Effects of Alcohol Intoxication on Sexual Decision-Making among Men Who Have Sex with Men (MSM): Alcohol’s Influences on Self-Control Processes

Abstract

This experiment tested mechanisms linking alcohol intoxication and analogue determinants of condomless anal intercourse (CAI) in a sample of 257 men who have sex with men (MSM). The two mechanisms tested were implicit approach biases toward CAI stimuli and executive working memory. Participants were randomized to 3 conditions (water control, placebo, or alcohol) and following beverage administration completed a working memory task, an Approach Avoidance Task of sexual vs. condom stimuli, and two video role-play vignettes of high-risk sexual scenarios. Sexual arousal and CAI intentions were assessed by self-report, and behavioral skills and risk exposure were derived from participants’ role-play behavior. Estimation of four path models showed that the hypothesized mechanisms were supported for the CAI intention outcome, but the findings for the skills and risk exposure outcome were mixed. Implications for development and enhancement of HIV prevention interventions were discussed.

Despite considerable advances in HIV prevention interventions, HIV and other sexually transmitted infections (STIs) remain significant public health problems (Jiang et al., 2020). Data suggest that, in the United States, the extent of the public health problem varies by population subgroup, but the most acute problem remains in men who have sex with men (MSM; Centers for Disease Control and Prevention (CDC), 2020). Accordingly, the National Institutes of Health (U.S.) Office of AIDS Research 2021–2025 Strategic Plan emphasizes primary and secondary prevention efforts that target subpopulations that remain heavily affected by HIV/AIDS (National Institutes of Health, 2019). Furthermore, unprotected sexual intercourse is the way that HIV most often is contracted in the US according to the CDC. Therefore, increasing the effectiveness of behavioral interventions, or combined behavioral and biomedical interventions, first requires an understanding of mechanisms that cause, in the case of MSM, condomless anal intercourse, or CAI.

Much research has been devoted to identifying variables that predict the likelihood and frequency of condomless sex and has shown that alcohol consumption is one of the most prominent among them. Research published over the last four decades across multiple countries has consistently shown a positive association between aggregate alcohol consumption (quantity, frequency) over a defined interval of time and aggregate frequency of condomless sex occurrences (Williams et al., 2016). Event level studies have similarly demonstrated an association between alcohol use and the occurrence of condomless sex, although they also have identified important contextual moderators of the association (Freeman, 2016; Wray et al., 2019). The causal influence of alcohol on sexual risk has been supported in experimental analogue studies, which have consistently shown that acute alcohol consumption increases the degree of intention to engage in condomless sex and decreases the use of strategies related to condom use (e.g., role-play performance of condom use negotiation skills; Berry & Johnson, 2018; Scott-Sheldon et al., 2016). Alcohol is an especially important variable to consider in studying CAI in MSM, because MSM have a higher prevalence of alcohol use than individuals demographically matched in the general population (Maisto & Simons, 2016; Medley et al., 2016).

Unfortunately, there is a major gap in the literature regarding empirical evidence for mechanisms underlying the alcohol-condomless sex relation (Berry & Johnson, 2018; Wray et al., 2021). This research gap is critical for at least two reasons. First, understanding mechanisms is essential to advancing knowledge and theory regarding alcohol’s effects on condomless sexual behavior. Second, knowledge of mechanisms of behavior is fundamental to developing more effective behavior change interventions (Kazdin, 2007). This latter point is critical in the context of mechanisms underlying alcohol’s relation to condomless sex and HIV prevention interventions that target MSM, because such interventions have largely neglected to incorporate the alcohol consumption-condomless sex relation (Vagenas et al., 2015; Williams et al., 2016; Woolf-King et al., 2020).

Self-control mechanisms in health behaviors

Self-control is a key component of interventions designed to modify behaviors that affect health, such as tobacco use, excessive eating, and engaging in condomless sex. Self-control has been characterized as a set of processes that underlie one’s ability to override thoughts, feelings, and action tendencies that compete with a specific goal (e.g., choosing an activity to pursue a health goal rather than pleasure; Inzlicht & Berkman, 2015). There is substantial empirical evidence that shows an inverse relation between self-control abilities and the occurrence of behaviors that tend to result in adverse health-related consequences (Protogerou et al., 2020). Dual-process models posit that self-control may consist of two dimensions, a fast-acting, reflexive, “automatic” dimension and a slower-acting, deliberative or effortful dimension (Heather, 2017; Lieberman, 2007; Osgood & Muraven, 2018; Wiers et al., 2007). The automatic, reflexive dimension emphasizes affective processes, while the effortful control dimension may largely consist of higher order “executive function” (EF) processes (Lieberman, 2007).

Automatic processes are based on activated associations between risk stimuli and their evaluation. For sexual stimuli, these evaluative associations may include feelings, cognitions and action tendencies that are represented in memory and indexed with implicit measures. Research has shown that implicit measures of approach - sexual stimuli associations are correlated with sexual activity (Hinzmann et al., 2020; Hofmann, Friese, & Strack, 2009). When an individual’s EF capacities or motivation to exert self-control is compromised, behavior becomes more strongly influenced by implicit evaluations that may not align with his/her explicit self-control goals (Hofmann, Friese, & Strack, 2009). The ability to exercise self-control varies widely, both between individuals (e.g., Maisto et al., 2021; Simons et al., 2010; Wills et al., 2006) and within individuals across time as a function of contextual factors (Baumeister et al., 2007; Lieberman, 2007). Two such contextual factors are alcohol consumption and sexual arousal.

Alcohol intoxication is associated with pronounced deficits in EF processes, which include working memory (WM), attention, and processing speed (Curtin & Fairchild, 2003; Pihl et al., 2003). Given this, we posit that alcohol may shift the relative influences of automatic vs. effortful decision-making processes on sexual behavior. When sober, individuals may tend to use deliberative processes to make decisions about sexual behavior based on personal standards and perceived costs and benefits. Such processing of goals may increase the tendency to engage in condom use. However, as degree of intoxication increases, the role of deliberative processes decreases and automatic processes increasingly drive behavior leading to a higher likelihood of condomless sex. In this regard, higher EF has been shown to be associated with decreased occurrence of condomless sex, but acute intoxication diminishes this effect (Ralevski et al., 2012). Accordingly, change in the influence of approach biases due to alcohol’s impairing effects on EF may mediate and adds to alcohol’s effects on implicit processes described earlier to influence the relation between alcohol consumption and condomless sex. However, research has yet to examine the effect of alcohol in moderating associations between implicit measures of sexual approach and sexual decision-making.

Sexual arousal has both motivational and emotional properties (Janssen, 2011; Toates, 2009) and thus is essential to consider in understanding the relations among alcohol, implicit motivation, executive function, and sexual decision-making. Experimental and observational research on alcohol and the occurrence of condomless sex in both individuals identifying as heterosexual or gay/bisexual (the latter, men only) has shown that sexual arousal can be either a mediator (George, 2019) or a moderator (Maisto et al., 2012; Maisto & Simons, 2016) of the relation between acute intoxication and having condomless sex. This work suggests that (1) increasing the influence to implicit biases toward sexual stimuli has one immediate effect of increasing sexual arousal, and (2) alcohol’s effect on implicit processes and their motivating effects on sexual behaviors are enhanced with a greater degree of sexual arousal (moderator) and can mediate the association between alcohol consumption and the occurrence of condomless sex, or act as a second, more proximal mediator of the occurrence of condomless sex as part of an alcohol→sexual risk approach biases→sexual arousal→condomless sex pathway.

How best to represent these multiple control processes is difficult in the context of an experiment, which does not include sufficient time to measure a dose of alcohol’s acute effects on all of the components that have been hypothesized to constitute EF nor multiple measures of automatic cognitive processes, and then in turn to study how any changes affect decisions regarding having condomless sex. Consequently, it is necessary to select those elements that have demonstrated clearest empirical support in model testing. Overall, the literature suggests that WM is a particularly important component of EF given evidence that WM capacity is related to successful self-control efforts (e.g., Bridgett et al., 2013; Hofmann et al., 2012; Kemps et al., 2020). In this regard, those with lower WM capacity exhibit stronger associations between automatic cognitive processes and drug and alcohol use, eating sweets, and sexual behavior (Grenard et al., 2008; Hofmann et al., 2008; Houben et al., 2011), and training to improve WM has been linked to reduced alcohol use and delay discounting (Bickel et al., 2011; Houben et al., 2011).

Summary and integration

Figure 1A depicts hypothesized inter-relationships among key variables of executive WM, automatic approach biases toward sexual stimuli, acute alcohol intoxication (“beverage condition” in the experimental context), sexual arousal, and sexual risk behaviors. The purpose of this study was to conduct an alcohol challenge experiment designed to test the hypotheses described immediately above and represented in Figure 1A in a sample of MSM. In this study, a community sample of MSM were randomly assigned to one of three beverage conditions, alcohol (target blood alcohol concentration [BAC] = .075%), placebo, or water control following completion of baseline assessments. After beverage consumption, participants completed measures of automatic approach biases (toward sexual stimuli) and working memory, respectively. Finally, participants watched two videos, each of which was followed by collection of role-play measures of condom use negotiation skills, participant-determined “exposure” to sexual risk, and intentions to have condomless sex with specified characters depicted in the role-plays.

Figure 1. Hypothesized Conceptual Model and Structural Models of the Relations among Alcohol Consumption, Approach Biases, Working Memory, and Sexual Arousal on Sexual Decision-Making Outcomes.

Note. Solid lines are hypothesized direct effects in the model. Dashed lines are moderating effects. Figure A = conceptual model. Figure B = Structural Model of CAI Intentions Analysis ( χ2(22, N = 257) = 16.88, p = .7699, CFI = 1.00). Figure C = Structural Model of Behavioral Skills Prompt 1 Analysis (χ2(22, N = 257)= 16.09, p = .8116, CFI = 1.00. RMSEA =0.00 90% CI[0.00,0.03], SRMR = 0.017). Figure D = Structural Model of Risk Exposure Analysis (χ2(22, N = 257)= 15.41, p = .8440, CFI = 1.00. RMSEA =0.00 90% CI[0.00,0.03], SRMR = 0.017). PrEP use, Site, and Age are included as covariates with paths to each endogenous variable, but omitted from the figures. Effects are standardized.

† p <.10, * p < .05, ** p < .01, *** p < .001

As represented in Figure 1A, intoxication was expected to result in a higher likelihood of sexual risk behaviors via increasing automatic approach biases toward sexual stimuli and simultaneously decreasing WM capacity. Stronger automatic approach biases were expected to increase responsiveness to sexual cues in the AAT and the resulting increase in sexual arousal (the latter was not manipulated in this experiment but was enhanced in all participants by the experimental stimuli) was expected to further increase effects of implicit cognition on behavior. In contrast, a decrease in WM due to intoxication was expected to reduce ability to access and utilize more distal health-promoting information (e.g., health risks, learned skills, etc.) and thus was expected to result in stronger associations between automatic approach biases to sexual stimuli and sexual decision-making (Hofmann et al., 2008). Sexual arousal was hypothesized to be a key factor linking alcohol intoxication and sexual decision-making (Maisto et al., 2012; Norris et al., 2009), and arousal increases the impact of automatic appetitive processes and decreases effortful control of behavior (Lieberman, 2007, Volkow, 2010). Hence, it is an essential factor to incorporate sexual arousal into the model. We expected intoxication and stronger automatic approach biases to be associated with higher reported sexual arousal when the sexual risk outcomes were measured. The increased arousal was predicted to partially mediate intoxication effects and to moderate the hypothesized effects of automatic and controlled (executive) processes.

Method

Participants

Participants were 257 men aged 21 to 50 (M =28.14, SD = 6.90) who were recruited from Syracuse, NY and Boston, MA using flyers, advertisements, and social networking sites (e.g., Facebook, Grindr, Tinder). Approximately 64.71% were White, 13.33% were Black, 7.45% were Asian, 0.78% were American Indian or Alaska Native, 0.39% were Hawaiian or Pacific Islander, 3.53% were mixed race, and 9.80% designated other. Approximately 18.43% were Hispanic/LatinX. Average yearly income was $32,583.60 (SD = $26500.10; median = $30,000.00).

Inclusion criteria were: 21–50 years old, a moderate or heavy drinker (according to the Quantity Frequency Variability [QFV]; Cahalan et al., 1969), sexually active with men (at least once a month in the 3 months prior to study enrollment), identify as gay or bisexual (three or higher on the Kinsey Scale; Kinsey et al., 1948), and denied being in a committed monogamous relationship. Exclusion criteria included: current medical or psychiatric problems, use of a vitamin/herb or medication for which alcohol use is contraindicated, current alcohol or other substance use disorder, alcohol treatment within the past three years, substance use disorder or mental health treatment in the past three months, or any lifetime history of treatment for bipolar disorder or schizophrenia. Four previous papers have reported findings from this study (Anderson et al., 2020; Luehring-Jones et al., 2019; Maisto et al., 2021; Rowland et al., 2021; Simons et al., 2019, 2021; Tahaney et al., 2020).

Measures

Screening.

Individuals interested in the study contacted the laboratory and completed initial telephone screening with a trained research assistant. Telephone screening assessed for age, recent sexual behavior, relationship status, alcohol use patterns, mental health and SUD treatment, current medications, and medical conditions. If eligible, participants were invited to schedule an initial laboratory session.

Demographics.

A demographics questionnaire was administered during Session 1 to collect information about participant characteristics. The items assessed age, race, ethnicity, annual income, and whether the participant was currently prescribed and taking PrEP.

Manipulation Checks.

Blood Alcohol Concentration.

Actual blood alcohol concentration (BAC) was measured using breath analysis (Alcosensor FST, Intoximeters, Inc.). BAC was assessed at six time points: upon arrival to the laboratory-prior to beverage administration, approximately 30-minutes after beverage administration (post—alcohol absorption period), after WM task, after AAT, after the set of interactive videos, and prior to leaving the laboratory to ensure a safe exit.

Perception of Intoxication and Alcohol Consumption.

Perception of intoxication was measured by asking participants “How intoxicated do you think you are?” on a 0 (Not at all intoxicated) – 10 (the most intoxicated I’ve ever felt) scale. Perception of the amount of alcohol that was consumed was measured by asking participants “How many shots of vodka do you think you consumed?” on a 0 – 10 scale. These questions were administered at three time points: approximately 30-minutes after beverage administration, after WM and approach avoidance tasks both were completed, and after the interactive videos.

Endogenous Predictor Variables.

Sexual arousal.

Prior to the interactive videos, current level of arousal was measured with a scale of 1 (not aroused at all) to 6 (extremely aroused). Post-video ratings of sexual arousal were obtained after each interactive video using the same scale. The two post-video arousal scores were averaged to form a single post-video arousal score.

Approach Avoidance Task.

The AAT was used to assess implicit biases toward sexual stimuli vs condom stimuli (Hofmann, Friese, & Gschwendner, 2009; Wiers et al., 2011). Stimuli were randomly presented and included 20 sexual images (men engaging in sexual behavior), 20 condom images, and 20 neutral images, for a total of 132 trials. Stimuli were preceded by a 500 ms white cross (+) fixation point in the center of the screen. Participants had 1700 ms to respond to the stimulus, with a 1000 ms intertrial interval. Each stimulus was presented in both a landscape and portrait format. Participants were instructed to pull the joystick towards them when an image in portrait format appeared while simultaneously imagining they were pulling that image to them. They were instructed to push the joystick away when an image in landscape format appears while simultaneously imagining they are pushing that image away from them. Pushing the joystick away caused the image to recede on the screen and pulling the joystick towards them caused the image to zoom, thus providing a sense of avoidance and approach, respectively. To maintain focus on image content and not simply the orientation, participants were instructed not to respond when the image was a tree, regardless of orientation. The instructions were followed by 8 practice trials (of grey boxes rather than images) during which they receive feedback on their responses. An index of sexual stimuli approach tendency was calculated following the procedures of Zvielli and colleagues (Maisto et al., 2021; Simons et al., 2010; Wills et al., 2006; Zvielli et al., 2015). Higher scores indicate faster reaction times when approach is paired with the sexual stimuli and avoid is paired with the condom stimuli, reflecting a bias to approach sexual stimuli relative to condom stimuli. The strength of approach versus avoidance toward the sexual stimuli as well as condom stimuli is relevant to the behavioral outcome, and our measure quantifies the relative response bias. Split-half reliability for the bias scores indicated good reliability both pre-beverage and post-beverage (Spearman-Brown Prophecy Reliability Estimate = 0.83, and 0.84, respectively).

Working Memory.

The Automated Operation Span (O-Span; Unsworth et al., 2005) was used to measure WM. This task requires participants to solve a series of math operations while trying to remember a set of unrelated letters. On this task, participants are given three practice trials: (1) letters only, (2) math problems only, and (3) letters and math problems together. Response time during the practice trial is used to set that participant’s time limit for the experimental trials (M response time+ 2.5 SD). The task consists of a series of blocks containing letter sets of 3–7 letters that participants must remember in order. Letters are presented one at a time and alternated with a math problem. During the task, participants are shown their math accuracy as a percentage in red font at the corner of the screen. The task correlates well with other measures of WM capacity and has good internal consistency and test-retest reliability (Unsworth et al., 2005). Brief versions comprised of blocks of this task have been shown to correlate highly with full versions of this measure (Foster et al., 2015). This permitted the presentation of target blocks both pre- and post-beverage administration that could be completed along with other measures. The measure used in the pre- and post-O-Span tasks were based on total letters recalled in order (Foster et al., 2015; Unsworth et al., 2005). This served as the index of WM for pre- and post-beverage assessment.

Dependent Variables.

Behavioral Skills and Risk Exposure.

Behavioral skills and risk exposure were measured with two interactive videos developed by Maisto et al. (2012). The videos were selected to elicit moderate sexual arousal, to limit the potential for overpowering the effect of the beverage manipulation. Both interactive videos produce indicators of the behavioral skills needed for safer sex in situations that would (1) be familiar to participants, (2) pose moderate difficulty to communicate feelings about condom use, and (3) elicit moderate sexual interest. The first interactive video depicted two men who had recently met and the second depicted two men who were friends. In both situations, the men had not previously had anal sex and were eventually faced with the decision of having CAI. The videos were enacted by professional actors according to a script and were filmed by professional videographers.

Each interactive video has a risk exposure component and a behavioral skills component. For the risk exposure component, each video began by setting a scene in which “Jim” (the protagonist) and “Dave” (the character with whom the participants were asked to identify) meet each other. The participant was asked to make a series of binary choices (yes/no) about engaging in various increasingly high-risk sexual activities with Jim, which constituted the basis of the interactive risk exposure measure. The choice points were (1) “Do you go with Jim to his apartment?”; (2) “Do you accept a drink from Jim?”; (3) “Do you go upstairs with Jim (video 1)?” or “Do you get in the hot tub with Jim (video 2)?”; (4) “Do you have anal sex with Jim?”; (5) “Do you have unprotected, receptive anal sex with Jim?” Each affirmative response was scored as one point and summed to create a total score. Video 1 and Video 2 total scores were averaged together to create a single variable for analyses. The risk exposure portion of the video terminated with the first “no” response and transitioned to the behavioral skills component of the video.

The behavioral skills portion of the video required participants to verbally negotiate sexual situations in an interactive role-play. Participants were asked to respond first to Jim’s comment that he desires to have CAI and that there is no cause for concern because he is safe (prompt 1). The video then paused for 60 seconds to allow the participant to respond. Subsequently, Jim delivered a second, more insistent comment that reminded the participant that CAI would not be risky, would be pleasurable, and that he could be trusted (prompt 2).

Participants’ qualitative responses to each of the prompts were audio-recorded and scored (on a 0–2 scale) on the following five dimensions (higher score = better communication skills): (1) use of an “I” statement of intention of safer sexual behavior or refusal of unsafe sexual behavior; (2) presence of a positive statement about Jim; (3) provision of a reason for safer sexual behavior; (4) suggestion of a specific safer alternative behavior; and (5) indications that the participant’s response was direct, serious, and clear. Each of the response dimensions was scored according to a previously established rating manual (Maisto et al., 2002, 2004). Two raters independently coded each of these five dimensions across the two prompts and the two videos. The two raters agreed on 95.15% of the codes, and the discrepancies on the remaining codes were resolved through discussion. An average behavioral skills score across the 5 dimensions in the two videos was created separately for each of the two prompts.

CAI intentions.

Self-reported likelihood of engaging in CAI after viewing each of the two interactive videos was measured by a 6-point rating scale anchored at “not likely at all” and “extremely likely” (George et al., 2009). A single score was created as the average CAI across the two prompts.

Procedure

Experimental Study

Session One.

This session began with a research assistant verifying participant’s age using government-issued photo identification and confirming the participant’s BAC was zero upon arrival to the laboratory. Two participants showed a BAC > 0 at this point and were rescheduled. After consenting to participate, individuals completed self-report questionnaires to confirm eligibility. Eligible participants completed computerized behavioral tasks and additional self-report questionnaires that were not included in the analyses reported here. Participants were compensated $50 for completing Session One, and those deemed ineligible were compensated $20.

Session Two.

Participants were instructed that they may be consuming alcohol at Session Two, and thus should not drive, or ride a bicycle to the laboratory. Participants were also instructed to not use any substance for 24 hours before the session, nor eat or drink anything other than water for four hours before the session. Compliance was assessed via self-assessment and using a breathalyzer to verify absence of alcohol consumption. Four participants who were non-compliant with pre-session instructions were rescheduled. Two research assistants facilitated Session Two: one (RA1) was blinded to the beverage condition and the other (RA2) administered beverages and was not present for the assessments.

Participants were randomly assigned to one of three beverage conditions: alcohol (dose designed to raise BAC to 0.075%), placebo, or water control. Prior to beverage consumption, participants completed baseline measures of O-Span and the AAT. Following consumption of their beverages, participants repeated the AAT and completed a second O-Span task. After post-beverage completion of these two measures, current sexual arousal was assessed and then participants watched two videos, each of which was designed to be moderately sexually arousing and to depict a scenario that required a decision to be made about engaging in CAI. Video sequence order was randomly balanced across beverage conditions. Each interactive video role-play was followed by one role-play and two self-report measures relevant to their decisions to engage in CAI (if they actually were in the situations depicted in the videos), and post-interactive video ratings, including sexual arousal. After the final study measure, participants who consumed alcohol were strongly discouraged from leaving the lab until two breathalyzer readings demonstrated a BAC ≤0.02%. Participants were compensated $90 for completing Session Two. One participant vomited during the course of the beverage administration portion of the experiment and was disenrolled from the study after a breathalyzer reading of BAC <0.02%.

Beverage administration.

Based on random assignment to beverage condition, RA2 told participants assigned to the alcohol and placebo conditions that they were being given an alcoholic beverage and participants assigned to the control condition were told that they were being given water. All participants received an equivalent volume of beverage based on calculations using their reported height and recorded weight (measured in the lab). All drinks were mixed (except control condition) and poured in front of the participants. Those assigned to the alcohol group received .70 g alcohol/kg body weight in the form of a chilled beverage of 80-proof vodka mixed with tonic water and lime juice in a 1:4 ratio. In the placebo group flat tonic water was substituted for vodka and poured from a vodka bottle; vodka was rubbed around the rim of the glass to enhance alcohol cues. Participants assigned to the control condition received water. Drinks were divided into three equal doses and participants were asked to consume all within 20 minutes and each at the same pace. They sat alone while consuming their drinks and RA2 intermittently (i.e., every 6–7 minutes) confirmed that the participant followed these instructions. General entertainment magazines that did not include mention of HIV or AIDS were available to the participants while they consumed their beverages. Upon finishing their drinks, participants began the 10-minute “initial (alcohol) absorption” period. RA2 obtained breath tests after the absorption period and each subsequent computerized measure. While maintaining blinding, RA1 assessed post-beverage perceptions (i.e., perceived intoxication and perceived consumption of alcohol) via a paper questionnaire.

Analysis Plan

We tested the hypothesized path models in Mplus 8 (Muthén & Muthén, 2019) with the maximum likelihood robust (MLR) estimator. Four models were tested, one for each outcome: CAI intentions, behavioral skills prompt 1, behavioral skills prompt 2, and risk exposure. The final models are depicted in Figure 1B–D and Figure 1S. Though not shown in the figures, each model included site, age, and PrEP use as covariates with paths to all endogenous variables. Experimental condition was an exogenous variable represented by two dummy coded variables reflecting the alcohol and placebo conditions. All other predictor variables were centered at their mean. Interaction terms were created by forming the cross-product of the respective centered variables. For interactions involving endogenous variables, interaction terms were allowed to freely covary with their constituent parts, or, for endogenous variables, with their disturbance terms (Preacher et al., 2007). The interactions and all other exogenous variables were freely covaried. Three hypothesized interaction effects were iteratively dropped for parsimony as they were not significant in any of the models: experimental condition x post-beverage AAT predicting pre-video and post-video sexual arousal; post-beverage WM x post-beverage AAT predicting the outcomes (e.g., CAI intentions).

In summary, the models included site, age, PrEP, alcohol condition, placebo condition, pre-beverage AAT, pre-beverage WM, alcohol condition x post-beverage AAT interaction, placebo condition x post-beverage AAT interaction, post-beverage AAT x post-video sexual arousal, and post-beverage WM x post-video sexual arousal as freely covarying exogenous variables. The site, age, and PrEP covariates and the experimental condition indicators had direct paths to all endogenous variables; post-beverage AAT, post-beverage WM, sexual arousal pre- and post-video, and the outcome variable (e.g., CAI intentions). In addition, post-beverage AAT had direct paths to sexual arousal pre- and post-video, and the outcome variable. Sexual arousal pre-video had a direct path to sexual arousal post-video, which, in turn, predicted the outcome variable. Associations between both post-beverage AAT and post-beverage WM and the outcome were moderated by post-video sexual arousal. The association between post-beverage AAT and the outcome was moderated by experimental condition. Hence, the models include direct paths from post-beverage AAT, post-beverage WM, post-video sexual arousal, experimental condition indicators, the experimental condition indicator x post-beverage AAT interactions, post-beverage AAT x post-video sexual arousal, post-beverage WM x post-video sexual arousal, and the site, age, and PrEP covariates to the outcome (e.g., CAI intentions).

Results

Descriptive Statistics

Summary statistics and correlations are presented in Table 1. Twenty-one percent of the sample reported using PrEP. PrEP use was positively correlated with CAI intentions and post-beverage WM and inversely correlated with sexual arousal. CAI intentions were moderately inversely correlated with behavioral skills and positively correlated with risk exposure. Sexual arousal (post-video) was moderately positively correlated with CAI intentions and risk exposure. Approach biases (post-beverage) exhibited a modest positive correlation with pre-video sexual arousal and an inverse association with behavioral skills at prompt 1. WM did not exhibit significant bivariate correlations with the dependent variables. Bivariate correlations indicate significant small-to-medium effects of alcohol vs. water control on CAI intentions and both behavioral skills outcomes (|r|’s .18 – .24).

Table 1.

Summary Statistics and Correlation Matrix

	M (SD)	1.	2.	3.	4.	5.	6.	7.	8.	9.	10.	11.	12.	13.	14.	15.
1. Age (n = 257)	28.14 (6.90)	1
2. Site (n = 257)	0.74 (0.44)	−.18**	1
3. PrEP (n = 257)	0.21 (0.41)	.02	.15*	1
4. Alc vs Pla (n = 172)	0.50 (0.50)	−.02	.00	−.06	1
5. Alc vs Wat (n = 171)	0.50 (0.50)	−.04	.00	−.02	.	1
6. Pla vs Wat (n = 171)	0.50 (0.50)	−.02	.00	.04	.	.	1
7. SA - 1 (n = 257)	2.47 (1.27)	−.08	−.17**	−.13*	−.01	.01	.02	1
8. SA - 2 (n = 257)	2.48 (1.23)	−.11	−.07	−.12*	.14	.10	−.05	.66***	1
9. CAI Intent (n = 255)	2.27 (1.34)	−.07	−.01	.20**	.08	.18*	.11	.11	.25***	1
10. Skill- P1 (n = 196)	0.53 (0.33)	−.02	.05	−.01	−.14	−.24**	−.11	.02	−.04	−.44***	1
11. Skill- P2 (n = 197)	0.60 (0.35)	−.02	.03	−.01	−.01	−.21*	−.21*	.09	.02	−.42***	.59***	1
12. Rexp (n = 256)	3.08 (1.27)	−.10	.20**	.09	.14	.07	−.07	.13*	.29***	.43***	−.17*	−.21**	1
13. AAT-1 (n = 256)	435.66 (166.60)	.08	−.25***	−.07	−.17*	.10	.27***	.10	.06	.08	−.10	−.12	.03	1
14. AAT- 2 (n = 256)	423.42 (173.66)	.00	−.12*	−.12	.01	.19*	.18*	.17**	.11	.05	−.20**	−.06	.12	.45***	1
15. WM- 1 (n = 252)	13.55 (6.96)	−.14*	.20**	.10	.15*	−.01	−.16*	−.07	−.03	−.08	.08	.02	.14	−.07	−.04	1
16. WM- 2 (n = 251)	29.78 (12.44)	−.22***	.17**	.15*	.05	−.14	−.19*	−.07	−.05	−.06	.10	.11	.10	−.13*	−.11	.53***

Note. Alc vs Pla = alcohol versus placebo contrast, Alc vs Wat = alcohol vs water contrast, Alc vs Pla = alcohol versus placebo contrast, SA1 = sex arousal pre video, SA2 = sex arousal post video, CAI intent = Condomless anal intercourse intentions, Skill- P1 = Behavioral Skills Prompt 1, Skill-P2 = Behavioral Skills Prompt 2, Rexp=risk exposure, AAT-1= Approach Avoidance Biases prebeverage, AAT-2= Approach Avoidance Biases postbeverage, WM-1=Working memory prebeverage, WM-2 = Working memory postbeverage. Site is coded 0 = Syracuse, 1 = Boston.

^*p < .05

^**p < .01

^***p < .001

Manipulation Checks

We tested a 3 (condition) x 3 (time) mixed model of perceived intoxication (χ² = 680.19, Ν = 257, p < .0001). There were significant main effects for time, condition, and the time x condition interaction (p’s < .0001). Pairwise contrasts indicated perceived intoxication was higher in the alcohol condition than placebo condition (contrast = 2.17, z = 9.46, p <.001) and higher in the placebo condition relative to the control condition (contrast = 2.67, z = 11.60, p <.001) at time 1. Effects remained significant, though diminished at time 2 and 3 (time 3 alcohol vs. placebo contrast = 1.73, z = 7.52, p <.001; placebo vs. control contrast = 1.45, z = 6.27, p <.001). Perceived intoxication exhibited a curvilinear decreasing trend in both the alcohol (time b =−0.69, p <= .001, time² b = −0.17, p = .002) and placebo conditions (time b = −0.51, p < .001, timê2 b = −0.19, p = .001). The results of the analysis of the perception of amount of alcohol consumed data paralleled those of the perception of intoxication data. Pairwise contrasts showed significant differences that were maintained across time (p’s < .001), though effects were slightly smaller. A repeated measures analysis of BAC in the alcohol condition, indicated that BAC exhibited a cubic growth trend over time (linear contrast = 0.003, z = 7.32, p < .001; quadratic −0.000, z = −0.44, p = .693; cubic contrast −0.002, z = −4.23, p < .001). At time 2, BAC M = 0.065, SD = 0.02; time 3 BAC M = 0.065, SD = 0.01; time 4 BAC M = 0.072, SD = 0.01; time 5 BAC M = 0.071, SD = 0.01.

Alcohol vs. placebo condition contrasts

Table 2 presents the results of alcohol vs. placebo condition contrasts for direct effects across the intermediate and sexual decision-making outcomes. Although mean differences between the alcohol and placebo groups were always in the expected rank order, they were statistically significant only for sexual arousal, as alcohol predicted higher sexual arousal (post video) relative to placebo. The latter result was due to a decrease in arousal in the placebo condition. As will be discussed later, control group means also were in the predicted rank order compared to alcohol and placebo groups but generally statistical differences were found only for the alcohol vs. control contrasts.

Table 2.

Alcohol vs Placebo Contrasts

CAI Intentions Model
Outcome	PC M(SD)	AC M(SD)	b	se	p
CAI Intention	2.28(1.33)	2.51(1.44)	0.13	0.10	.494
Sexual Arousal postvideo	2.32(1.86)	2.67(1.35)	0.36	0.14	.009
Sexual Arousal prevideo	2.49(1.29)	2.47(1.35)	−0.04	0.19	.824
Approach Avoidance postbeverage	442.83(175.18)	445.62(186.23)	0.26	0.24	.289
Working Memory postbeverage	27.79(13.27)	29.18(12.70)	−0.03	0.17	.842
Behavioral Skills Prompt 1 Model
Outcome			b	se	p
Behavioral Skills Prompt 1	0.53(0.34)	0.44(0.34)	−0.08	0.06	.159
Sexual Arousal postvideo	2.32(1.86)	2.67(1.35)	0.36	0.14	.009
Sexual Arousal prevideo	2.49(1.29)	2.47(1.35)	−0.04	0.19	.823
Approach Avoidance postbeverage	442.83(175.18)	445.62(186.23)	0.26	0.24	.288
Working Memory postbeverage	27.79(13.27)	29.18(12.70)	−0.04	0.17	.832
Behavioral Skills Prompt 2 Model
Outcome			b	se	p
Behavioral Skills Prompt 2	0.55(0.34)	0.54(0.38)	−0.02	0.07	.818
Sexual Arousal postvideo	2.32(1.86)	2.67(1.35)	0.36	0.14	.009
Sexual Arousal prevideo	2.49(1.29)	2.47(1.35)	−0.04	0.19	.823
Approach Avoidance postbeverage	442.83(175.18)	445.62(186.23)	0.26	0.24	.288
Working Memory postbeverage	27.79(13.27)	29.18(12.70)	−0.03	0.17	.836
Risk Exposure Model
Outcome			b	se	p
Risk Exposure	2.90(1.26)	3.26(1.31)	0.24	0.18	.176
Sexual Arousal postvideo	2.32(1.86)	2.67(1.35)	0.36	0.14	.009
Sexual Arousal prevideo	2.49(1.29)	2.47(1.35)	−0.04	0.19	.824
Approach Avoidance postbeverage	442.83(175.18)	445.62(186.23)	0.26	0.24	.290
Working Memory postbeverage	27.79(13.27)	29.18(12.70)	−0.04	0.17	.827

Note. PC = placebo condition, AC = alcohol condition, Approach Avoidance = Approach avoidance bias.

Structural Equation Models

Figures 1B–D present the complete results of the model tests for each of the sexual decision-making outcomes. Tables 3–5 summarize the results of tests of conditional indirect effects involving alcohol vs. control for each of the sexual decision-making outcomes except for behavioral skills prompt 2 (omitted due to the nonsignificant R² for the outcome). The remaining results of model testing are presented in Tables 1S to 6S and in Figure 1S in the online supplementary material.

Table 3.

Conditional Indirect and Total Effects of Alcohol Condition for Condomless Anal Intercourse Intention Analysis (unstandardized)

Pathway	Moderator / Level	b[95%CI]
Alc→WM2→CAI	SA2 M-1SD	0.08 [0.01, 0.25]
Alc→WM2→CAI	SA2 M	0.03 [−0.01, 0.12]
Alc→WM2→CAI	SA2 M+1SD	−0.02 [−0.13, 0.02]
Alc→AAT2→CAI	SA2 M-1SD	−0.06 [−0.20, 0.00]
Alc→AAT2→CAI	SA2 M	0.03 [−0.02, 0.14]
Alc→AAT2→CAI	SA2 M+1SD	0.12 [0.01, 0.31]
Alc→AAT2→SA1→SA2→CAI	AAT2 M-1SD	0.00 [−0.01, 0.00]
Alc→AAT2→SA1→SA2→CAI	AAT2 M	0.01 [0.00, 0.03]
Alc→AAT2→SA1→SA2→CAI	AAT2 M+1SD	0.02 [0.00, 0.05]
Alc→AAT2→SA1→SA2→CAI	WM2 M-1SD	0.00 [0.00, 0.02]
Alc→AAT2→SA1→SA2→CAI	WM2 M+1SD	0.01 [0.00, 0.05]
Total Effects, Alc→CAI	SA2 M-1SD, AAT2 M, WM2 M	0.46 [0.08, 0.86]
Total Effects, Alc→CAI	SA2 M, AAT2 M, WM2 M	0.50 [0.13, 0.88]
Total Effects, Alc→CAI	SA2 M+1SD, AAT2 M, WM2 M	0.54 [0.13, 0.93]
Total Effects, Alc→CAI	SA2 M, AAT2 M-1SD, WM2 M	0.26 [−0.25, 0.75]
Total Effects, Alc→CAI	SA2 M, AAT2 M+1SD, WM2 M	0.74 [0.14, 1.41]
Total Effects, Alc→CAI	SA2 M, AAT2 M, WM2 M-1SD	0.46 [0.10, 0.83]
Total Effects, Alc→CAI	SA2 M, AAT2 M, WM2 M+1SD	0.54 [0.14, 0.96]

Note. AAT2 = Approach Bias (postbeverage), Alc = Alcohol, CAI = Condomless Anal Intercourse, SA1 = Sexual arousal (prevideo) SA2=Sexual arousal (postvideo), WM2= Working memory (post beverage). Bold coefficients are significant based on bias-corrected bootstrapped 95% confidence intervals.

Table 5.

Conditional Indirect and Total Effects of Alcohol Condition for the Risk Exposure Analysis (unstandardized)

Pathway	Moderator / Level	B[95%CI]
Alc→WM2→REXP	SA2 M-1SD	0.00 [−0.11, 0.06]
Alc→WM2→REXP	SA2 M	−0.02 [−0.10, 0.01]
Alc→WM2→REXP	SA2 M+1SD	−0.04 [−0.13, 0.00]
Alc→AAT2→REXP	SA2 M-1SD	0.04 [−0.04, 0.20]
Alc→AAT2→REXP	SA2 M	0.07 [0.00, 0.21]
Alc→AAT2→REXP	SA2 M+1SD	0.10 [0.00, 0.26]
Alc→AAT2→SA1→SA2→REXP	AAT2 M-1SD	0.01 [0.00, 0.03]
Alc→AAT2→SA1→SA2→REXP	AAT2 M	0.01 [0.00, 0.03]
Alc→AAT2→SA1→SA2→REXP	AAT2 M+1SD	0.01 [0.00, 0.04]
Alc→AAT2→SA1→SA2→REXP	WM2 M-1SD	0.01 [0.00, 0.03]
Alc→AAT2→SA1→SA2→REXP	WM2 M+1SD	0.01 [0.00, 0.04]
Total Effects, Alc→REXP	SA2 M-1SD, AAT2 M, WM2 M	0.20 [−0.19, 0.60]
Total Effects, Alc→REXP	SA2 M, AAT2 M, WM2 M	0.21 [−0.16, 0.60]
Total Effects, Alc→REXP	SA2 M+1SD, AAT2 M, WM2 M	0.22 [−0.16, 0.61]
Total Effects, Alc→REXP	SA2 M, AAT2 M-1SD, WM2 M	−0.06 [−0.52, 0.40]
Total Effects, Alc→REXP	SA2 M, AAT2 M+1SD, WM2 M	0.48 [−0.13, 1.17]
Total Effects, Alc→REXP	SA2 M, AAT2 M, WM2 M-1SD	0.20 [−0.17, 0.58]
Total Effects, Alc→REXP	SA2 M, AAT2 M, WM2 M+1SD	0.22 [−0.16, 0.63]

Note. AAT2 = Approach Bias (postbeverage), Alc = Alcohol, REXP = Risk Exposure, SA1 = Sexual arousal (prevideo) SA2=Sexual arousal (postvideo), WM2= Working memory (post beverage). Bold coefficients are significant based on bias-corrected bootstrapped 95% confidence intervals.

CAI Intentions

The structural model is shown in Figure 1B. The model was a good fit to the data (χ2(22, N = 257) = 16.88, p = .7699, figure = 0.00 90% CI [0.00, 0.04], CFI = 1.00, SRMR = 0.018) and accounted for 22% of the variance in CAI intentions. In respect to the covariates, PrEP was associated with higher CAI intentions (β = 0.27, p <. 001) and higher WM (post-beverage; β = 0.10, p = .037). The Boston site was associated with less arousal (pre-video; β = −0.16, p = .037). Age was associated with lower WM (post beverage; β = −0.15, p = .010). Other covariate effects were not significant. These covariate effects, except for any related to the decision-making outcome, are the same for all subsequent models and are not repeated in later description of model testing results.

Conditional indirect effects.

Select conditional indirect effects involving alcohol vs. control are reported in Table 3. On average, alcohol condition (relative to water control) was indirectly associated with CAI intentions via approach biases and sexual arousal. Alcohol’s relation to CAI was stronger when approach biases were higher. There were other significant indirect effects as well that supported the hypothesized conditional mechanisms. For example, when arousal was elevated, there was a significant indirect alcohol effect via automatic control processes. In contrast, when arousal was low, alcohol’s effect was mediated through diminished executive function.

Behavioral Skills Prompt 1

The structural model is shown in Figure 1C. The model was a good fit to the data (χ²(22, N = 257) = 16.09, p = .8116, RMSEA = 0.00 90% CI [0.00, 0.03], CFI = 1.00, SRMR = 0.017) and accounted for 12% of the variance in behavioral skills at prompt 1. There were no significant relations between covariates and behavioral skills prompt 1.

Indirect effects.

Select conditional indirect effects involving the alcohol vs. control contrasts are reported in Table 4. Alcohol condition had significant inverse total indirect effects on behavioral skills prompt 1 when either sexual arousal or approach biases were elevated (i.e., M + 1 SD). This is primarily due to the significant specific indirect effect of alcohol condition → approach biases → behavioral skills when sexual arousal was at the mean + 1 SD. Other specific indirect effects were not significant. The total effect of alcohol on behavioral skills prompt 1 was significant except under conditions when approach biases were low (M –1 SD). Specific indirect effects and total indirect of the placebo condition vs. water control were generally not significant. The total effect of placebo vs water control was significant only when approach biases were elevated (M + 1 SD).

Table 4.

Conditional Indirect and Total Effects of Alcohol Condition for Behavioral skills Prompt 1 Analysis (unstandardized)

Pathway	Moderator / Level	b[95%CI]
Alc→WM2→BSK1	SA2 M-1SD	0.00 [−0.04, 0.01]
Alc→WM2→BSK1	SA2 M	−0.01 [−0.03, 0.00]
Alc→WM2→BSK1	SA2 M+1SD	−0.01 [−0.04, 0.00]
Alc→AAT2→BSK1	SA2 M-1SD	−0.01 [−0.06, 0.01]
Alc→AAT2→BSK1	SA2 M	−0.02 [−0.06, 0.00]
Alc→AAT2→BSK1	SA2 M+1SD	−0.03 [−0.08, 0.00]
Alc→AAT2→SA1→SA2→BSK1	AAT2 M-1SD	0.00 [0.00, 0.01]
Alc→AAT2→SA1→SA2→BSK1	AAT2 M	0.00 [0.00, 0.00]
Alc→AAT2→SA1→SA2→BSK1	AAT2 M+1SD	0.00 [0.00, 0.00]
Alc→AAT2→SA1→SA2→BSK1	WM2 M-1SD	0.00 [0.00, 0.00]
Alc→AAT2→SA1→SA2→BSK1	WM2 M+1SD	0.00 [0.00, 0.00]
Total Effects, Alc→BSK1	SA2 M-1SD, AAT2 M, WM2 M	−0.16 [−0.28, −0.04]
Total Effects, Alc→BSK1	SA2 M, AAT2 M, WM2 M	−0.17 [−0.28, −0.06]
Total Effects, Alc→BSK1	SA2 M+1SD, AAT2 M, WM2 M	−0.18 [−0.29, −0.07]
Total Effects, Alc→BSK1	SA2 M, AAT2 M-1SD, WM2 M	−0.03 [−0.18, 0.14]
Total Effects, Alc→BSK1	SA2 M, AAT2 M+1SD, WM2 M	−0.32 [−0.51, −0.14]
Total Effects, Alc→BSK1	SA2 M, AAT2 M, WM2 M-1SD	−0.17 [−0.29, −0.06]
Total Effects, Alc→BSK1	SA2 M, AAT2 M, WM2 M+1SD	−0.17 [−0.28, −0.05]

Note. AAT2 = Approach Bias (postbeverage), Alc = Alcohol, BSK1 = Behavioral Skills Prompt 1, SA1 = Sexual arousal (prevideo) SA2 = Sexual arousal (postvideo), WM2 = Working memory (post beverage). Bold coefficients are significant based on bias-corrected bootstrapped 95% confidence intervals.

Behavioral Skills Prompt 2

The structural model is shown in Figure 1S. The model was a good fit to the data (χ²(22, N = 257) = 17.41, p = .7405, RMSEA = 0.00 90% CI [0.00, 0.04], CFI = 1.00, SRMR = 0.018). However, the model R² was not significant (R² = .07, p = .058). Therefore, additional results of model testing are not presented here but as noted are available in the supplementary material.

Risk exposure

The structural model is shown in Figure 1D. The model was a good fit to the data (χ²(22, N = 257) = 15.41, p = .8440, RMSEA = 0.00 90% CI [0.00, 0.03], CFI = 1.00, SRMR = 0.017 and accounted for 19% of the variance in risk exposure. In respect to the effect of covariates on risk exposure, PrEP was associated with more risk exposure (β = 0.12, p = .037). The Boston site was associated with more risk exposure (β = 0.21, p = .002). Other covariate effects on risk exposure were not significant.

Indirect effects.

Select conditional indirect effects involving the alcohol vs. control contrasts are reported in Table 5. The placebo condition did not have significant indirect or total effects on risk exposure. On the other hand, the alcohol condition exhibited some significant positive indirect effects on risk exposure via approach biases and sexual arousal. However, the total indirect and total effects of alcohol condition (vs water control) were not significant.

Discussion

This experiment tested two mechanisms derived from a dual process model of self-control linking alcohol intoxication and analogue factors contributing to CAI among MSM. First, alcohol intoxication was hypothesized to increase CAI risk factors via increases in automatic, reflexive, control processes, operationalized as implicit approach biases toward MSM sexual stimuli versus condom stimuli. Second, alcohol intoxication was hypothesized to increase CAI risk factors via decreases in controlled, reflective, processes, operationalized as executive working memory. Consistent with dual process theories, arousal was hypothesized to increase the effect of automatic processes on behavior and decrease the effect of controlled processes on behavior (Lieberman, 2007; Metcalfe & Mischel, 1999; Volkow, 2010). Thus, mechanisms linking alcohol intoxication and CAI risk factors were expected to be conditional upon level of subjective sexual arousal, an important factor underlying sexual behavior and linked to alcohol effect’s on sexual behavior (Maisto & Simons, 2016; Maisto et al., 2012). Consistent with these predictions, the results showed the following: First, alcohol condition (versus water control) was indirectly associated with higher CAI intentions via increases in implicit approach biases when subjective sexual arousal was elevated (M +1 SD) but not when arousal was low (M −1 SD). This same pattern held for the behavioral skills1 and risk exposure outcomes. Second, alcohol condition was indirectly associated with higher CAI intentions via impairments in executive working memory when subjective sexual arousal was low (M −1 SD), but not when arousal was elevated (M +1 SD). The finding that impairment of WM mediated alcohol’s relation to CAI only when arousal was low suggests that, at higher levels of arousal, the influence of EF on sexual decision-making declines. No other indirect or conditional indirect effects involving alcohol and WM were significant for outcomes besides CAI.

It is important to view the findings of this study in the context of a study published recently by Wray et al. (2021), who reported the results of an alcohol challenge experiment that was similar in design and conceptualization to this study but whose pattern of CAI results did not fully align with those of this study. In the Wray et al. study, 121 heavy, “high-risk” drinking MSM aged 21–50 were randomly assigned to one of three (alcohol [BAC target = .08%]), placebo, or water control) beverage conditions, and after beverage consumption completed tasks of attention bias to MSM-related sexual stimuli and inhibitory control, respectively. After task performance, participants viewed the same two sexual risk videos that were used in this study and after each completed CAI ratings. No other sexual decision-making outcomes were reported. Consistent with this study’s findings, Wray et al. found that alcohol (vs. water; no alcohol vs. placebo contrasts were reported) was associated with increased CAI, and sexual arousal ratings over the course of the experiment were positively related to CAI. However, contrary to their hypotheses, Wray et al. found no direct effects of alcohol on the performance of either task, and no indirect or conditional indirect effects involving alcohol, attention bias, inhibitory control, and CAI. The discrepancies in results between Wray et al. and this study could be due to several differences between them, including use of two different behavioral tasks, neither of which showed an alcohol effect despite participants’ reaching the target BAC of .08% on average, including being on PrEP as an exclusion criterion, and a smaller sample size in Wray et al., which was reduced further for analysis to N = 83 due to task performance-related attrition.

The results of the present study involving alcohol consumption cannot be interpreted as a pharmacological effect of alcohol but rather as the combined influence of expectations that alcohol will be consumed and the consumption of a moderate (peak BAC averaging about .072%) dose of alcohol. This interpretation follows from direct alcohol vs. placebo condition contrasts that show no differences between them in any of the primary sexual decision-making or task performance outcomes. Such a pattern of findings has also been observed in some alcohol challenge/HIV-related sexual decision-making (Berry & Johnson, 2018) and in alcohol-cognitive task performance (Hoffman & Nixon, 2015) studies. It is plausible that these results were due to small effect sizes/insufficient statistical power, as the ordering of the outcome means for the beverage conditions was consistent with expectations. Unfortunately, it is difficult to identify other plausible explanations for the absence of alcohol-placebo condition differences in this study. For example, the dose of alcohol used was similar to that used in our own and past experiments that did show alcohol-placebo differences in sexual decision-making outcomes in both MSM and heterosexual samples. Furthermore, the sexual decision-making outcomes in this study are similar to those used in our and others’ prior research and have been shown to be sensitive to beverage condition manipulations. Similarly, participants’ BAC when performing the post-beverage consumption O-Span was sufficiently high to impair performance on a complex working memory task such as the O-Span. From a practical standpoint, however, failure to find pure pharmacological effects of alcohol is of less significance, because in real-life situations, drinking alcohol typically is inseparable from knowledge and associated expectancies of outcomes of its consumption in real-life sexual decision-making situations.

This study demonstrated how alcohol may alter and combine with implicit associations, working memory, and sexual arousal in relation to in-the-moment decision-making about having condomless sex. As George (2019) noted, this is the point at which HIV prevention interventions aim ultimately to be effective. Nevertheless, such interventions have predominantly been derived from a social-cognitive perspective and have not addressed the dynamics of in-the-moment sexual decision-making (Johnson, 2019). Although many HIV prevention interventions have met formal empirical criteria for being “evidence-based” (Protogerou et al., 2020), a considerable amount of variance in outcomes of HIV prevention intervention trials is not explained by intervention effects. The data of this and prior experimental and other event-level (e.g., EMA studies (Maisto et al., 2021; Simons et al., 2018) suggest that interventions with a focus on in-the-moment dynamics to complement social-cognitive interventions that emphasize skills building and determinants more distal from making decisions about sex could increase the efficacy of HIV prevention interventions. This is especially the case because drinking alcohol can alter mediators and moderators of relations between alcohol consumption and decision-making outcomes and results in a less-than-perfect correlation between intentions expressed when not intoxicated and intentions while intoxicated. Johnson (2019) formally recognized this conclusion by proposing a model of determinants of sexual intentions and behavior into “motivation” (pre-consumption) and consumption phases.

The importance of developing intervention models based on in-the-moment sexual decision-making has been discussed in the literature for decades (Gold et al., 1991; Maisto & Simons, 2016). Unfortunately, it seems that, for the most part, such interventions have not been established, particularly to account for the proximal effects of alcohol on sexual decision-making. George (2019) provided an excellent review and discussion of interventions that have attempted to address in-the-moment effects of alcohol. This work suggests that at least two types of intervention approaches may be relevant and may be combined. The first is to use strategies to modify the learned associations that underlie implicit biases toward sexual stimuli and away from condom use. The literature on cognitive bias modification (CBM) is relevant in this regard (Friese et al., 2011). Although there have been recent efforts to improve procedures (Wiers et al., 2020), reviews of CBM addiction (alcohol and tobacco use, respectively) studies have shown that they often are below clinical trial methodology current standards and vary considerably in efficacy findings that average in the small effect size range (Boffo et al., 2019). As Wiers et al. (2020) noted, this is a body of research still in its early stages that would benefit from methodological improvement. The second approach has been to focus on the in-the-moment event itself via use of education (about how alcohol can change critical determinants of sexual decision-making) and techniques such as imagery, vignette or role-play simulation, and discussion/review and processing of past incidents when a decision to have condomless sex was made while under the influence of alcohol (George, 2019). The testing of “microinterventions” (Strauman et al., 2013) is another possibility that has been applied in interventions for mood disorders (Strauman et al., 2013) and alcohol-related sexual aggression (Davis et al., 2020). One possible application to alcohol and decision-making about condomless sex is an intervention that focuses on the key mechanism of implicit approach associations regarding condomless sex. Such an intervention could be administered and then evaluated by requesting that participants complete analogue sexual decision-making tasks such as those used in this study under conditions in which participants are either intoxicated (or are instructed that they are consuming alcohol but are not) or sober, with feedback and discussion following each test. Such an approach takes into account that alcohol changes mechanisms underlying its association with having condomless sex and individualizes the intervention as well, or makes it more “patient-centered” (Pantelic et al., 2018).

There are a few limitations of this study that should be considered in interpreting its findings. Most important, it was an experimental analogue and for ethical reasons did not measure outcomes of actual in-the-moment sexual decision-making. Therefore, the generalizability to real sexual events is an empirical question. Another limitation is that the design of this study did not take into account the increased availability of PrEP to individuals such as MSM who are considered “high risk” for contracting HIV. In fact, 21 percent of our participants reported at baseline that they were prescribed PrEP (we have no data on how adherent the participants were to their prescriptions). In this study, “taking PrEP” was used as a covariate in testing the SEMs and showed a significant direct relation to one of the four main sexual decision-making outcomes, CAI intentions. Whether PrEP is important in estimating the relevance of this study’s findings because its availability has increased substantially in the last few years, and the concept of what is “sexual risk” when adhering to a PrEP regimen differs from when not taking PrEP (Maisto et al., 2021). Accordingly, the relevance of the sexual decision-making outcomes used in this study to current efforts at HIV prevention is not straightforward. In fact, the issue of the relevance of this study’s data to sexual decision-making is multifaceted. Essentially strict adherence to a PrEP regimen lowers the likelihood of HIV transmission to a partner virtually to zero (Mayer et al., 2019), thereby greatly decreasing the importance of factors that affect the decision to have CAI, though PrEP does not provide protection against contracting other STIs. Unfortunately, the large majority of individuals who are at high risk to contract HIV do not initiate PrEP uptake (Jenness et al., 2018) or maintain it if they do (Mayer et al., 2019) Furthermore, there are racial and ethnic subgroup differences (disparities) in MSM who initiate PrEP (Kanny et al., 2019). As a result, multiple modalities of PrEP delivery are being developed with the aim of increasing its uptake and maintenance, but their success in doing so still is under evaluation (Beymer et al., 2019). Therefore, this study’s findings for now may be viewed as highly relevant to a large majority of MSM, particularly in specific racial and ethnic subgroups. However, this conclusion is subject to change with patterns of PrEP uptake and use and as research findings emerge.

In summary, the results of this study show that implicit approach associations, executive functioning (WM) and subjective sexual arousal are critical mediators and moderators of the alcohol-CAI association in MSM. In particular, alcohol has direct effects on implicit approach biases to sexual stimuli, WM, and sexual arousal, and implicit associations and arousal combine to potentiate alcohol’s relation to decisions regarding unprotected sex in MSM, so that the latter is most likely to occur when sexual arousal high and implicit associations are strong. In this study WM played a less pervasive role, in that its relation to CAI was most evident when arousal was lower. This pattern of findings has clear implications for development and enhancement of HIV prevention interventions.

The results of this experiment also suggest several directions for future research. An overarching question is the generalizability of this study’s results to populations other than MSM. The theoretically-derived mechanisms whose actions were tested in this experiment have not been considered population-specific. Nevertheless, the replicability of this study’s findings in populations besides MSM is a topic for future research. Another priority is further clarification of the role of executive function in combination with acute alcohol effects and implicit associations in sexual decision-making. As our and Wray et al.’s (2021) data show, such investigations will require a conceptualization of executive function that has empirical support in the cognitive science literature and then a derivative operationalization that can be used in the context of either an alcohol challenge experiment or, more complex, in the natural environment in a longitudinal EMA study, for example. Related to this point, WM was related to CAI only when sexual arousal was lower. This pattern is consistent with dual-process models and suggests that, to the extent that alcohol acts to increase arousal, it somewhat paradoxically may decrease the impact of alcohol-induced impairments in executive function as “cold-cognition” may be a less powerful influence on decision making when aroused. In contrast, alcohol in this study both increased risk promoting implicit biases, and increased arousal, which further enhanced the impact of such automatic processes. These results are consistent with and enhance the findings of previous experimental analogue studies on the effects of alcohol, implicit biases, and sexual arousal and the mechanisms of their action on decisions about having unprotected sex. Future research should clarify executive function, and consider other factors at the event level, including partner characteristics, setting variables, and individual differences variables. Finally, ultimately this research is geared to increasing the evaluation of an increased number of HIV prevention interventions that incorporate the role of alcohol in combination with mechanisms underlying its association with in-the-moment decisions about sex, and implementing those that are empirically supported.