Effects of Optimism and Stage of Change on Alcohol Use and Problems among Sexual Minority Men with HIV participating in a Brief Motivational Interviewing Intervention

Abstract

Introduction:

Disseminating effective alcohol interventions for sexual minority men (SMM) with HIV remains a crucial public health endeavor. Motivational interviewing (MI) interventions are an established approach to reducing alcohol use, yet more research is needed to determine predictors of MI treatment outcomes and underlying mechanisms related to sustained behavior change among SMM with HIV. This pre-registered secondary analysis tested whether action-related stage of change mediated effects of a MI intervention on future alcohol use and problems among SMM with HIV, and whether individual differences in trait optimism moderated these associations.

Methods:

SMM with HIV who engaged in frequent alcohol use (N=180) were randomized to MI or assessment-only treatment as usual (TAU). Participants completed a semi-structured Timeline Follow-Back interview to measure past-month alcohol use as well as self-reports assessing stage of change, trait optimism, and alcohol problems at baseline and 3- and 12-months post-baseline.

Results:

Structural equation models controlling for baseline alcohol use and stage of change indicated that 3-month action significantly mediated effects of MI on 12-month drinks per week. Likewise, the indirect effect of 3-month action was moderated by higher levels of trait optimism. When employment status, education level, and annual family/household income were included as covariates in the model, being employed significantly predicted 12-month alcohol use, and mediation and moderated mediation effects were no longer statistically significant. Stage of change did not mediate effects of MI on 12-month alcohol problems, and this indirect effect was not moderated by trait optimism.

Conclusions:

The present study provides further evidence supporting action-related stage of change as a mechanism linking MI to alcohol use reductions. Results demonstrated that SMM with HIV who were more optimistic tended to take more action towards reducing their alcohol use and suggest that MI-based interventions may benefit from integrating components aimed at augmenting patients’ optimism. Yet, covarying for current economic status substantially impacted findings and underscores the need to critically consider how broader socioecological contexts can impact treatment outcomes.

1. Introduction

Sexual minority men (SMM; inclusive of men who have sex with men and/or identify as gay, bisexual, or otherwise not heterosexual) accounted for 67% of new HIV infections in 2021 and continue to have the highest HIV incidence and prevalence in the United States (Centers for Disease Control and Prevention, 2023). SMM are more likely to meet criteria for alcohol use disorder (AUD) than heterosexual men with or without HIV (McCabe et al., 2019; Crane et al., 2017; Womack et al., 2019). Moreover, alcohol is an established barrier across the HIV care continuum (for review see Vagenas et al., 2015) and its deleterious effects on HIV-related outcomes are well-documented (e.g., Hendershot et al., 2009; Kahler et al., 2017). Thus, developing and delivering effective alcohol interventions tailored for SMM with HIV remains a public health priority.

Motivational interviewing (MI), which uses person-centered and directive techniques to catalyze behavior change (Miller & Rollnick, 2002), is an empirically validated approach to reducing alcohol use across many populations (Wray et al., 2016; Borsari & Carey, 2000; McDevitt-Murphy et al., 2014) and may have pragmatic advantages for persons with HIV generally, and SMM specifically. For example, due to its relative brevity versus cognitive behavioral therapy, MI can be more readily integrated into existing HIV-related care components. Further, MI-based interventions are rooted in “an atmosphere of acceptance and compassion” (Miller & Rollnick, 2013, p. 29), a crucial element when considering that sexual identity- and HIV-related stigmas persist across various socioecological levels (Bogart et al., 2008; Stangl et al., 2023). Of particular note, MI-based interventions have been shown to increase pre-exposure prophylaxis adherence (Dangerfield et al., 2023), decrease heavy episodic alcohol use in persons with HIV (Kahler et al., 2018), and reduce frequency of condomless sex (Monti et al., 2016), although evidence of MI’s effects relative more limited and scripted brief intervention in SMM with HIV is equivocal (Kahler et al., 2024). Despite their promise, MI and MI-based interventions typically produce modest reductions in alcohol use that dissipate over time (for reviews see Huh et al., 2015; Vasilaki et al., 2006). Likewise, very few studies implementing alcohol-specific MI have focused on SMM with HIV. As such, more research is needed to determine salient predictors of MI treatment outcomes and underlying mechanisms related to sustained behavior change in this population (Pantalone et al., 2020).

Given the emphasis within MI to enhance intrinsic motivation for change, increasing one’s action-oriented change behavior is a logical mechanism underlying MI’s potential to impact subsequent alcohol-related behaviors. The Transtheoretical Model (Prochaska & DiClemente, 1983) posits that behavior change occurs along a continuum, with three core stages including precontemplation (i.e., initial ambivalence with no immediate intentions for behavior change), contemplation (i.e., problem recognition with intentions for behavior change some time in the future), and action (i.e., active engagement in behavior change). Despite a strong theoretical rationale (Magill et al., 2015; Miller & Rose, 2009; Borsari et al., 2009), empirical evidence supporting the action stage of change as a mechanism of change is equivocal. For example, among young adult college students, results from a meta-analysis using individual participant data and systematic review of 61 alcohol intervention randomized controlled trials (RCTs) found no overall effect of brief MI-based interventions on increased motivation for alcohol-related change (Tan et al., 2023) and that motivation for alcohol-related change did not mediate effects of intervention on subsequent alcohol use reductions (Reid & Carey, 2015), respectively.

Alternatively, two separate studies using data from the United Kingdom Alcohol Treatment Trial (UKATT Research Team, 2005) found that higher post-treatment progression in the stages of change predicted more days abstinent and fewer drinks per drinking day at 9- and 12-month follow-ups among primarily middle-aged adults (Heather & McCambridge, 2013; Cook et al., 2015). Moreover, higher scores in the action stage of change predicted lower alcohol use and problems three months later, regardless of intervention condition, in a sample of non-treatment-seeking adults receiving a brief motivational intervention (Richards et al., 2020). Using those same data, Field and colleagues (2020) also found that participants randomized to the brief motivational intervention plus telephone booster condition reported higher 3-month scores in the action stage of change which, in turn, predicted less alcohol use and fewer problems at the 6- and 12-month follow-ups. Taken together, despite some support that greater progression through the stages of change is a relevant mechanism for MI-based interventions, no investigations to our knowledge have examined whether stage of change mediates MI effects on future alcohol outcomes among persons with HIV generally or SMM specifically. Additionally, little research has examined individual difference factors that may moderate these relations.

Individual differences in trait optimism may be a key factor related to MI’s ability to help individuals take actionable steps towards behavior change. Broadly characterized as positive expectations about future outcomes (Hoeppner et al., 2021; Scheier & Carver, 1987), greater optimism is associated with better health and well-being, at least in part because those with higher optimism tend to engage in more proactive problem solving and show better adaptive coping skills, such as seeking social support and focusing on the positives during stressful life circumstances (Carver et al., 2010; Mink et al., 2014). Prior research shows that optimism is associated with slower HIV-related disease progression (Ironson & Hayward, 2008), and that HIV-related stigma and discrimination are negatively associated with optimism (Ammirati et al., 2015). Thus, for members of marginalized groups (e.g., SMM with HIV), alcohol-focused MI may more effectively increase motivation that leads to sustained behavior change among individuals who are more optimistic. Yet, to our knowledge, no published study has examined optimism, stages of change, and alcohol use among SMM with HIV or in the context of MI.

The present secondary analysis prospectively examined whether greater action, based on the stages of change, mediated the effect of MI on alcohol use and problems among a sample of SMM with HIV. In addition, we examined whether intervention effects on the progression towards more action-specific stages of change were more pronounced for those with higher levels of trait optimism at baseline. We hypothesized that (1) participants randomized to the treatment (vs. control) condition would report more action-specific change at a 3-month follow-up which, in turn, would relate to less alcohol use and fewer problems at a 12-month follow-up; and (2) this indirect effect would be stronger for those with higher levels of baseline optimism.

2. Material and methods

2.1. Transparency and openness

This study preregistered the design, hypotheses, and analysis plan (https://osf.io/w9e4z/) after data collection was completed. Materials and analysis code for this study are available by emailing the corresponding author. This secondary analysis used data collected from a larger RCT (ClinicalTrials# NCT01328743). Findings from these data have been reported previously (Zelaya et al., 2022; Guy et al., 2022; Surace et al., 2022; Kahler et al., 2024).

2.2. Participants and procedures

Study data come from an RCT testing the effects of MI that was tailored for delivery to SMM with HIV who engaged in heavy alcohol use (N = 180; Kahler et al., 2018). Participants were recruited from a community health center that specialized in the care of sexual and gender minority patients in the northeastern U.S. from December 2011 through March 2015. Eligibility criteria included: (a) being ≥ 18 years of age; (b) male sex assigned at birth; (c) reporting ≥ 14 drinks per week or ≥ 5 drinks on one occasion 1+ times in a typical month; (d) currently receiving HIV care; and (e) reporting past-year engagement in oral or anal sex with a male partner, or self-identifying as gay/bisexual. Exclusion criteria included: (a) current intravenous drug use; (b) current psychosis, suicidality, or mania at the baseline assessment; (c) past 3-month treatment for an HIV-related opportunistic infection, and; (d) currently receiving treatment for alcohol and/or drug concerns. The parent study was approved by the Brown University and Fenway Health Institutional Review Boards.

After an initial preliminary screener questionnaire, eligible participants were invited to an in-person baseline appointment. Participants provided written informed consent followed by structured clinical interviews and self-report assessments. Participants were then randomized to receive MI in addition to their ongoing HIV treatment or only their ongoing HIV treatment as usual (TAU). Participants in the TAU condition did not receive any additional counseling and were informed that their participation in the baseline appointment was complete. Those in MI immediately received a brief (up to 60 minutes) counseling session that integrated the core elements from the FRAMES (i.e., Feedback, Responsibility, Advice, Menu Options, Empathy and Self-Efficacy) approach (Miller & Rollnick, 2002) and included components focused on alcohol-related normative and personalized feedback, other health related information concerning HIV, and alcohol/non-alcohol-related personal goals. Participants in the intervention condition also received brief telephone check-ins approximately two weeks later and in-person booster sessions three and six months postbaseline. All participants completed the same structured clinical interviews and self-report assessments during in-person appointments at 3, 6, and 12 months postbaseline. Participants received up to $250 for completing all study appointments. Combined retention rates at the 3- and 12-month appointments were 93% and 89%, respectively. The present study used data from the baseline, 3-, and 12-month appointments.

2.3. Measures

2.3.1. Demographic characteristics and socioeconomic position

Participants reported on their age, sex assigned at birth, gender identity (given inclusion criteria, all participants were cisgender men), race, and ethnicity. Socioeconomic position was assessed by level of education, total family/household annual income, and employment status via three single items. Education level was defined as the highest grade or degree completed, rated on a 10-point scale ranging from “None” to “Graduate or Professional Degree”. Total family/household income was measured with 10 response options ranging from “$0-$9,999” to “$100,000 or more” in standard $10,000 increments. Employment status included response options of “Unemployed”, “Employed part-time (1–30 hours/week)”, “Employed full-time (30+ hours/week)”, “Full-time student”, “Homemaker”, “Part-time student”, and “Retired”. For analytic purposes, employment was operationalized as a binary variable (i.e., employed [part- or full-time] or not employed [homemaker, retired, part- or full-time student]).

2.3.2. Trait optimism

The Revised Life Orientation Test (LOT-R; Scheier et al., 1994) measured trait optimism. The LOT-R consists of 10 items rated on a 5-point scale ranging from Disagree a lot to Agree a lot, with higher score indicating greater optimism (e.g., “Overall, I expect more good things to happen to me than bad.”). The LOT-R contains four filler items and, of the remaining six items, three capturing pessimism are reversed scored. The LOT-R is a valid and reliable measure (Chiesi et al., 2013) that has been previously used in studies with SMM (Hickson et al., 2022). All analyses used a sum score, and this measure demonstrated good internal consistency reliability in the present study (α = .81).

2.3.3. Stage of change

The Readiness to Change Questionnaire (RCQ; Rollnick et al., 1992) is a reliable and validated measure (Forsberg et al., 2003) that assessed baseline stage of change and contains 12 items rated on a 5-point scale ranging from Strongly disagree (−2) to Strongly agree (+2). Based on the Transtheoretical Model (Prochaska & DiClemente, 1983), the RCQ includes three subscales with four items each pertaining to Precontemplation (e.g., “I don’t think I drink too much”), Contemplation (e.g., “Sometimes I think I should cut down on my drinking”) and Action (e.g., “I am actually changing my drinking habits right now”) stages of change. Sum scores for each subscale ranged from −8 to 8, and participants were assigned to a specific stage based on the scale with the highest sum score at baseline (i.e., “quick method”; Heather & Hönekopp, 2008). Participants were assigned to the stage of change farthest along the continuum (i.e., Action > Contemplation > Precontemplation) when they had equal scores across multiple subscales.

The RCQ subscales demonstrated acceptable-to-good internal consistency reliability at baseline (Precontemplation α = .69; Contemplation α = .80; Action α =.85) and 3-months (Action α = .89) in the present study. Consistent with prior longitudinal research examining stage of change and alcohol use in the context of treatment (Cook et al., 2015), participants’ baseline stage of change was characterized dichotomously (i.e., action vs. pre-action [precontemplation or contemplation]).

2.3.4. Alcohol use indices

Drinks per week and alcohol-related problems were assessed at baseline and 12-month follow-up. The Timeline Follow Back (TLFB; Sobell & Sobell, 1992) assessed past-month use frequency and quantity to calculate participants’ average number of drinks per week. For participants without complete data, drinks per week was defined as the number of alcoholic drinks consumed divided by the number of weeks of available data. The Short Inventory of Problems (SIP; Miller et al., 1995) assessed past-month alcohol-related problems. The SIP contains 15 items rated on a 4-point scale ranging from Never to Daily or Almost Daily, with higher scores indicating more alcohol-related problems (e.g., “I have been unhappy because of my drinking”). The SIP is a reliable and valid measure (Alterman et al., 2009) and demonstrated excellent internal consistency reliability at baseline and 12-months in the present study (αs = .95).

2.4. Analytic plan

2.4.1. Preliminary analyses

Preliminary analyses used SPSS version 29 to examine variable distributions and normality assumptions, and test bivariate correlations among sociodemographic variables, trait optimism, treatment condition, baseline RCQ stage of change (i.e., action vs. pre-action), 3-month RCQ action subscale scores, and alcohol use indices (i.e., baseline and 12-month drinks per week and problems). Chi-square tests and independent samples t-tests examined whether baseline stage of change and 3-month RCQ action subscale scores significantly differed among participants randomized to TAU or MI. Trait optimism was grand-mean centered and multiplied by a centered binary treatment condition variable (−1 = TAU, 1 = MI) to create an interaction term for use in subsequent path analyses.

2.4.2. Prospective mediation models

Path analyses tested indirect effects of treatment condition on 12-month drinks per drinking day and alcohol-related problems via 3-month action (i.e., RCQ action subscale sum score) using maximum-likelihood (ML) estimation and the MODEL INDIRECT command in Mplus 8.1 (Muthén & Muthén, 1998–2018). Path analyses also tested whether trait optimism (w) moderated the effect of treatment condition on 3-month action (a path) and the association of this indirect effect on 12-month alcohol use indices (b paths). We used the MODEL CONSTRAINT command to calculate the overall index of the moderated mediation effect (a₃*b₁) and conditional indirect effects (a₁*b₁ + a₃* b₁*w) at −1 SD, mean, and +1 SD levels of w (i.e., trait optimism). Loop plots were used to probe conditional indirect effects of treatment condition and high/medium/low levels of trait optimism on 12-month alcohol use indices via 3-month action. Covariates included baseline stage of change, education level, employment status, and annual family/household income. Likewise, models also covaried baseline levels of the corresponding 12-month alcohol use outcome.

All exogenous variables were allowed to correlate with each other. Following established recommendations (Rucker et al., 2011), 10,000 bootstrapped estimates computed 95% bias-corrected confidence intervals (CI) for the main and indirect effects to account for non-normal parameter estimates and mediated effects obtained by the product of coefficients (MacKinnon, 2012). Indirect effects with confidence intervals that do not contain the value zero are statistically significant (Hayes, 2017), regardless of the total effect’s statistical significance (Rucker et al., 2011; Hayes, 2009).

Using a stepped approach, we initially tested path models predicting 12-month drinks per week and alcohol-related problems adjusting for baseline levels of the dependent variable and stage of change (Models 1A and 2A, respectively). We then added education level, employment status, and annual family/household income to account for factors related to trait optimism and isolate its effect on stage of change (Models 1B and 2B, respectively). Acceptable model fit was determined via Root Mean Square Error of Approximation (RMSEA) and Standardized Root Mean Square Residual (SRMR) values ≤ 0.08, and Comparative Fit Index (CFI) values ≥ 0.90 (Hu & Bentler, 1999).

3. Results

3.1. Preliminary analyses

Sample descriptives are presented in Table 1. The proportion of participants in action vs. pre-action stages of change at baseline did not significantly differ between treatment conditions, χ²(1, 179) = 1.83, p = .176. Alternatively, using contemporary guidelines based on Funder and Ozer (2019), there was a medium-to-large effect size difference in 3-month RCQ action subscale scores among participants in the MI condition (M = 2.91, SD = 3.77) compared to those in TAU (M = 0.29, SD = 4.14), t(165) = 4.26, p < .001; Cohen’s d = 0.66.

Table 1.

Sample Demographics, Personality, and Alcohol-Related Characteristics

n(%) or M (SD), sample range	Full Sample N=180	Motivational Interviewing (MI) n = 89	Treatment as Usual (TAU) n = 91
Sociodemographics
Age, years	42.11 (10.43), 20 – 66	41.03 (10.26), 20 – 60	43.16 (10.55), 20 – 66
Race / Ethnicity
Asian, Non-Hispanic/Latine	1 (0.6)	0	1 (1.1)
Black, Hispanic/Latine	4 (2.2)	2 (2.2)	2 (2.2)
Black, Non-Hispanic/Latine	31 (17.2)	12 (13.5)	19 (20.9)
White, Hispanic/Latine	14 (7.8)	9 (10.1)	5 (5.5)
White, Non-Hispanic/Latine	113 (62.8)	58 (65.2)	55 (60.4)
More than 1 race	9 (5.0)	4 (4.5)	5 (5.5)
Employment status, % employeda	113 (62.7)	47 (52.8)	58 (63.8)
Annual family/household income, USD
$19,999 or less	56 (31.1)	31 (34.8)	25 (27.5)
$20,000 – 39,999	41 (22.8)	19 (21.4)	22 (24.2)
$40,000 – 59,999	24 (13.4)	11 (12.4)	13 (14.3)
$60,000 – 79,999	12 (6.7)	5 (5.6)	7 (7.7)
$80,000 or higher	47 (26.1)	23 (25.8)	24 (26.4)
Level of education
≤ High school grad/GED	27 (15.0)	14 (15.7)	13 (14.3)
Some post-high school education	69 (38.3)	34 (38.2)	35 (38.5)
College graduate	84 (46.7)	41 (46.1)	43 (47.3)
Trait Optimism (LOT-R)	14.75 (5.28), 0 – 24	14.84 (5.58), 0 – 24	14.66 (4.99), 0 – 24
Stage of Change
Precontemplation (baselineb/3-monthc)	44 (24.6) / 45 (26.9)	17 (19.1) / 12 (13.5)	27 (29.7) / 33 (36.3)
Contemplation (baselineb/3-monthc)	81 (45.3) / 41 (24.6)	41 (46.1) / 15 (16.9)	40 (44.0) / 26 (28.6)
Action (baselineb/3-monthc)	54 (30.2) / 81 (48.5)	31 (34.8) / 51 (58.4)	23 (25.3) / 29 (31.9)
Past-month Alcohol use and Problems
Avg. drinks per week (TLFB)
Baseline	20.77 (16.94), 1.40 – 152.68	21.50 (19.85), 2.10 – 152.68	20.06 (13.58), 1.40 – 55.48
12-monthd	12.70 (13.32), 0 – 78.89	10.46 (11.91), 0 – 62.24	14.85 (14.30), 0 – 78.79
SIP total score
Baseline	9.28 (8.51), 0 – 44	9.15 (8.02), 0 – 37	9.42 (9.01), 0 – 44
12-monthd	5.61 (7.20), 0 – 45	4.99 (7.28), 0 – 45	6.22 (7.11), 0 – 33

Notes:

^aEmployed = working part- or full-time;

^bn=179, n=89, and n=90 for full sample, MI, and TAU conditions, respectively;

^cn=167, n=79, and n=88 for full sample, MI, and TAU conditions, respectively;

^dn=161, n=79, and n=82 for full sample, MI, and TAU conditions, respectively. LOT-R = Lifetime Orientation Test-Revised (Scheier et al., 1994), full possible range is 0 – 24; TLFB = Timeline Follow Back (Sobell & Sobell, 1992); SIP = Short Inventory of Problems, full possible range is 0 – 60

Bivariate correlations among study variables are presented in Table 2. Baseline trait optimism evinced statistically significant, small-to-medium effect positive associations with education level (r = .17) and total family/household income (r = .22). Alternatively, trait optimism was not significantly associated with employment status, baseline stage of change (i.e., action vs. pre-action), 3-month RCQ action subscale scores, or 12-month alcohol use indices (rs ≤ .13). Three-month RCQ action subscale scores evinced statistically significant, small-to-medium effect bivariate relations with baseline stage of change (r = .29) and 12-month drinks per week (r = −.22) but not with alcohol-related problems.

Table 2.

Bivariate correlations among study variables

	1.	2.	3.	4.	5.	6.	7.	8.	9.	10.	11.	12.
Baseline
1. Age	--
2. Ethnoracial identitya	−.10	--
3. Education	.12	−.22**	--
4. Incomeb	.08	−.34**	.40**	--
5. Employment statusc	−.14	−.23**	.22**	.39**	--
6. Treatment conditiond	−.10	−.05	−.01	−.05	−.11	--
7. Trait optimisme	.16*	−.01	.17*	.22**	.02	.02	--
8. RCQ stage of changef	.09	.24**	−.06	−.11	−.16*	.10	.13	--
9. Drinks per weekg	.08	.05	−.21**	−.12	.02	.04	−.10	−.10	--
10. Alcohol-related problemsh	−.01	.12	−.21**	−.32**	−.20**	−.02	−.18*	−.06	.35**	--
Three-months
11. RCQ – Actioni	.09	.01	−.06	−.18*	−.25**	.32**	−.04	.29**	−.05	.14	--
Twelve-months
12. Drinks per weekg	−.05	−.02	−.01	.10	.23**	−.17*	.07	−.15	.48**	.05	−.22**	--
13. Alcohol-related problemsh	−.03	.06	−.17*	−.17*	−.17*	−.09	−.07	−.14	.16*	.55**	.09	.17*

Notes:

^*p<.05,

^**p<.01;

^aNon-Latinx White=1, Minoritized Ethnoracial Identity=0;

^bTotal annual family/household income;

^cEmployed=1, Not employed=0;

^dMotivational Interviewing=1; Treatment as Usual=0;

^eLOT-R total score;

^fBased on whether participants were characterized as in Action (1) or Pre-Action (0) based on Readiness to Change Questionnaire (RCQ; Rollnick et al., 1992) using the “quick method” (Heather & Hönekopp, 2008);

^gPast-month drinks per week from Timeline Follow Back (Sobell & Sobell, 1992);

^hShort Inventory of Problems total score;

ⁱRCQ-Action subscale total score

3.2. Hypothesis 1: Indirect effects of treatment condition on 12-month alcohol use and problems via 3-month stage of change

12-month drinks per week.

The hypothesized model fit the data well (Model 1A: χ²[10] = 13.25, p = .21; CFI = .96; RMSEA = .04, 90% CI (.00, .10); SRMR = .05; Model 1B: χ²[16] = 17.56, p = .35; CFI = .98; RMSEA = .02, 90% CI (.00, .08); SRMR = .05). In the base model without socioeconomic variables, and that adjusted for only baseline stage of change and baseline drinks per week, three-month RCQ action subscale scores significantly mediated the effect of treatment condition on 12-month drinks per week (Table 3, Model 1A). That is, participants in the MI condition (vs. TAU) reported higher levels of RCQ action at 3-months that, in turn, predicted fewer drinks per week at 12-months (Figure 1, Panel A).

Table 3.

Indirect Effects from Prospective Mediation and Moderated Mediation Models

Outcome: 12-month drinks per week (DPW)	Model 1A: Base Model	Model 1B: Socioeconomic Position
Treatment condition → 3-month RCQ action subscale → 12-mo. DPW	β (SE) = −0.04 (0.02) 95% CI = (−0.09, -0.01)	β (SE) = −0.02 (0.02) 95% CI = (−0.07, 0.01)
Treatment condition * Trait optimism → 3-month RCQ action subscale → 12-mo. DPW	B (SE) = −0.12 (0.06) 95% CI = (−0.27, -0.02)	B (SE) = −0.07 (0.05) 95% CI = (−0.20, 0.02)
Outcome 12-month alcohol-related problems (ARP)	Model 2A: Base Model	Model 2B: Socioeconomic Position
Treatment condition → 3-month RCQ action subscale → 12-mo. ARP	β (SE) = 0.02 (0.02) 95% CI = (−0.02, 0.06)	β (SE) = 0.01 (0.02) 95% CI = (−0.02, 0.06)
Treatment condition * Trait optimism → 3-month RCQ action subscale → 12-mo. ARP	B (SE) = 0.001 (0.002) 95% CI = (−0.001, 0.10)	B (SE) = 0.001 (0.002) 95% CI = (−0.002, 0.10)

Notes: Estimates for indirect effects reported are standardized (β); estimates for the index of moderated mediation effects are unstandardized (B). SE = standard error; 95% confidence intervals (CI) for indirect effects. Treatment condition (MI [Motivational Intervention] =1, TAU [HIV treatment as usual] = 0). Models 1A and 2A represent mediator models after adjusting for baseline stage of change (action [1] vs. pre-action [0]) and baseline levels of the 12-month outcome variable. Models 1B and 2B also includes education level, employment status, and annual family/household income as covariates. Significant indirect effects are in bold typeface.

Figure 1.

Models examining 3-month action scores (Readiness to Change Questionnaire; RCQ) as a mediator of the effect of treatment condition (Motivational Interviewing [MI]=1; Treatment as Usual [TAU]=0) on past-month average drinks per week (Panel A) and alcohol-related problems (Panel B) at the 12-month follow-up. Solid and dashed lines indicate standardized estimates from the base and socioeconomic-adjusted models, respectively. Base model covariates include baseline drinks per week or alcohol-related problems, and stage of change (action [1] vs. pre-action [0]). Socioeconomic-adjusted model covariates also include education level, employment status, and annual family/household income. Covariate estimates are not depicted in the figure for parsimony. *p<.05; **p<.01; ***p<.001.

The indirect effect was no longer statistically significant with education level, employment status, and annual family/household income in the model (Table 3, Model 1B). In this model, being employed (vs. not employed) significantly predicted 12-month drinks per week (β = 0.16, 95% CI [.03, .30]), whereas neither education level nor annual family/household income significantly predicted 3-month RCQ action subscale scores or 12-month drinks per week (Figure 1, Panel A).

12-month alcohol-related problems.

The hypothesized model fit the data well (Model 2A: χ²[10] = 14.03, p = .17; CFI = .96; RMSEA = .04, 90% CI (.00, .10); SRMR = .05; Model 2B: χ²[16] = 19.23, p = .26; CFI = .97; RMSEA = .03, 90% CI (.00, .08); SRMR = .05). While treatment condition significantly predicted 3-month RCQ action subscale scores (Figure 1, Panel B), the indirect effect of treatment condition on 12-month alcohol related problems via 3-month action was not statistically significant in the base or socioeconomic-adjusted models (Table 3, Models 2A and 2B). Only baseline alcohol-related problems significantly predicted 12-month alcohol-related problems in the base (β = 0.53, 95% CI [.36, .68]) and socioeconomic-adjusted models (β = 0.53, 95% CI [.34, .69]). Likewise, education level, employment status, and annual family/household income were not significantly associated with 3-month RCQ action subscale scores or 12-month alcohol-related problems.

3.3. Hypothesis 2: Indirect effects of stage of change moderated by trait optimism

12-month drinks per week.

The overall moderated mediation effect was statistically significant, but only in the base model without socioeconomic covariates included (Table 3, Models 1A and 1B). Results from the base model indicated that the indirect effect of treatment condition on 12-month drinks per week via 3-month action was moderated by baseline levels of optimism. Loop plots probed conditional indirect effects for models predicting 12-month drinks per week at low (i.e., −5.28 [−1 SD]), medium (i.e., mean = 0), and high (i.e., 5.28 [+1 SD]) levels of trait optimism (Figure 2). There were statistically significant conditional indirect effects at low (B = −0.09, SE = 0.06, 95% CI [−0.25, −0.01]), medium (B = −0.11, SE = 0.06, 95% CI [−0.26, −0.02]), and high (B = −0.14, SE = 0.08, 95% CI [−0.33, −0.03]) levels of trait optimism.

Figure 2.

Loop plot examining baseline optimism (LOT-R) as a moderator of the indirect effect of treatment condition (Motivational Interviewing [MI]=1; Treatment as Usual [TAU]=0) on past-month average drinks per week at the 12-month follow-up from the base model. Base model covariates include baseline drinks per week or alcohol-related problems, and stage of change (action [1] vs. pre-action [0]). Red line represents the unstandardized estimate for the overall moderated mediation effect at different values of optimism. Low (≤ −5.28), medium (0), and high (≥ 5.28) values on the x-axis represent baseline optimism at −1 SD, Mean, and +1 SD, respectively. Blue lines represent 95% CIs for the overall moderated mediation effect. Values were the unstandardized moderated mediation effect and 95% CIs do no cross 0 on the y-axis are statistically significant.

12-month alcohol-related problems.

The overall moderated mediation effect was not statistically significant in the base or socioeconomic-adjusted models (Table 3, Models 2A and 2B).

3.4. Post-hoc analyses predicting 12-month drinks per week

Given that mediation and moderated mediation were no longer supported after including socioeconomic covariates, we conducted post hoc analyses to attempt to ascertain how and why current economic status impacted study results. We first conducted chi-square tests to determine whether SMM with HIV who were in action vs. pre-action at baseline differed in employment status, and independent samples t-tests to determine whether baseline alcohol-related problems differed based on employment status. Results indicated that SMM with HIV in pre-action were significantly more likely to be employed than those in action, χ² (1, N=179) = 4.43, p=.035, and that participants who were employed (M=7.84, SD=7.36) reported significantly fewer baseline alcohol-related problems than those who were not employed (M=11.31, SD=9.59), t(132.61) = 2.63, p = .01.

Next, we tested three additional sets of path models. The first set of models adjusted for potentially relevant demographic factors (i.e., age, race, ethnicity) in addition to the preregistered covariates specified a priori. Age and ethnoracial identity (Non-Latinx White = 1; Minoritized ethnoracial identity = 0) were not significantly associated with 3-month action or 12-month drinks per week. Consistent with primary mediation analyses, the indirect effect of treatment condition on 12-month drinks per week via 3-month action was not statistically significant (Supplemental Table 1).

The second set of models removed the binary employment variable while retaining all other covariates specified a priori to determine whether null results were driven primarily by employment status or one’s broader socioeconomic position. Annual income was negatively associated with 3-month action and positively associated with 12-month drinks per week, whereas education level was not associated with the mediator or outcome variable. Consistent with primary mediation analyses, the indirect effect of treatment condition on 12-month drinks per week via 3-month action was not statistically significant (Supplemental Table 2).

The third set of models tested separate models with subsamples based on employment status (n = 105/75 participants who were/were not employed, respectively). Being in the MI condition significantly predicted greater 3-month action and lower 12-month drinks per week, but only for participants who were employed. Consistent with primary mediation analyses, neither the indirect effect of treatment condition on 12-month drinks per week via 3-month action nor the overall moderated mediation effect were statistically significant for either subgroup (Supplemental Table 3).

4. Discussion

The present study extends the literature concerning mechanisms of behavior change and the extent to which individual difference factors affected these processes in a sample of SMM with HIV enrolled in an RCT testing brief motivational interviewing versus an assessment-only HIV treatment as usual control. Results were partially consistent with a priori hypotheses and extend prior research in two primary ways. First, results demonstrate that MI may be effective at reducing alcohol use (but not necessarily alcohol problems) by increasing the extent to which SMM with HIV actively engaged in change efforts. These results align with prior research demonstrating that progressing towards more action-specific stages of change is a salient predictor of future alcohol outcomes among middle aged adults enrolled in a MI (Heather & McCambridge, 2013; Cook et al., 2015; Field et al., 2020). Second, this study provides initial evidence that the ability of MI to help SMM with HIV take greater action towards behavior change may depend, in part, on individual differences in trait optimism. However, both of these findings were impacted substantially by covarying current economic status, with both findings no longer remaining statistically significant.

We decided a priori to conduct analyses of mediation and moderation that accounted for variables that also might be related to optimism, in order to provide maximal clarity around the specific role for this variable. Considerable cross-sectional and prospective evidence, including from this study, indicates that optimism is positively associated with broad socioeconomic indicators – including, but not limited to, income, education, and employment – across diverse samples (e.g., Robb et al., 2009; Heinonen et al., 2006; Boehm et al., 2015). In the present study, optimism was significantly positively correlated with both education and income but was not correlated significantly with employment. At the same time, optimism was not significantly correlated with greater baseline or 3-month scores on the RCQ action subscale.

Conversely, those with higher income and who were employed tended to report less action-related behavior change at 3 months and more alcohol use at 12 months. Further, in post hoc analyses separated by employment status, the standardized effect of 3-month action on 12-month drinks per week was smaller for participants who were (β = −0.05) versus were not (β = −0.10) employed. Although randomization balanced employment and income across the two conditions, those who were employed reported greater 12-month drinks per week, and this robust prediction completely obscured any mediated or moderated effects of action-related behavior change. This reflects a challenge in randomized controlled trials attempting to assess mediation, because the level of the mediator is not randomized and factors that influence the mediator may impact outcomes outside of the intervention itself. A stratified analysis that examines mediation across levels of a moderator can help disentangle potential confounding of mediated effects, but in the current study, sample size limitations preclude a complete examination of mediation and moderated mediation across levels of employment or income. Furthermore, the parent study did not have a priori hypotheses about types of employment and alcohol outcomes, and therefore did not collect more detailed data about employment that might have helped explain these results (e.g., job type [Thompson & Pirmohamed, 2021]).

One potential – albeit speculative – explanation for discrepant results when comparing the base and socioeconomic-adjusted models is that participants who experienced fewer problems may have been less motivated to reduce their alcohol use because they remained able to fulfill their occupational roles. As such, SMM with HIV who engaged in frequent alcohol use at baseline while experiencing relatively fewer socioeconomic consequences (e.g, employment or income), may have had less interest in taking concrete action towards reducing their alcohol use over time. Indeed, post hoc analyses indicated that at baseline, participants in pre-action were more likely to be employed than those in action stages of change, and that those who were employed reported fewer baseline alcohol-related problems than those who were not employed. Although a higher socioeconomic position is often associated with lower rates of alcohol consumption, problems, and AUD (Calling et al., 2019; Mulia & Karriker-Jaffe, 2012), more research is needed to understand how these factors may influence active treatment ingredients within motivational interventions.

Relatedly, while the parent study did not assess protective behavioral strategies, how SMM with HIV consumed alcohol may be one potential reason why the intervention in the present study reduced alcohol use but not problems. That is, despite consuming few drinks per week, participants may not have frequently engaged in protective behavioral strategies known to reduce alcohol-related problems (e.g., alternating alcohol and nonalcoholic beverages). As such, MI-based interventions that implement components focused on protective strategies may be more effective at reducing both alcohol use and problems among SMM with HIV.

Although the present study results should be interpreted cautiously and warrant future replication, these findings do also have potential clinical and theoretical implications. As noted in a prior review on behavioral alcohol interventions, Magill et al. (2015) underscore the need to identify optimal conditions to engender behavior change within motivational interventions. In this regard, individuals with higher levels of optimism may be particularly responsive to MI given their positive outlook on life and confidence in being able to change behavior. Moreover, knowing whether or not a client has a generally optimistic worldview may help inform how a clinician tries to elicit change and sustain talk. For example, MI counselors could try to frame discussions around alcohol use in terms of gains or losses (e.g., emphasizing the benefits or costs of changing or not changing behavior, respectively; Gerend & Cullen, 2008) when providing personalized feedback in a way that helps empower patients to be more optimistic about the future due to changing their behavior now. Conversely, loop plots indicated that mediation was not supported at very low levels of optimism (i.e., approximately ≤1.3 standard deviations below the mean). Thus, MI-based interventions could integrate components aimed to increase optimism (e.g., Best Possible Self interventions; Meevissen et al., 2001) to help highly pessimistic individuals become motivated to change.

The present study had both strengths and limitations. First, these data came from a relatively large RCT focused on reducing alcohol use among SMM with HIV. Not only is research on effective interventions for this population lacking, but this secondary analysis sought to address a sizable literature gap by examining underlying mechanisms affecting treatment outcomes for SMM with HIV. Likewise, the parent study included validated measures of alcohol use and trait optimism over a 12-month period with approximately 90% retention rates throughout. However, sample characteristics limit the generalizability of study findings. Specifically, participants were all were cisgender SMM with HIV recruited exclusively from one community health center in the Northeast region of the United States, and most identified as Non-Latine and White. Thus, it remains unclear whether trait optimism and action-specific stage of change are germane factors related to MI efficacy. Indeed, future research should endeavor to examine relevant mechanisms of behavior change with persons with HIV from diverse racial and ethnic backgrounds and varying sexual identities. In addition, the parent trial only collected data on self-reported alcohol use. Future research would benefit by integrating technologies that can passively, and objectively, measure alcohol use in real time (e.g., biosensors).

Although it is not inherently a limitation, the RCQ action subscale broadly measures conscious action to change one’s alcohol use. It does not, however, provide details about the processes of change they are employing, such as those assessed by the Processes of Change questionnaire (DiClemente et al., 1991). Finally, the study focused on trait optimism. It is possible that state optimism at the time of the MI session may have had a stronger moderating effect. Indeed, emerging evidence suggests that rather than being an immutable trait, optimism can change over time and in response to positive or negative life events (Hoeppner et al., 2021; Chopik et al., 2015; Segerstrom, 2007). Thus, it is likely important to assess situationally-relevant and broader trait-based optimism. Specifically, if an individual believes that their circumstances are unlikely to materially change if they reduce their alcohol use, they may be unlikely to move from recognizing potential harms of drinking to taking concrete actions to change drinking.

This study extends the prior literature on mechanisms of behavior change and identifies potentially germane factors affecting these processes among SMM with HIV who received an alcohol-focused MI. Results demonstrate that SMM with HIV who were more optimistic tended to take more action towards reducing their alcohol use and underscore the need to critically consider how one’s broader socioecological context can impact treatment outcomes. If future studies replicate findings that optimism helps to potentiate greater action to change during MI, then interventions may also endeavor to integrate components aimed at augmenting patients’ optimism (e.g., visualization exercises focused on one’s best possible future self [Meevissen et al., 2011], episodic future thinking exercises based on individually-relevant goals [Voss et al., 2022]). Likewise, more research is needed to determine whether optimism affects other relevant mechanisms of change, which in turn, will help to inform ways to implement more effective interventions.