Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2025 Feb 1.
Published in final edited form as: Psychol Addict Behav. 2023 May 18;38(1):36–46. doi: 10.1037/adb0000931

Mechanisms of Behavior Change during Alcohol Treatment among Negative Affect Drinkers: A Time-Varying Effect Model Analysis Using 84 Consecutive Days of Ecological Momentary Assessment

Junru Zhao 1, Braden K Linn 1, Paul R Stasiewicz 1, Gregory E Wilding 2, Charles LaBarre 1, Clara M Bradizza 1
PMCID: PMC10656359  NIHMSID: NIHMS1890716  PMID: 37199961

Abstract

Objective:

To better understand the timing and unique contribution of four potential mechanisms of behavior change (MOBC) during alcohol use disorder (AUD) treatment (negative affect, positive affect, alcohol craving, adaptive alcohol coping), we used a time-varying effect modeling (TVEM) analytic approach to examine the change trajectories of alcohol abstinence, heavy drinking, the hypothesized MOBCs, and the time-varying associations between the MOBCs and alcohol outcomes.

Method:

Participants (N = 181; Mage = 50.8 yrs, SD = 10.6; 51% women; 93.5% Caucasian) were enrolled in a 12-week randomized clinical trial of cognitive-behavioral outpatient treatment program for AUD. For 84 consecutive days, participants provided self-reports of positive and negative affect, craving, alcohol use and adaptive alcohol coping strategies employed.

Results:

Throughout the 84-day treatment window, higher daily average craving levels were associated with both decreased likelihood of alcohol abstinence and increased odds of heavy drinking, whereas higher adaptive alcohol coping was associated with increased odds of abstinence and decreased odds of heavy drinking. Higher negative affect was associated with decreased odds of abstinence in the first 10 days of treatment and increased odds of heavy drinking before day 4 or day 5. Higher positive affect was associated with decreased odds of heavy drinking during the first four or five days.

Conclusions:

The differential time-varying associations between negative affect, positive affect, alcohol craving, adaptive alcohol coping, and alcohol use provide insights into how and when each of the MOBCs is active during AUD treatment. These findings can help optimize the efficacy of future AUD treatments.

Keywords: mechanisms of behavior change (MOBC), time-varying effect model (TVEM), alcohol use disorder (AUD), negative affect, positive affect

Each year more than two million adults ages 18 and older present for alcohol use disorder (AUD) treatment in the United States (National Institute on Alcohol Abuse and Alcoholism [NIAAA], 2019). Whereas treatment is effective for some, it is not effective for everyone (Miller & Wilbourne, 2002). To improve treatment effectiveness, it has been suggested that research identify processes that mediate the effects of treatment on an outcome (Finney, 2018; Kazdin, 2007; Longabaugh et al., 2005). Such mediators are termed mechanisms of behavior change (MOBC). MOBC studies have been a top priority of the alcohol treatment field for over 20 years (Mechanisms of Behavior Change Satellite Subcommittee, 2018). Although several variables have been theorized to be MOBCs, consensus on MOBCs in alcohol treatment has not yet been reached, inhibiting the development of new treatments, tailored treatments, and research that determines treatment response (Hallgren et al., 2018; Kazdin, 2007).

Craving, affect, and coping each have emerging empirical or theoretical support as possible MOBCs in alcohol treatment and other areas of behavioral health (Bresin et al., 2018; Kiluk et al., 2010; Linn et al., 2021; Magill et al., 2015; Petry et al., 2007; Roos et al., 2017). However, simultaneously examining multiple MOBCs is critical because most treatments are theorized to have more than one MOBC operating at any given time (Kazdin & Nock, 2003). A major limitation in the field of MOBC studies is that researchers have utilized clinical trial data, which remains focused primarily on post- vs. pre-test change in each specified clinical outcome (e.g., drinking intensity/frequency, alcohol-related problems). Putative MOBCs may only be measured at baseline, which not only underestimates the potential mediating effects (Cohen, 2001) but also prohibits modeling of how and when the MOBC is active during treatment (Hallgren et al., 2018). Knowledge of how and when a MOBC is active may help inform the optimal temporal ordering of treatment content or may offer ideas for how to personalize treatment content to match a client’s alcohol problem phenotype. Indeed, a recent study found that CBT may work, in part, to “decouple” the relationship between negative emotions and alcohol use (e.g., Linn et al., 2021).

The current study leverages recent advances in data collection and analytic techniques to extend the literature on MOBCs in alcohol treatment. Specifically, ecological momentary assessment (EMA; Schiffman et al., 2008) was used to collect 84 consecutive days of within-treatment data on four hypothesized MOBCs (i.e., craving, negative affect, positive affect, and adaptive alcohol coping) and two alcohol outcome variables (i.e., alcohol abstinence and heavy drinking). Indeed, the use of EMA data has the potential to advance AUD research on within-person associations between MOBCs and drinking (e.g., Morgenstern et al., 2016). In addition, time-varying effect models (TVEM; Hastie & Tibshirani, 1993) were used to analyze the data.

Data collected at a high resolution (i.e., daily) and analyzed using TVEM has advantages over previous MOBC studies. First, high-resolution EMA data offers the opportunity to capture more nuanced change. For example, a MOBC’s change trajectory may be linear, non-linear or it may have sudden changes known as saltuses (Snippe et al., 2017). The ability to capture these nuances has future implications for how treatments might be tailored to optimize clinical outcomes. Secondly, using a spline-based estimation procedure, TVEM is a nonparametric statistical technique that flexibly models temporal change, especially when the shape of change is unknown (Tan et al., 2012). Finally, examining the time-varying associations between a MOBC and a drinking outcome can help better understand both the timing (e.g., when) and the strength or direction of contribution (e.g., how) of a MOBC. TVEM has been used in prior studies investigating alcohol treatment mechanisms (Linn et al., 2021), albeit with lower- intensity temporal resolution data. TVEM has also been used in smoking cessation studies to examine the relationships of different MOBCs to smoking cessation outcomes (Dermody & Shiffman, 2020; Shiyko et al., 2012).

In the present study, we examined: (1) 84 consecutive days of within-treatment change trajectories of four hypothesized MOBCs (i.e., negative affect, positive affect, adaptive alcohol coping, and alcohol craving) and alcohol outcome variables (i.e., alcohol abstinence and heavy drinking), and (2) the change in association between these hypothesized MOBCs and alcohol outcome variables.

Method

We report how we determined our sample size, all data exclusions (if any), all manipulations, and all measures in the study.

Participants

The parent study from which the current analyses are derived was a randomized clinical trial (NIH R01 AA024628) comparing the efficacy of 12 weekly sessions of CBT for AUD plus emotion regulation treatment (CBT + ERT) supplement to 12 sessions of CBT for AUD plus a healthy lifestyles supplement (CBT + HLS) intended as a time, attention and expectation of benefit control. Participants (N = 181) for the current analyses included men and women seeking outpatient treatment for an alcohol use disorder who reported drinking heavily in response to negative affect situations. Inclusion criteria included: a) a current DSM-5 diagnosis of alcohol use disorder, moderate or severe, b) reported drinking heavily in response to negative affect (see Measures – IDTS-A), c) consumption of any alcohol in the past 3 months, and d) lived within commuting distance of the program site. Exclusion criteria included: a) severe mental illness (e.g., psychosis or active bipolar disorder), b) changes in the past 3 months in the dose or type of prescription medication that affects mood (e.g., antidepressants or anxiolytics), d) any drug use disorder, other than nicotine or mild cannabis dependence, and e) mandated to attend treatment. Thus, the study sample comprises individuals who reported experiencing significant problems due to alcohol and frequently drank heavily in response to negative affect. By including individuals who reported frequent heavy drinking in negative affect situations we achieve a more sensitive test of the parent study aims (focused on the efficacy of the CB+ERT intervention) and increase the intervention’s relevance for this population. The University at Buffalo Institutional Review Board approved the study protocols.

Of the 530 men and women who were phone screened, 354 (66.8%) were eligible to participate in an in-person screening/baseline assessment. Of the 248 individuals who completed the in-person assessment, 194 (78.2%) were eligible to participate and were enrolled in the study. Twelve of the 194 individuals never attended the initial treatment session and one additional individual attended one treatment session but never initiated the EMA and subsequently withdrew from the study. Thus, the current analyses were based on the 181 participants with EMA data (see CONSORT Figure 1).

Figure 1.

Figure 1

Open in a new tab

Flow Chart

Procedure

Daily ecological momentary assessment (EMA) data were collected over 12 weeks of cognitive-behavioral AUD treatment. Interested individuals completed an initial phone screening, followed by an in-person screening for diagnosis and subsequently, a comprehensive in-person baseline assessment, within the next week. Starting in April 2020, all sessions took place over video conferencing to comply with COVID-19 social distancing measures and public safety precautions. Eligible individuals were then randomly assigned to one of the two treatment conditions: cognitive behavioral treatment for AUD plus an emotion regulation treatment supplement (CBT + ERT) or CBT plus a Healthy Lifestyles treatment supplement (CBT + HLS). Treatment assignment was based on the urn randomization procedure (Wei, 1978), which balanced participant gender, AUD severity, presence of a comorbid mood or anxiety disorder, and trait mindfulness. Treatment sessions were conducted by licensed, trained therapists with experience conducting alcohol treatment who received weekly supervision by an experienced clinical study supervisor.

All participants received 12 weekly sessions of standard CBT for AUD adapted from Project MATCH (Kadden et al., 1992). In addition, participants were randomized to receive either an emotion regulation (ERT; Stasiewicz et al., 2013; 2018) treatment supplement or a health and lifestyle (HLS) treatment supplement (Stasiewicz et al., 2013). Each weekly treatment session began with 45 minutes of CBT followed by either 45 minutes of ERT or HLS. The ERT treatment supplement includes adaptive emotion regulation strategies designed to help individuals develop the capacity to regulate negative affect in adaptive ways. The HLS treatment supplement is an active control that provides education about various health-related topics.

After completion of the baseline assessment, individuals received orientation and training in the use of the daily assessment technology. Participants used a smartphone to access a secure link once per day in the morning. They were asked to report on levels of positive and negative affect, alcohol craving, which adaptive alcohol coping skills they had employed (from a list of 9), and alcohol consumption on the previous day. Daily monitoring of EMA reports by research staff was conducted throughout the 12-week treatment period. If a report was missed, staff called the participant the same day to encourage regular reporting. During this period, 123 participants (68%) completed ≥ 80% of daily reports (67 or more days) and 21 completed all 84 days. Participants were compensated $1 per day for reporting. If they completed at least six daily reports within a week, they received a $10 bonus. They also received a $50 bonus for completing all 12 weeks of reporting.

Measures

Demographics

Participant demographic information was collected at the phone screen and baseline assessment. As shown in Table 1, the current sample was predominately Caucasian. Approximately half of the sample was married, had a university or post-graduate degree, and had a personal income above the median US annual salary of $53, 924 (Bureau of Labor Statistics, 2022).

Table 1.

Sociodemographic Characteristics of Study Participants

Variable n/M %/SD χ 2
Age (M, SD) 50.8 10.6
Gender (n, %) χ2 (1) = .14
 Male 88 48.6
 Female 93 51.4
Hispanic/Latino (n, %) χ2 (1) = 156.93***
 Yes 6 3.3
 No 175 96.7
Race (n, %) N/A
 White/Caucasian 170 93.5
 American Indian/Alaskan Native 0 .0
 Asian 1 1.0
 Native Hawaiian/other Pacific Islander 0 .0
 Black/African-American 10 5.5
 Other 0 .0
Highest educational level (n, %) χ2 (2) = 49.14***
 ≤ High school 22 12.5
 Some college 56 35.8
 University or post-graduate degree 98 55.7
Marital status (n, %) χ2 (5) =128.00***
 Single, never married 40 22.5
 Single, divorced 34 19.1
 Single, widowed 4 2.2
 Single, married but separated 8 4.5
 Living with partner as if married 12 6.7
 Married and living with spouse 80 44.9
Income (n, %) χ2 (1) = 10.32***
 ≥ $50,000 67 37.9
 > $50,000 110 62.1
Open in a new tab

Note. N/A: Not applicable. The chi-square statistics do not apply here, because when the expected frequency for a given cell is less than 1, the chi-square distribution will not be a reasonably accurate approximation of the distribution of the chi-square statistics (Cohen, 2001).

***

p < .001.

Baseline Psychological Symptoms

The overall severity of psychological symptoms was measured by the Brief Symptom Inventory (BSI; Derogatis, 1993) at the baseline assessment, which occurred approximately one week before the first treatment session. A total symptom score was computed from 53 items (Cronbach’s α = .96) reflecting nine domains (anxiety, hostility, somatization, obsessive-compulsive, interpersonal sensitivity, depression, phobic anxiety, paranoid ideation, and psychoticism). Each item was rated on a 5-point scale, ranging from 0 (not at all) to 4 (extremely). Higher scores indicated greater symptomology.

Inventory of Drug-Taking Situations-Alcohol Version (IDTS-A)

The IDTS-A is a 50-item measure that assesses situations in which individuals reported frequently drinking heavily over the past six months (Annis et al., 1995). Participants indicated the extent to which they drank heavily in each situation ranging from 1 (never) to 4 (almost always). This measure consists of eight different drinking situations: unpleasant emotions, physical discomfort, testing personal control, urges and temptations to use, conflict with others, social pressure to use, pleasant emotions, and pleasant times with others. Participants who scored above the mid-point on either the unpleasant emotions (10 items; Cronbach’s α = .91) or conflict with others (10 items; Cronbach’s α =.89) subscales met the study inclusion criteria of negative affect drinking.

Alcohol Abstinence and Heavy Drinking

During EMA assessments, alcohol use was assessed by the following two questions: (1) Did you consume any alcoholic beverages yesterday? and (2) How many standard alcoholic beverages did you consume yesterday? A response of no to the first question was coded as 1 (vs. 0) for the dichotomized alcohol abstinence variable on that day. A dichotomized variable of heavy drinking was coded as 1 (vs. 0) (≥ 4 standard drinks per day for women and ≥ 5 standard drinks per day for men) based on responses to the second question; heavy drinking was coded as 0 for responses of no to the first question and for responses to the second question that did not cross the above thresholds.

Positive and Negative Affect

Daily reports of positive affect (10 items; Cronbach’s α = .93) and negative affect (10 items; Cronbach’s α = .92) were assessed by the Positive and Negative Affect Scale (PANAS; Watson et al., 1988). Higher sum scores calculated from a 5-point scale (1 = very slightly to not at all and 5= extremely, respectively) indicated both higher positive affect and higher negative affect.

Alcohol Craving

Daily average levels of alcohol craving were assessed by the following question: Thinking about yesterday, what was your average level of craving or desire to drink over the course of the day? This question was rated on a scale from 1 to 100, such that 1–10 (coded as 1) indicated none/very low craving and 91–100 (coded as 10) indicated extreme craving, respectively.

Adaptive Alcohol Coping

Daily adaptive alcohol coping was assessed by nine yes or no items asking participants if they used each of the skills to help them avoid drinking alcohol, including I talked to someone about it; I waited out the urge or craving; I reminded myself of the benefits of quitting; I reminded myself of the negative consequences of drinking; I substituted a nonalcoholic beverage, food or activity; I thought about the situation in a different way; I said no when a drink was offered to me (drink refusal skills); I avoided coming into contact with alcohol, or people and places associated with alcohol; and I kept busy with other activities. Scores across the nine items were summed with higher scores indicating greater adaptive alcohol coping.

Analytic Strategy

First, potential confounds were examined. As shown in Table 2, participants’ baseline psychological symptoms, education level, income, and marital status were correlated with alcohol abstinence and heavy drinking during treatment. Also, throughout the treatment period, number of treatment sessions attended was positively correlated with adaptive alcohol coping, positive affect, and alcohol abstinence, and negatively correlated with craving, negative affect, and heavy drinking. Thus, these variables were controlled for in subsequent model testing. Next, preliminary analyses were conducted on all study variables. On average, alcohol abstinence during treatment was positively correlated with positive affect and adaptive alcohol coping and negatively correlated with negative affect and alcohol craving. Heavy drinking revealed a reversed pattern of these associations (see Table 2).

Table 2.

Descriptive Statistics and Zero-Order Correlations for Model Variables (N = 181)

Variable 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11.
1. Alcohol abstinence −
2. Heavy drinking −.71*** −
3. Negative affect −.24*** .17*** −
4. Positive affect .23*** −.18*** −.28*** −
5. Alcohol craving −.45*** .29*** .44*** −.20*** −
6. Adaptive alcohol coping .38*** −.12*** .00 .26*** −.14*** −
7. Psychological symptoms .02 −.07*** −.31*** .20*** −.09*** −.04***
8. Education −.07*** .08*** .09*** −.04*** .12*** −.04***
9. Income −.06*** .01 −.07*** .01 .08*** .01
10. Marital statusa .06*** −.20*** −.14*** .15*** −.10*** −.01
11. Number of treatment sessions attended .11*** −.11*** −.08*** .14*** −.05*** .17***
M .64 .61 16.31 28.52 2.40 2.55 51.81 2.43 3.14 1.54 8.67
SD .48 .49 7.06 8.51 2.35 2.10 2.81 .71 .94 .50 3.98
Open in a new tab

Note.

a

Marital status of single, never married; single, divorced; single, widowed; and single, married but separated were collapsed as single (coded as 1). Living with partner as if married and married and living with spouse were collapsed as cohabited (coded as 2).

*

p < .05.

**

p < .01.

***

p < .001.

Finally, the SAS macro %TVEM (Li et al., 2017) was used to fit a series of eight TVEMs. In each model, the P-spline basis and 10 knots (knots are the “joints” of a spline) were specified, which generates smooth estimated functions by automatically balancing a good model fit and over-fitting caused by a large number of knots.

Using the intercept-only TVEM models, we first explored the change trajectories of alcohol abstinence, heavy drinking, negative affect, positive affect, adaptive alcohol coping, and alcohol craving separately as a function of time (days) in treatment (Aim 1). In these models, the pseudo time-varying covariate intercept, which is a constant of 1, is included for estimation purpose. Next, we examined the time-varying concurrent daily associations between the criterion (alcohol abstinence or heavy drinking) and four MOBCs (negative affect, alcohol craving, adaptive alcohol coping, and positive affect) separately in each of the two models, controlling for participants’ psychological symptoms, education, income, marital status, and number of sessions attended during treatment (Aim 2).

For spline regression models in which the criterion was alcohol abstinence or heavy drinking, TVEM generated odds ratios corresponding to each of the time points assessed and the 95% pointwise confidence intervals. At a particular timepoint during treatment, the confidence interval on an odds ratio for a nonsignificant covariate included 1. Odds ratios greater than 1 indicated increased odds of alcohol abstinence or heavy drinking associated with a given covariate. Odds ratios falling between 0 and 1 reflected the decrease in odds of alcohol abstinence or heavy drinking associated with a given covariate. For spline regression models in which the criterion was negative affect/positive affect/adaptive alcohol coping/alcohol craving, TVEM generated regression coefficients corresponding to each of the time points assessed and the 95% pointwise confidence intervals. At a particular timepoint during treatment, the confidence interval on a coefficient for a nonsignificant covariate included 0. In TVEMs, explanation of the change trajectories is based on visual inspection of the overall shapes/curves of a time-varying construct or the time-varying associations.

TVEMs were conducted to examine if those assigned to the CBT+ERT condition had different changes in these MOBCs than those assigned to CBT+HLS condition. The only MOBC that exhibited group differences was adaptive alcohol coping, with individuals in the CBT+ERT group endorsing a greater number of adaptive coping skills during the first five weeks of treatment. TVEMs were also conducted to examine if treatment group status was predictive of any change trajectory of either of the two drinking outcomes. No group differences were found. On average, participants completed a mean of 8.67 (SD = 3.98) treatment sessions and the number of sessions attended were balanced between the two treatment conditions (r = .13, p = .076), with 9.07 (SD = 3.65) in the CBT+ERT condition and 9.77 (SD = 3.43) in the CBT+HLS condition, respectively. Further, logistic regression analyses suggested that the effects of treatment condition on alcohol use did not depend on the number of sessions attended (B = .11, p = .199 for alcohol abstinence and B = −.07, p = .483 for heavy drinking, respectively). Taken together, the lack of observed associations between treatment condition and drinking outcomes may not be masked by the extent to which the participants adhered to the treatment.

We also explored potential group differences in the time-varying associations between various proposed MOBCs and daily drinking outcomes and as expected, the findings did not differ across the two treatment groups1. These findings were consistent with those reported in a previous study from our research group (Linn et al., 2023), in which we tested structural invariance across two treatment groups by examining differences in the associations between MOBCs (e.g., negative emotionality) and drinking outcomes and found no group differences.

Missing Data and Sample Size Justification

Across the 84-day treatment period, sample sizes ranged from 124 to 172 for alcohol abstinence and heavy drinking. At this point, the sample size requirements for TVEM, are not yet well understood. However, there are some general guidelines in place (Lanza & Linden-Carmichael, 2021). Given that TVEM is a direct extension of multiple linear regression, the sample size determined for a particular regression at a particular time point with power of .80 and above can shed light on the adequacy of statistical power for other points in time. In addition, the pointwise confidence intervals used for significance testing in TVEM are sensitive to sample size, such that the intervals are wider with smaller samples. Using the G*Power software, a sample of 120 was found to have adequate power (.85) to detect a moderate effect (.15) (Cohen, 2001) for a multiple regression with eight covariates (four focal covariates and four control variables). Thus, the current sample sizes across different time points suggested adequacy for the planned analyses.

In TVEM, missing data are accounted for using full-information maximum likelihood (FIML) estimation procedure (Li et al., 2017). FIML accommodates missing-at-random (MAR; Shafer, 1997) and it can reduce potential bias on parameter estimates and standard errors even if MAR is violated (Arbuckle, 1996).

Results

Change Trajectories of Alcohol Use and Time-Varying MOBCs

Figure 2 depicts the intercept functions and the corresponding 95% confidence bands for within-treatment alcohol abstinence and heavy drinking. Starting from day four or five, the likelihood (odds) of alcohol abstinence continually increased through to the end of treatment in a roughly linear manner. Across the 84 days of treatment, there was a decreased likelihood of heavy drinking. As shown in Figure 3, throughout the 84-day treatment window, both adaptive alcohol coping and positive affect remained unchanged. There was a slight decrease in negative affect during the first three weeks, after which negative affect leveled off. There was a continuous decrease in alcohol craving across the 84 days.

Figure 2. Time-Varying Effect Model Intercept Function for Alcohol Abstinence and Heavy Drinking during Treatment.

Figure 2

Open in a new tab

Note. At a particular time point, a confidence interval not including 1 indicates a statistically significant odds (likelihood) of being abstinent or heavy drinking. An odds greater than 1 indicates increased likelihood whereas an odds falling between 0 and 1 indicates decreased likelihood.

Figure 3. Time-Varying Effect Model Intercept Function for Negative Affect, Positive Affect, Adaptive Alcohol Coping, and Alcohol Craving during Treatment.

Figure 3

Open in a new tab

Note. To fit the intercept function models, positive and negative affect items were recoded as ranging from 0 to 4 and the craving item was recoded as ranging from 0 to 9, respectively. At a particular time point, a confidence interval not including 0 indicates a statistically significant (nonzero) mean level of negative affect, positive affect, adaptive coping, or alcohol craving.

Time-Varying Associations between MOBCs and Alcohol Abstinence

Higher negative affect was related to decreased likelihood (odds) of abstinence during roughly the first 10 days of treatment and the period between day 44 and day 63; no associations were found for other time points (Figure 4A). During the entire treatment period, positive affect was not related to alcohol abstinence (Figure 4B). Throughout the 84 days of treatment, there was a relatively stable association between higher adaptive alcohol coping and increased odds of alcohol abstinence (Figure 4C). Higher daily average craving levels were associated with decreased odds of alcohol abstinence across all timepoints, and the decrease became even greater for the last three weeks of treatment (Figure 4D).

Figure 4. Time-Varying Associations between Negative Affect, Positive Affect, Adaptive Alcohol Coping, Alcohol Craving and Alcohol Abstinence during Treatment.

Figure 4

Open in a new tab

Note. OR= Odds ratio. NA= Negative affect. PA= Positive affect. AA= Alcohol abstinence. At a particular time point, the confidence interval for a significant time-varying association does not include 1.

Time-Varying Associations between MOBCs and Heavy Drinking

Higher negative affect was related to increased likelihood (odds) of heavy drinking during the first four or five days, and the associations were decoupled afterward until the end of treatment (Figure 5A). Higher positive affect was associated with decreased odds of heavy drinking during the first four or five days; there were no associations at other time points (5B). Throughout the 84 days of treatment, higher adaptive alcohol coping was associated with decreased odds of heavy drinking (Figure 5C). Higher daily average craving levels were associated with increased odds of heavy drinking across the entire treatment period and the shape of change in the associations manifested in a roughly quadratic form. That is, these associations were relatively stable for the first eight weeks, and then became increasingly stronger throughout the remainder of treatment (Figure 5D).

Figure 5. Time-Varying Associations between Negative Affect, Positive Affect, Adaptive Alcohol Coping, Alcohol Craving and Heavy Drinking during Treatment.

Figure 5

Open in a new tab

Note. OR= Odds ratio. NA= Negative affect. PA= Positive affect. HD= Heavy Drinking. At a particular time point, the confidence interval for a significant time-varying association does not include 1.

Discussion

MOBCs are factors posited to account for changes in outcomes during treatment (Kazdin, 2007). Knowing when, in what direction, and the strength of a given MOBC operates could be helpful when developing new treatment approaches for AUD. Using a TVEM analytic approach and daily EMA data spanning 84 consecutive days of AUD treatment, we aimed to explore: (1) within-treatment change trajectories of alcohol outcomes (alcohol abstinence and heavy drinking) as well as four hypothesized MOBCs (negative affect, positive affect, alcohol craving, adaptive alcohol coping) and (2) the time-varying associations between the MOBCs and alcohol outcomes. The current findings provide insight into how AUD treatment works and have implications for tailoring AUD treatment and possibly improving AUD treatment outcomes.

After the first four or five days, the likelihood of alcohol abstinence appears to be ever-increasing to the end of treatment. Across the 84 days of treatment, there was a decreased likelihood of heavy drinking. The change trajectories of alcohol abstinence and heavy drinking seem to be attributed to the overall decrease in negative affect during the first few weeks of treatment and the continuous decrease in craving throughout. It is also possible that all four proposed MOBCs work together in some way and contribute to the change trajectories of alcohol abstinence and heavy drinking. Though speculative, this explanation reflects Kazdin and Nock’s (2003) theory that multiple mechanisms acting simultaneously are responsible for behavior change.

The second aim of the study was to examine the unique daily associations of each of the four MOBCs with daily alcohol use during treatment. Despite its conceptual complexity and controversy, craving is currently viewed as an affective construct that is either ambivalently valanced (e.g., can be experienced as positive or negative depending on the context) (Sayette, 2006) or negatively valanced (Baker et al., 2004). In the current study, alcohol craving emerged as a salient risk for alcohol use. Over and above all other MOBCs, greater alcohol craving was associated with increased odds of heavy drinking, as well as a decreased odds of alcohol abstinence. These findings are in line with the supported association between craving and substance use when using EMA (Serre et al., 2015). These findings also lend additional support to existing calls for the use of addiction interventions targeting craving (e.g., Unrod et al., 2014).

Self-report of more types of daily coping was related to increased odds of abstinence and decreased odds of heavy drinking, highlighting the importance of adaptive coping in the maintenance of treatment effects. Indeed, there has been accumulating evidence of the mediating role of coping skills in CBT or CBT-based AUD treatment (Magill et al., 2020). However, it is likely that individuals used a particular skill more than once a day. Thus, the use of one’s preferred skill(s) multiple times in a day can be equally adaptive in coping, as compared to the use of several different skills.

In this sample of negative affect drinkers (i.e., report frequent heavy drinking in negative affect situations), it is understandable that negative affect was associated with an decreased odds of abstinence during the first few days of treatment, prior to the participants receiving a substantial dose of treatment. However, the recurrence of an association between negative affect and a decreased odds of abstinence during weeks seven through nine is difficult to interpret. Although speculative, the return of a relationship between negative affect and drinking could reflect the challenges of learning and implementing new ways of managing unpleasant emotions, and such challenges could be greater in a sample of people who drink heavily in response to negative affect. However, it is equally likely that the re-emergence of this relationship is a statistical artifact that warrants investigation in future research.

Despite higher positive affect being associated with decreased odds of heavy drinking for the first four or five days, it was not associated with alcohol abstinence at any timepoint in treatment. These findings are somewhat inconsistent with a previous study demonstrating that positive affect was active in influencing percent days abstinent at the beginning and end of AUD treatment (Linn et al., 2021). These discrepancies may be due, in part, to the composition of the current sample, who were all negative affect drinkers. Alternatively, when positive affect was tested simultaneously with negative affect, adaptive alcohol coping and craving in the model, the role of positive affect as a MOBC may have been overpowered by these other factors. For instance, positive affect may interact with adaptive alcohol coping in an antagonistic fashion (Cohen, et al., 2003), such that adaptive alcohol coping diminishes the influence of positive affect in predicting alcohol use. However, this speculation may be better supported by either investigating interactions between positive affect, negative affect, adaptive alcohol coping and craving, or identifying distinct profiles/classes of the time-varying associations using a person-centered finite mixture version of TVEM (MixTVEM; Dziak, 2015; Yang et al., 2018).

There are several clinical implications to the current study. Greater negative affective is related to lower odds of abstinence during the beginning (first 10 days) and middle-end of treatment (days 44–63; corresponding to weeks 7 through 9 of the 12-week treatment). Greater negative affect is related to greater odds of heavy drinking up to day 5, and this relationship is “decoupled” until the end of treatment. During the treatment period, negative affect had a greater impact on whether a person drank on a specific day; however, treatment appears to mitigate the impact of negative affect on heavy drinking. Thus, individuals may still drink in response to negative affect but are less likely to drink heavily. Given the recent focus on reductions in heavy drinking as a desirable clinical endpoint (Witkiewitz et al, 2017), treatments that address negative affect as it relates to heavy drinking behavior are critical to ensuring positive outcomes.

Additionally, higher daily craving levels were associated with decreased odds of abstinence and increased odds of heavy drinking throughout treatment. Thus, addressing craving in alcohol treatment remains an important treatment target. Finally, throughout the treatment period, there was a stable association between higher adaptive coping and increased odds of alcohol abstinence and decreased odds of heavy drinking. However, treatment did not appear to strengthen this association as might be expected if individuals improved adaptive coping skills over the course of treatment.

Limitations

There are several limitations to the current study. First, information about putative MOBCs and the dependent alcohol variables are collected at the same point in time. Consequently, caution should be exercised when making inferences regarding the directionality of associations. For example, abstinence may drive affect change, rather than the reverse, given the potential for some portion of the current sample to be in physiological withdrawal. Alternative analytic approaches such as vector autoregression multivariate time-series analysis (VAR; Liu, 1986) may be employed to investigate the lagged relations and could provide greater certainty regarding directionality among the posited MOBCs and drinking outcomes. Second, some data were collected during the COVID-19 pandemic in which factors not measured may have impacted changes in the MOBCs, drinking behaviors, or both. Although we separated the participants into three groups (reported data via in-person only, video-conferencing only, and a mixture of in-person and video-conferencing), we did not find group differences in any of the MOCBs or drinking variables mentioned. However, the unpredictable influence of the pandemic on treatment delivery and other aspects of people’s lives cannot not be ruled out completely.

Third, all data used in the current analyses were self-reported and thus findings may be biased by shared method variances. Fourth, TVEM is a novel approach and remains an active area of research. Little is known about the inter-person variability on the robustness of model estimation (Tan et al., 2012). In other words, all the estimates (either intercept-only models or prediction models) derived from the current TVEM are average timing effects across people at any given time point. Unlike multilevel models, the averaged estimates of TVEM may have masked important individual differences and thus may also have biased the robustness of model estimation. Finally, the generalizability of the current findings may be limited by our sample of negative affect drinkers with moderate to severe AUD and by a sample of predominantly White men and women with higher SES.

Future Directions

Psychosocial treatments are theorized to have multiple MOBCs acting simultaneously (Kazdin, 2007). The current study tested the potential mechanisms of craving, adaptive alcohol coping, negative affect, and positive affect. The results indicate that craving, adaptive alcohol coping, negative affect, and positive affect are all active at the beginning of treatment. Craving and adaptive alcohol coping remaining active throughout, that is, the associations between craving/adaptive alcohol coping and drinking outcomes held throughout treatment. For future MOBC research, knowing when and how a MOBC is active permits the use of random assignment to different conditions that seek to vary the order of treatment content to manipulate a particular MOBC (Finney, 2018). If MOBCs can be directly manipulated, it may then be possible to optimize treatment by focusing on MOBCs that drive desirable outcomes.

Next, the current findings have implications for modifying current AUD treatments to focus on reducing negative affect and craving, and increase use of adaptive alcohol coping skills. Finally, over and above the additive effects of craving, negative and positive affect, and adaptive alcohol coping, these variables may operate jointly in their prediction of drinking. For example, the risks of craving and negative affect on drinking may be mitigated by high levels of adaptive alcohol coping.

Conclusion

This study advances the field of cognitive-behavioral alcohol treatment by depicting the 84 consecutive daily during-treatment change trajectories of alcohol use (abstinence and heavy drinking), four MOBCs (craving, negative affect, positive affect, alcohol coping) of such changes, and the time-varying associations between alcohol drinking and MOBCs. These findings have the potential to help optimize the future AUD treatments by reducing both negative affect and craving. Future research should also focus on gaining a better understanding of the interactive effects of these MOBCs.

Public health significance statements.

This study highlights the importance of reducing both negative affect and craving when treating negative affect drinkers with moderate to severe alcohol use disorder (AUD).

Acknowledgements

Research reported in this article was supported by the National Institute on Alcohol Abuse and Alcoholism (NIAAA) grants R01 AA024628 and R01 AA024628-S1 awarded to Paul R. Stasiewicz and Clara M. Bradizza, and T32 AA007583 awarded to Charles LaBarre. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.

Footnotes

1

Two sets of multi-group TVEM were conducted. Despite being conservative (e.g., Afshartous & Preston, 2010), the overlapping 95% confidence intervals for the separate means suggest that the two treatment groups did not differ for either alcohol abstinence or heavy drinking.

University IRB procedures and ethics with participants were followed. Data are available upon request.

References

  1. Afshartous D, & Preston RA. (2010). Confidence intervals for dependent data: Equating non-overlap with statistical significance. Computational Statistics & Data Analysis, 54, 2296–2305. 10.1016/j.csda.2010.04.011. [DOI] [Google Scholar]
  2. Annis HM, & Graham JM. (1995). Profile types on the inventory of drinking situations - implications for relapse prevention counseling. Psychology of Addictive Behaviors, 9, 176–182. 10.1037/0893-164x.9.3.176 [DOI] [Google Scholar]
  3. Arbuckle JL. (1996). Full information in the presence of incomplete data. In Marcoulides GA & Schumacker RE (Eds.), Advanced Structural Equation Modeling: Issues and Techniques (pp. 243–277). Lawrence Erlbaum Associates. [Google Scholar]
  4. Baker TB, Piper ME, McCarthy DE, Majeskie MR, & Fiore MC. (2004). Addiction motivation reformulated: An affective processing model of negative reinforcement. Psychological Review, 111(1), 33–51. 10.1037/0033-295X.111.1.33 [DOI] [PubMed] [Google Scholar]
  5. Bresin K, Mekawi Y, & Verona E. (2018). The effect of laboratory manipulations of negative affect on alcohol craving and use: A meta-analysis. Psychology of Addictive Behaviors : Journal of the Society of Psychologists in Addictive Behaviors, 32(6), 617–627. 10.1037/adb0000383 [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Bureau of Labor Statistics. (2022). Usual weekly earnings of wage and salary workers first quarter 2022. U.S. Department of Labor. https://www.bls.gov/news.release/pdf/wkyeng.pdf [Google Scholar]
  7. Cohen B. (2001). Explaining Psychological Science (2nd ed.). New York: John Wiley & Sons. [Google Scholar]
  8. Cohen J, Cohen P, West SG, & Aiken LS. (2003). Applied multiple regression/correlation analysis for the behavioral sciences (3rd ed.). Routledge. [Google Scholar]
  9. Dermody SS, & Shiffman S. (2020). The time-varying effect of alcohol use on cigarette smoking relapse risk. Addictive Behaviors, 102, 106192. 10.1016/j.addbeh.2019.106192 [DOI] [PubMed] [Google Scholar]
  10. Derogatis LR. (1993). The Brief Symptom Inventory (BSI) administration, scoring, and procedures manual. National Computer Systems. [Google Scholar]
  11. Dziak JJ, Li R, Tan X, Shiffman S, & Shiyko MP. (2015). Modeling intensive longitudinal data with mixtures of nonparametric trajectories and time-varying effects. Psychological Methods, 20(4), 444–469. 10.1037/met0000048 [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Finney JW. (2018). Identifying mechanisms of behavior change in psychosocial alcohol treatment trials: Improving the quality of evidence from mediational analyses. Journal of Studies on Alcohol & Drugs, 79(2), 163–170. 10.15288/jsad.2018.79.163 [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Hallgren KA, Wilson AD, & Witkiewitz K. (2018). Advancing analytic approaches to address key questions in mechanisms of behavior change research. Journal of Studies on Alcohol & Drugs, 79(2), 182–189. 10.15288/jsad.2018.79.182 [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Hastie T, & Tibshirani R. (1993). Varying-coefficient models. Journal of the Royal Statistical Society. Series B ( Methodological), 55(4), 757–779. 10.1111/j.2517-6161.1993.tb01939.x [DOI] [Google Scholar]
  15. Jahng S, Wood PK, &Trull TJ. (2008). Analysis of affective instability in ecological momentary assessment: Indices using successive difference and group comparison via multilevel modeling. Psychological Methods, 13, 354–375. 10.1037/a0014173. [DOI] [PubMed] [Google Scholar]
  16. Kadden R, Carroll K, Donovan D, Cooney N, Monti P, Abrams D, ... & Hester R. (1995). Cognitive-Behavioral Coping Skills Therapy Manual: A Clinical Research Guide For Therapists Treating Individuals With Alcohol Abuse And Dependence (NIH Publication No. 94–3724, Project MATCH Monograph Series, Vol. 3). National Institute on Alcohol Abuse and Alcoholism. [Google Scholar]
  17. Kahler CW, Lachance HR, Strong DR, Ramsey SE, Monti PM, Brown RA. (2007). The commitment to quitting smoking scale: initial validation in a smoking cessation trial for heavy social drinkers. Addictive Behavior, 2420–2424. 10.1016/j.addbeh.2007.04.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Kazdin AE, & Nock MK. (2003). Delineating mechanisms of change in child and adolescent therapy: Methodological issues and research recommendations. Journal of Child Psychology & Psychiatry & Allied Disciplines, 44(8), 1116–1129. 10.1111/1469-7610.00195 [DOI] [PubMed] [Google Scholar]
  19. Kazdin AE. (2007). Mediators and mechanisms of change in psychotherapy research. Annual Review of Clinical Psychology, 3, 1–27. 10.1146/annurev.clinpsy.3.022806.091432 [DOI] [PubMed] [Google Scholar]
  20. Kiluk BD, Nich C, Babuscio T, & Carroll KM. (2010). Quality versus quantity: Acquisition of coping skills following computerized cognitive behavioral therapy for substance use disorders. Addiction, 105(12), 2120–2127. 10.1111/j.1360-0443.2010.03076.x [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Lanza ST, & Linden-Carmichael AN. (2021). Time-varying effect modeling for the behavioral, social, and health sciences. Springer. [Google Scholar]
  22. Li R, Dziak JJ, Tan X, Huang L, Wagner AT, & Yang J. (2017). TVEM (time-varying effect model) SAS macro users’ guide (Version 3.1.1). University Park: The Methodology Center, Penn State. http://methodology.psu.edu [Google Scholar]
  23. Linn BK, Stasiewicz PR, Zhao J, Lucke JF, Ruszczyk MU, LaBarre C, & Bradizza CM. (2021). The shape of change: Determining when mechanisms of behavior change are active in cognitive behavioral treatment for alcohol use disorder using time-varying effect modeling (TVEM). Journal of Studies on Alcohol and Drugs, 82(5), 629–637. 10.15288/jsad.2021.82.629 [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Linn BK, Zhao J, Stasiewicz PR, LaBarre C, Wilding GE, & Bradizza CM. (2023). Relationship of negative emotionality, NIAAA recovery, and 3- and 6-month drinking outcomes among adults in treatment for alcohol use disorder. Drug and Alcohol Dependence, 242, 109695. 10.1016/j.drugalcdep.2022.109695 [DOI] [PubMed] [Google Scholar]
  25. Liu LM. (1986). Multivariate Time Series Analysis using Vector ARMA Models. Scientific Computing. [Google Scholar]
  26. Longabaugh R, Donovan DM, Karno MP, McCrady BS, Morgenstern J, & Tonigan JS. (2005). Active ingredients: How and why evidence-based alcohol behavioral treatment interventions work. Alcoholism: Clinical and Experimental Research, 29(2), 235–247. 10.1097/01.ALC.0000153541.78005.1F [DOI] [PubMed] [Google Scholar]
  27. Magill M, Kiluk BD, McCrady BS, Tonigan JS, & Longabaugh R. (2015). Active ingredients of treatment and client mechanisms of change in behavioral treatments for alcohol use disorders: Progress 10 years later. Alcoholism, Clinical and Experimental Research, 39(10), 1852–1862. 10.1111/acer.12848 [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Magill M, Tonigan JS, Kiluk B, Ray L, Walthers J, & Carroll K. (2020). The search for mechanisms of cognitive behavioral therapy for alcohol or other drug use disorders: A systematic review. Behaviour Research & Therapy, 131, 1–10. 10.1016/j.brat.2020.103648 [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Miller WR, & Wilbourne PL. (2002). Mesa Grande: A methodological analysis of clinical trials of treatments for alcohol use disorders [Article]. Addiction, 97(3), 265–277. 10.1046/j.1360-0443.2002.00019.x [DOI] [PubMed] [Google Scholar]
  30. Morgenstern J, Kuerbis A, Houser J, Muench FJ, Shao S, & Treloar H. (2016). Within-person associations between daily motivation and self-efficacy and drinking among problem drinkers in treatment. Psychology of Addictive Behaviors, 30(6), 630–638. 10.1037/adb0000204 [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. National Institute on Alcohol Abuse and Alcoholism. (2019). 2019 National Survey on Drug Use and Health. https://www.niaaa.nih.gov/publications/brochures-and-fact-sheets/alcohol-facts-and-statistics
  32. Petry NM, Litt MD, Kadden R, & Ledgerwood DM. (2007). Do coping skills mediate the relationship between cognitive-behavioral therapy and reductions in gambling in pathological gamblers? Addiction, 102(8), 1280–1291. 10.1111/j.1360-0443.2007.01907.x [DOI] [PubMed] [Google Scholar]
  33. Roos CR, Maisto SA, & Witkiewitz K. (2017). Coping mediates the effects of cognitive-behavioral therapy for alcohol use disorder among out-patient clients in Project MATCH when dependence severity is high. Addiction, 112(9), 1547–1557. 10.1111/add.13841 [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Sayette MA. (2016). The role of craving in substance use disorders: Theoretical and methodological issues. Annual Review of Clinical Psychology, 12, 407–433. 10.1146/annurevclinpsy-021815-093351 [DOI] [PubMed] [Google Scholar]
  35. Schiffman SA, Stone AA, & Hufford MR. (2008). Ecological Momentary Assessment. Annual Review of Clinical Psychology, 4, 1–32. 10.1146/annurev.clinpsy.3.022806.091415 [DOI] [PubMed] [Google Scholar]
  36. Serre F, Fatseas M, Swendsen J, Auriacombe M. (2015). Ecological momentary assessment in the investigation of craving and substance use in daily life: A systematic review. Drug and Alcohol Dependence, 148, 1–20. 10.1016/j.drugalcdep.2014.12.024 [DOI] [PubMed] [Google Scholar]
  37. Schafer JL. (1997). Analysis of incomplete multivariate data. CRC press. [Google Scholar]
  38. Shiyko M, Lanza S, Tan X, Li R, & Shiffman S. (2012). Using the time-varying effect model (TVEM) to examine dynamic associations between negative affect and self confidence on smoking urges: Differences between successful quitters and relapsers. Prevention Science, 13(3), 288–299. 10.1007/s11121-011-0264-z [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Silins E, Horwood LJ, Patton GC, Fergusson DM, Olsson CA, Hutchinson DM, Spry E, Toumbourou JW, Degenhardt L, & Swift W. (2014). Young adult sequelae of adolescent cannabis use: An integrative analysis. The Lancet Psychiatry, 1(4), 286–293. [DOI] [PubMed] [Google Scholar]
  40. Snippe E, Dziak JJ, Lanza ST, Nyklíček I, & Wichers M. (2017). The shape of change in perceived stress, negative affect, and stress sensitivity during mindfulness-based stress reduction. Mindfulness, 8(3), 728–736. 10.1007/s12671-016-0650-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Stasiewicz PR, Bradizza CM, Schlauch RC, Coffey SF, Gulliver SB, Gudleski GD, & Bole CW. (2013). Affect regulation training (ART) for alcohol use disorders: Development of a novel intervention for negative affect drinkers. Journal of Substance Abuse Treatment, 45, 433–443. 10.1016/j.jsat.2013.05.012 [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Stasiewicz PR, Bradizza CM, & Slosman KS. (2018). Emotion regulation treatment of alcohol use disorders: Helping clients manage negative thoughts and feelings. Routledge. [Google Scholar]
  43. Tan X, Shiyko MP, Li R, Li Y, & Dierker L. (2012). A time-varying effect model for intensive longitudinal data. Psychological Methods, 17(1), 61–77. 10.1037/a002581410.1037/a0025814.supp (Supplemental) [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Unrod M, Drobes DJ, Stasiewicz PR, Ditre JW, Heckman B, Miller RR, Sutton SK, & Brandon TH. (2014). Decline in cue-provoked craving during cue exposure therapy for smoking cessation. Nicotine & Tobacco Research, 16(3), 306–315. 10.1093/ntr/ntt145 [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Watson D, Clark LA, & Tellegen A. (1988). Development and validation of brief measures of positive and negative affect: The PANAS Scales. Journal of Personality & Social Psychology, 54(6), 1063–1070. doi: 10.1037/0022-3514.54.6.1063 [DOI] [PubMed] [Google Scholar]
  46. Wei LJ. (1978). Application of an Urn Model to Design of Sequential Controlled Clinical-Trials. Journal of the American Statistical Association, 73(363), 559–563. 10.2307/2286600 [DOI] [Google Scholar]
  47. Witkiewitz K, Hallgren KA, Kranzler HR, Mann KF, Hasin DS, Falk DE, Litten RZ, O’Malley SS, & Anton RF. (2017). Clinical validation of reduced alcohol consumption after treatment for alcohol dependence using the World Health Organization risk drinking levels. Alcoholism: Clinical and Experimental Research, 41(1), 179–186. 10.1111/acer.13272 [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Yang J, Shao M, & Cai G. (2019). On the performance of MixTVEM: A simulation study. Communications in Statistics-Simulation and Computation, 48(9), 2830–2844. 10.1080/03610918.2018.1468458 [DOI] [Google Scholar]
  49. Zhao J, Linn BK, Stasiewicz PR, Wilding GE, LaBarre C, & Bradizza CM. (2022). Predictors of alcohol use during treatment: A time-varying effect modeling (TVEM) analysis using 84 days of daily ecological momentary assessment (EMA) data. Alcoholism: Clinical and Experimental Research, 46 (S1), 70–289. [Google Scholar]
  50. The manuscript is based on data also used in a conference poster abstract that has been published in Zhao et al. (2022).

RESOURCES