant internal barrier to cessation among daily smokers, suggesting that individuals who struggle to cope with emotional distress may view tobacco use as an important affect regulation strategy. The negative reinforcement model for addiction provides a useful framework for understanding these findings (Baker, Piper, McCarthy, Majeskie, & Fiore, 2004).

1.2. Nicotine and the negative reinforcement model for addiction and cravings

The negative reinforcement model for addiction posits that, for dependent individuals, avoidance of negative affect (NA) is a core motivation for the use of psychoactive substances (Baker et al., 2004). Known as spiraling dysregulation, aversive withdrawal symptoms increase NA, prompting individuals to use drugs to mitigate NA (Koob & LeMoal, 2001). Baker et al. (2004) theorized that substance use in response to NA may be subconscious when NA is low, but transitions to a learned, cognitively-driven response at higher levels of NA. Substance users eventually generalize this response to non-withdrawal stressors (i.e. interpersonal conflict), such that these everyday stressors prime substance use (Baker et al., 2004).

In an ecological momentary assessment (EMA) study, Carter et al. (2008) found that smokers’ self-reported tobacco craving and NA ratings were lowest immediately after smoking compared with immediately before smoking and random times of day, suggesting that smokers may be motivated to smoke to reduce craving and improve their mood states. However, other studies have found that while smoking relieves tobacco withdrawal-related NA, it is less effective for NA related to other stressors (Perkins, Karelitz, Conklin, Sayette, & Giedgowd, 2010).

NA increases risk of relapse for individuals in OUD treatment, an effect that is mediated by increased drug craving (Fatseas et al., 2015; Huhn et al., 2016; Moore et al., 2014; Serre, Fatseas, Denis, Swendsen, & Auriacombe, 2018). Across EMA studies, craving is a strong predictor of relapse for individuals in treatment for alcohol and other substance use (Fatseas et al., 2015). Additionally, in an EMA study of individuals in treatment for cocaine and heroin, stress was positively correlated with momentary reports of both NA and illicit drug craving (Preston & Epstein, 2011). Therefore, if tobacco use significantly reduces stress or NA for individuals in treatment for OUD, it may have some utility as a strategy for relapse prevention.

However, some evidence suggests that tobacco use may increase craving for other drugs. In one EMA study, the percentage of random-prompt entries during which participants reported tobacco smoking increased linearly with ratings of tobacco craving, cocaine craving, and craving for both cocaine and heroin (Epstein, Marrone, Heishman, Schmittner, & Preston, 2010). Additionally, a tobacco cessation study of methadone-maintained OUD patients found those who used fewer cigarettes had less craving for their primary treatment substance at follow up (Reid et al., 2008; Zirakzadeh et al., 2013). For individuals who tend to use tobacco concurrently with other substances, tobacco consumption may cause cue-elicited cravings that motivate alcohol and other drug use (Drobes, 2002).

1.3. Study overview

EMA designs are necessary for parsing temporal associations between tobacco use and affect in non-experimental studies; however, existing EMA studies of tobacco use during OUD treatment have examined only concurrent associations (Epstein et al., 2010). This study uses a measure of tobacco use frequency since last assessment, centered to parse out effects of individual differences, day-level processes, and momentary effects in tobacco use frequency on real-time reports of affect and craving. This allows us to examine temporal associations, as well as the direction and nature of any significant correlations.

Our overarching goal is to determine whether tobacco use provides any unique benefit for individuals in recovery for OUD by reducing stress or NA (the negative reinforcement hypothesis; Fatseas et al., 2015; Moore et al., 2014; Preston and Epstein, 2011; Scott et al., 2018). We also examine support for the alternative hypothesis that tobacco use increases other drug craving.

2. Material and methods

2.1. Participants and setting

Participants were recruited from semi-independent inpatient and outpatient addiction recovery programs in the Triangle area of North Carolina. Participants were recruited via in-person clinic visits, flyers, brochures given at clinic intake, directed phone calls, and paper mailings. Prospective participants were screened to ensure that they met the following study criteria: (1) at least 18 years of age; (2) not actively using their primary treatment substance; (3) use of primary substance within the past 12 months; (4) access to a personal mobile phone; and (5) not currently pregnant. Pregnant women were excluded because of the parent study’s focus on dietary outcomes and their different nutritional requirements.

The parent study included N=128 individuals in treatment for SUD. Of the individuals in the parent study, N=73 reported that they were receiving treatment specifically for OUD (heroin or illicit use of prescription opioids). All OUD individuals who also reported current tobacco use (N=55) contributed data to this study (Table 1). Women were heavily represented in this sample, as one of the clinics we recruited from was a female-only clinic.

Table 1.

Sample demographic characteristics

	M (SD) or % (N)
Age	36.83 (9.83)
Sex
Male	16 (9)
Female	84 (46)
Race
Black	11 (6)
White	85 (46)
Other	4 (2)
Education
Less than high school	14 (7)
High school diploma or GED	24 (12)
Associates degree or higher	56 (31)
Employment
Unemployed	29 (13)
Not working - disability	35 (16)
Part-time	13 (6)
Full-time	11 (5)
Treatment substance(s)
Heroin	65 (36)
Prescription opioid	60 (33)
Cocaine	29 (16)
Alcohol	20 (11)
Marijuana	18 (10)
Methamphetamine	13 (7)
Other substance	2 (1)
Last use of primary treatment substance
1–3 months ago	64 (35)
3–6 months ago	18 (10)
6–12 months ago	18 (10)

Note: N’s range from 40 to 55 due to non-response (particularly for employment), and a survey programming error that was corrected during the study. Treatment substances are not mutually exclusive.

2.2. Procedures

This study was approved by the UNC IRB and participants provided written informed consent. During the intake session, participants completed a tablet-based baseline survey and indicated their preference for text messages that linked to an EMA survey versus an interactive voice messaging system.

Intake sessions initiated a series of three, week-long measurement bursts spaced six weeks apart, during which participants received four EMA surveys per day. Surveys were scheduled to occur within four random uniform intervals that were defined by each participant’s sleep and wake times, with a constraint that no two surveys could be sent within one hour of each other. Participants were compensated for each completed survey. Participants were instructed to skip an EMA survey if they received the next survey before they had an opportunity to respond. Any survey responses completed within 30 minutes of the previous survey were removed. Timestamps were recorded to ensure data quality.

The 55 individuals included provided a total of 1,763 unique observations. Descriptive statistics by burst are listed in Table 2. In total, there were an average of 37.24 EMA surveys per person (SD=21.31) across all individuals and all bursts.

Table 2.

Descriptive statistics by burst

	Burst 1	Burst 2	Burst 3
	M (SD) or % (N)	M (SD) or % (N)	M (SD) or % (N)
Individuals providing data (N)	55	43	35
Observations per person	14.36 (5.92)	15.07 (5.91)	15.83 (6.96)
Tobacco use frequency	4.37 (3.30)	3.95 (3.09)	3.52 (2.91)
Negative affect	0.53 (0.59)	0.40 (0.54)	0.45 (0.53)
Stress	29.21 (28.61)	29.20 (29.62)	32.75 (28.66)
Observations reporting any craving	35.65 (246)	36.85 (199)	29.76 (125)

2.3. Measures

To reduce participant burden, one-item assessments were used for all measures that were obtained through EMA, except for affect.

Negative Affect:

Participants responded to the 20-item Positive and Negative Affect Schedule (Watson et al., 1988) in each EMA, indicating “To what extent do you feel this way right now?” using a 4-point scale ranging from “Not at all” to “Very much.” Scores on the NA subscale were averaged to yield mean NA (range 0–3) for each EMA (α= 0.85).

Stress:

For each EMA, participants were asked: “How much stress are you experiencing right now on a scale of 0 (no stress) to 100 (extreme stress).” Participants responded by entering a number 0–100.

Cravings:

For each EMA, participants were asked: “How much do you crave drugs or alcohol right now?” Response options were: “Not at all,” “Just a little,” “A moderate amount,” “Quite a lot,” and “It is all I can think about.” Due to sparse cell counts, momentary craving at each EMA survey was recoded as a binary variable, where 36.71% of observations reflected episodes of craving.

Frequency of tobacco use:

At each EMA, participants were asked: “How many times have you used tobacco since you were last surveyed? Response options ranged from 0 to 9 or more. The structure of this question allowed us to establish temporality of the effect of prior tobacco use on current affect, stress, and craving.

2.4. Data Analysis

Repeated measures data were analyzed using three-level multilevel modeling to account for nesting of observations within days and days within individuals. All analyses were conducted using the GLIMMIX or MIXED procedures in SAS (Version 9.4).

For each outcome (NA, stress, craving), we used a hierarchical model-building approach using a series of three models. First, we estimated a null model with no predictors to estimate intraclass correlation coefficients (ICCs) across levels (Model 1). Second, we estimated a model including the raw measure of tobacco use frequency as a predictor (Model 2).

In its raw form, the tobacco use frequency measure contains variance from three levels: at the person-level some individuals smoke more frequently than others. At the day-level, smokers smoke more on some days than others. Finally, people smoke different amounts at different moments within a day. Only this last portion of variance in tobacco use can be unequivocally temporally related to subsequent affect, so it is necessary and informative for us to parse variation in tobacco use into person-level, day-level, and within-day components. Our third model disaggregates the main effect of tobacco use frequency on each outcome.

We followed the centering approach from Curran and Bauer (2011) to facilitate the disaggregated analysis. Tobacco use frequency was averaged within person, and then grand mean centered, to create a unique measure of person-mean tobacco use frequency (TobFx_GMC_j). Uncentered person mean tobacco use frequency was subtracted from each person’s mean tobacco use frequency within a day to calculate a unique measure of day-level variation in tobacco use frequency (TobFx_PMC_ij). These day-mean measures were subtracted from momentary assessments of tobacco use frequency to calculate a measure of within-day variation in tobacco use frequency (TobFx_DMC_tij).

The 3-level model for each outcome (excluding covariates), estimated for occasion t on day i, for person j was:

$$\text{Level 1}:{\widehat{y}}_{tij}={\pi }_{0ij}+{\pi }_{1ij}{\mathit{\text{TobFx}}}_{-}DM{C}_{tij}$$

$$\text{Level 2: }{\pi }_{0ij}={\beta }_{00j}+{\beta }_{01j}{\mathit{\text{TobFx}}}_{-}PM{C}_{ij}+r_{0ij}{\pi }_{1ij}={\beta }_{10j}$$

$$\begin{gathered} \text{Level }3:{\beta }_{00j}={\gamma }_{000}+{\gamma }_{001}{\mathit{\text{TobFx}}}_{-}GM{C}_j+{\gamma }_{002}{\mathit{\text{clinic}}}_j+u_{00j}{\beta }_{10j}={\gamma }_{100} \\ \left(r_{0ij}\right)~N\left(0,{\sigma }^2\right)\left(u_{00j}\right)~N\left(0,{\tau }_{00}\right) \end{gathered}$$

All three-level models included a random intercept at level-2 (varying across days within person) and a random intercept at level-3 (varying across persons) because the day-level and person-level ICCs were meaningfully large for all outcome variables (Table 3). We also tested for random slopes for all significant fixed effects in the final model. However, these models failed to converge to a proper solution; random slopes were omitted from final models. Model fit statistics are reported for each outcome (Table 4). All non-null models controlled for weekday versus weekend, time of day, length of time since last EMA survey, months since last use of treatment substance (abstinence), sex, and treatment clinic. Additionally, we conducted a sensitivity analysis including use of medication assisted treatment (MAT), number of treatment substances, and self-reported relapse as covariates. None of the covariates tested in the sensitivity analysis were significant predictors of stress, NA, or craving, and the significance of our results (Table 4) did not change.

Table 3.

Intraclass correlations for all outcome variables

Intraclass Correlations
	Stress	Negative Affect	Craving
Level 3 (Person-Level) ICC	.52	.52	.45
Level 2 (Day-Level) ICC	.12	.22	.13

Note: For craving, a binary outcome, the level-1 variance was fixed to π²/3 following the logistic latent variable approach specified by Snijders and Bosker (1999).

Table 4.

Model results and model fit statistics

	Null (Model 1)			Raw (Model 2)			Disaggregated (Model 3)
	Coef	SE	p	Coef	SE	p	Coef	SE	p
Stress
Fixed effects
Intercept	32.37*	2.86	<.0001	32.19*	3.13	<.0001	32.62*	3.00	<.0001
TobFx_rawtij				0.13	0.25	0.61
TobFx_GMCj							1.43	1.15	0.22
TobFx_PMCij							0.37	0.37	0.32
TobFx_DMCtij							−0.21	0.35	0.55
Random effects
Day	130.68	17.21		128.10	17.30		127.95	17.29
Person	414.03	86.08		409.65	85.29		408.83	85.63
Model Fit
−2LL	14941.8			14755.9			14750.6
AIC	14947.8			14761.9			14756.6
BIC	14953.8			14767.9			14762.6
Negative Affect
Fixed effects
Intercept	0.56*	0.06	<.0001	0.50*	0.06	<.0001	0.58*	0.06	<.0001
TobFx_rawtij				0.02*	0.004	<.0001
TobFx_GMCj							0.02	0.02	0.32
TobFx_PMCij							0.03*	0.01	0.001
TobFx_DMCtij							0.02*	0.01	0.005
Random effects
Day	0.08	0.01		0.08	0.01		0.08	0.01
Person	0.18	0.04		0.18	0.04		0.18	0.04
Model Fit
−2LL	1757.0			1708.3			1720.3
AIC	1763.0			1714.3			1726.3
BIC	1769.0			1720.3			1732.3
Craving
Fixed effects
Intercept	−0.75 0.27			−1.14	0.33		−0.80	0.31
TobFx_rawtij				0.08*	0.03	0.01
TobFx_GMCj							0.06	0.11	0.58
TobFx_PMCij							0.06	0.05	0.19
TobFx_DMCtij							0.11*	0.05	0.03
Random effects
Day	0.86 0.19			0.92	0.19		0.93	0.19
Person	3.48 0.83			3.51	0.84		3.58	0.86
Model Fit
−2PLL	8161.02			7948.65			7961.67

Note: In the column headings of this table, “coef” stands for coefficient, “SE” for standard error, and “p” for p-value.

^*Estimates are significant at p<.05. Significant covariates across models are discussed in the text.

3. Results

Model 1: Null models

The level 2 and level 3 ICCs for all outcomes, shown in Table 3, suggest that three level hierarchical models were appropriate. The level 3 ICC indicates the proportion of the variance in the dependent variable that is accounted for by individual differences. The level 2 ICC measures the additional portion of the variance in the dependent variable that is accounted for by similarities in the dependent variable measurements taken within the same day for a given person. The remaining momentary variance is what is predicted by day-mean-centered tobacco use frequency.

Model 2: Raw effects of frequency of tobacco use on outcome variables

There was a significant main effect of frequency of tobacco use on craving and NA (Table 4). Each additional instance of tobacco use was associated with a 1.08-factor (SE=0.03) increase in craving and a 0.02-unit increase in mean NA (SE=0.004). There was no main effect of raw tobacco use on stress. Additionally, time since last use of treatment substance significantly predicted craving and stress, such that fewer months of abstinence was associated with a 1.43 factor (SE=0.12) increase in craving and a 3.58-point (SE=1.25) increase in stress. No other covariates were significant for any outcome.

Model 3: Disaggregated effects of frequency of tobacco use on outcome variables

The disaggregated effects model for craving showed a significant momentary effect of frequency of tobacco use on craving (Table 4). Every additional instance of tobacco use relative to the day mean was associated with a 1.11 factor (SE=0.05) increase in subsequent craving for alcohol or other drugs. There were no significant within person or within day effects of tobacco use on cravings. Additionally, time since last use of treatment substance significantly predicted craving such that fewer months of abstinence was associated with a 1.44 factor (SE=0.12, p=0.0004) increase in craving. No other covariates were significant.

The disaggregated effects model for NA showed a significant momentary effect on NA (Table 4). Every additional instance of tobacco use relative to the day mean was associated with a 0.02-unit (SE=0.01) increase in mean NA. In standardized terms, a 1 standard deviation increase in tobacco use frequency is associated with a 0.06 standard deviation (SE= 0.02) increase in subsequent NA. There was also a significant day-level association between frequency of tobacco use on NA (β=0.03, SE=0.01), suggesting that individuals smoke more on days when they feel more negatively. No covariates were significant predictors of NA.

Consistent with the main effect model, there were no significant disaggregated effects of tobacco use frequency on stress. Time since last use remained a significant predictor or stress, such that fewer months of abstinence was associated with higher stress levels. No other covariates were significant predictors of stress.

4. Discussion

Tobacco use frequency had no impact on self-reported stress, however, momentary tobacco use was associated with small increases in subsequent NA and cravings. When we disaggregated smoking frequency, we found that momentary periods of heavier smoking were the primary drivers of this association, but that OUD individuals also tend to smoke more on days when they have higher NA.

These results contradict EMA findings among smoking-only populations; Carter et al. (2008) and Chandra, Scharf, and Shiffman (2011) found that smoking did precede lower NA. This discrepancy may be due in part to the type of stressors participants were experiencing. Perkins et al. (2010) found that smoking was effective for relieving NA related to nicotine withdrawal, but not from other stressors. Our findings support the association observed by Epstein et al. (2010) regarding craving, and further, establish temporality. Periods of heavier tobacco use were significantly and temporally associated with increased craving for alcohol or other drugs.

These findings have implications for tobacco cessation interventions in clinical settings. The ideal timing of tobacco cessation for OUD and other substance use patients is controversial among clinicians (Baca & Yahne, 2009; Gentry, Craig, Holland, & Notley, 2017). Clinicians have expressed fears that tobacco cessation may jeopardize recovery outcomes, although for the most part existing literature shows little impact (Baca and Yahne, 2009; Derefinko et al., 2018; Gentry et al., 2017; McKelvey et al., 2017). The results of this study suggest that tobacco use does not significantly, beneficially regulate patient affect. However, participants were active smokers, and results can only speak to the efficacy of tobacco use for managing non-pharmacologic stressors and negative affect related to withdrawal from opioids.

An additional EMA study focused on comparing stress, affect, and craving among nicotine-deprived and nicotine-using OUD patients is needed. Per the negative reinforcement model of addiction, tobacco cessation will induce nicotine withdrawal and increase negative affect. Future studies should determine if this increase in withdrawal-related negative affect is large enough to promote relapse, to tobacco alone or to both tobacco and opioids.

Limitations

Our sample was heavily female. This is a strength because females are traditionally underrepresented in studies of SUD and may have gender-specific treatment needs (Tuchman, 2010). However, it is also a limitation. Our findings may not be generalizable to male patients with OUD. Given evidence that the associations between craving, mood, and substance use behaviors may differ by gender, further research with a balanced gender sample is needed (Kennedy et al., 2013; MacLean, 2018; Moran et al., 2018; Shiffman & Rathbun, 2011).

Additionally, there was some non-response to EMA prompts (average study response rate=53.8%). It is possible that people did not respond at times of higher stress, craving, or NA. The FIML estimation method used in this study assumes that missing observations are conditionally missing at random. However, Gottfredson, Sterba, and Jackson (2017) show that non-ignorable missingness is least problematic in studies with intensive repeated measures (i.e. EMA), because random coefficients partially account for nonignorable missing data mechanisms.

Using random prompts, we were only able to report total tobacco use since last survey and not exact time since last use. We found small increases in negative affect with increased tobacco use, but it is possible that the effects of nicotine on mood are more short-lived. If so, time since last cigarette may be a more important predictor of current affect or craving than number of uses. Future studies with event-based surveys may be better poised to examine if tobacco use is linked to short-term affective changes of larger magnitude for individuals in recovery from OUD.

Finally, evidence suggests that pharmacological mechanisms may also contribute to OUD patients’ difficulty quitting tobacco. Opioid and nicotinic-cholinergic neurotransmitter systems interact to regulate opioid and nicotine effects. The brain’s opiodergic system is involved in the development of physical dependence on nicotine, as nicotine stimulates opioid receptors that produce nicotine-rewarding effects (Hadjiconstantinou & Neff, 2011). Furthermore, the nicotinic-acetylcholine system modulates self-administration of opioids (Yoon, Lane, & Weaver, 2015). In clinical studies of withdrawal symptoms, methadone administration and cigarette smoking had a significant, interactive effect on reducing opioid withdrawal symptoms (Elkader, Brands, Selby, & Sproule, 2009; Zirakzadeh et al., 2013). Medication assisted treatment (MAT) may moderate the association between tobacco use and the outcomes studied; however, we did not have enough non-MAT participants (n=9 of 55 individuals; n=286 of 2,048 observations) to test for group differences.

5. Conclusions

Frequency of tobacco use during recovery for OUD is significantly associated with small increases in subsequent negative affect and craving. Perceived effects of tobacco on affect regulation may be more important in understanding rates of tobacco use than the actual, observed effects. These results suggest that clinicians should not discourage tobacco cessation during OUD treatment on the basis of the belief that smoking is a harm reduction tool.

Highlights:

• For opioid use disorder patients, tobacco use was associated with increased craving

• Tobacco use frequency was also associated with small increases in negative affect

• Tobacco use frequency did not significantly affect patients’ stress levels

• Tobacco use does not provide affect regulatory benefits for these patients