This study was a randomized controlled trial of eSBI among YMSM and YTW seeking HIV testing [8]. Using an electronic web-based portal, all participants were screened for alcohol misuse and received immediate feedback regarding their level of use (eg, how their use compares to others, whether it exceeds “safe use” guidelines); those who screened for moderate to high alcohol use, including binge drinking, were then randomized to either electronic intervention or control modules. All participants, regardless of randomization status, were followed for 12 months with in-person visits conducted at 1, 3, 6, and 12-month intervals (see CONSORT diagram, Figure 1). All participants were recruited from HIV testing clinics that utilized a seek, test, treat, and retain (STTR) model of care [9].

The protocol was approved by the Ann & Robert H. Lurie Children’s Hospital Institutional Review Board (review #2015-703) with a waiver of parental permission for participation of minors (aged 16-17 years). All enrolled participants were consented prior to the start of any research activities using an electronic consent form. The consent form indicated that the purpose of the study was to learn more about whether screening and brief intervention for substance use is effective in reducing that use.

Youth were recruited from HIV testing centers at the 3 primary sites in Chicago: the Division of Adolescent Medicine at Lurie Children’s Hospital, Howard Brown Health, and the Village at the University of Chicago. Individuals were eligible if they were (1) aged 16 years to 25 years, (2) interested in testing for HIV infection, (3) HIV-negative or HIV status unknown (per self-report; verified at the point of HIV testing), (4) identified as a man who has sex with men or a transgender woman who has sex with men (ie, born male, identify as female or transgender, and at any point in the gender transition process), and (5) English-speaking. Individuals who were eligible and interested were enrolled by study staff.

Upon enrollment, all participants completed HIV testing (using standard of preventive care testing and counseling at each site); individuals who screened reactive on the rapid HIV test were immediately referred for confirmatory HIV testing and were withdrawn from the study. All participants were screened for alcohol misuse using the Alcohol Use Disorders Identification Test (AUDIT) via an electronic portal. The AUDIT is a widely used, 10-item, validated screening tool for the assessment of alcohol use [10]. It contains items that reflect consumption, dependency, and alcohol-related harms. The AUDIT total score ranges from 0 to 40, divided into 4 Zones (I-IV), reflecting increasing levels of risk and recommended intervention (ie, Zone 1: alcohol education; Zone II: simple advice; Zone III: simple advice plus brief counseling and monitoring; Zone IV: referral to specialist for diagnostic evaluation and treatment). In addition, one item on the AUDIT reflects binge drinking (“How often do you have six or more drinks on one occasion?”). For our study, those scoring in Zones II-IV or endorsing the binge drinking item were included.

After completion of the AUDIT, participants received immediate feedback regarding their level of use. Participants who screened moderate to high risk for alcohol misuse or endorsed binge drinking on the AUDIT were randomized via computerized assignment (1:1) to either brief intervention modules to reduce alcohol misuse, eSBI, or a time and attention-matched control (ie, promotion of good nutrition). Study staff were blinded to random assignments. The randomization sequence was generated by NSK, uploaded to the Radiant platform (Radiant Interactive Group), and assigned automatically in the order of enrollment.

Those randomized to the intervention were asked to complete lessons or exercises on 11 topical areas focused on alcohol use (eg, importance of change, downsides of drinking, barriers to change, strategies for cutting back, change motivators) in an MI format, each with a single web page. The intervention is not a full MI intervention, as it was relatively simple in its presentation, but it did have components of the Elicit-Provide-Elicit approach. The intervention had been widely used in employee assistance programs. The intervention content was adapted for late adolescents and young adults in terms of voice and language (eg, open, nonjudgmental, inclusive), contexts (eg, school as well as work contexts), and reading level. The intervention was presented as a series of slides that contained an interactive element in which participants could click on checkboxes, drop-down menus, and short text fields as part of the brief MI exercises. An example slide is shown in Figure 2. By design, the intervention lacked any specific cultural tailoring for YMSM or YTW other than the age-appropriate language. This was intended to facilitate rapid scaling and broad adoption across the general population if the intervention proved effective. The electronic intervention and control modules were delivered using a customized eSBI platform developed by Radiant Interactive Group (Laguna Niguel, CA). Radiant uses industry-standard data encryption and security to manage protected health information recorded by their system.

Those randomized to the control condition completed a brief time and attention-control intervention, also in an MI format and of equal length (ie, same number of intervention screens), which encouraged good nutrition but was not expected to impact alcohol use.

The standard HIV preventive care at each site included rapid point-of-care HIV testing and counseling and linkage navigation to PrEP or HIV care, depending on the test result.

Participants completed a baseline study visit that included a standardized assessment via computer-assisted self-interviewing (CASI). Follow-up CASI assessments were conducted at 1, 3, 6, and 12-month follow-up visits either in person or remotely via web-based assessment. Detailed contact information was collected to facilitate study retention.

The analysis used an intent-to-treat approach in which all randomized participants were included for analysis. Baseline characteristics of the intervention groups were inspected visually for imbalances; differences between the 2 intervention groups were assessed using chi-square or Fisher exact tests for categorical variables and t tests for numerical variables. Changes in drinking levels over time were examined descriptively using means with SDs and medians with IQRs. The targeted sample size of 450 was estimated to yield 80% power to detect effects for our primary (alcohol use) outcome under a range of retention levels (75%-85%).

To evaluate the efficacy of the intervention, generalized estimating equations (GEEs) with robust standard errors were conducted. Each outcome was regressed on intervention group, time, a group-by-time interaction, and the baseline value of the outcome, where the statistical significance of the interaction term indicates whether the groups differ in response to the intervention over time. The baseline outcomes were included in the model due to baseline imbalances in the outcomes between the groups. Unadjusted models are presented in Table S1 in Multimedia Appendix 1. Omnibus tests (Type III Score tests) were used to assess the overall effects of the intervention, time, and whether the differences between the groups varied across time points. Contrasts were performed to assess the effect of the intervention compared with the time and attention-matched control at each follow-up visit compared with first follow-up for exploratory purposes even when omnibus tests were not significant. Interaction contrasts represent the relative risk (RR) for binary outcomes or incidence rate ratio (IRR) for count outcomes in the intervention arm at that time point compared with the RR or IRR for the group comparison at first follow-up (ie, a ratio of ratios). All count outcomes were modeled as negative binomially distributed variables with log link functions to compute rate ratios; the binary outcome (having had more than 5 drinks in any single drinking session in the prior month) was modeled as a binomially distributed variable with log link function to estimate relative risk. An exchangeable working correlation was specified to account for repeated measurements on participants over time. All statistical analyses were conducted using SAS software version 9.4 (SAS Institute). Graphics were generated using STATA 17.

A total of 644 individuals were screened for the study: 549 were eligible, and 95 were not eligible. Study enrollment began in December 2016, and all follow-up visits were completed by October 2020. Among those eligible, 85 (85/549, 15.5%) were lost to follow-up prior to enrollment, resulting in 464 enrolled (see CONSORT diagram, Figure 1). Of those enrolled, 14 were withdrawn or considered incomplete, resulting in a final sample of 450 (450/549, 82% of eligible).

All enrolled participants were screened with the AUDIT; 120 screened low risk and were not randomized. Therefore, a total of 329 participants were included in the analysis described herein. Table 1 presents descriptive information on the 329 participants randomized to the intervention or attention control conditions. The average age of the participants was 22.8 years. Most participants identified as non-White (221/329, 67.2%). The majority identified as cisgender male (294/329, 89.4%) and gay (205/329, 62.3%). Most of the participants were employed either full-time or part-time (222/329, 67.7%) and had at least some college education or attended trade school (224/329, 68.3%). There were 70 (70/329, 21.3%) with a history of incarceration. A little more than one-half of participants had a primary care provider (174/329, 52.9%). There were no significant differences between the intervention groups in these characteristics at baseline. For each measure of drinking, there were reductions in the mean amounts of drinking over time in both groups (Table 2). The average duration of the eSBI intervention was just under 10 minutes and of the control condition was just over 10 minutes.

The results of the adjusted GEE models for primary outcomes are presented in Table 3. The omnibus tests (Type III Score tests) showed that the overall effect of the intervention on substance use was not statistically significant (ie, no significant differences between groups across all time points) for any of the primary outcomes. The overall effect of time was statistically significant only for binge drinking in a peak drinking week at 12 months compared with 1 month (ie, there was an increase in drinking outcomes over time across both groups; IRR=1.32, 95% CI 1.05-1.66), and the intervention group-by-time interaction was not statistically significant (ie, differences between groups did not vary over time) for any of the primary outcomes. Figure 3 displays the results of the GEE models with the predicted values and lower and upper 95% CIs.

The results of the adjusted GEE models for secondary outcomes are presented in Table 4. The intervention group showed a reduction in the number of condomless IAS acts while under the influence of alcohol or drugs, whereas the control group showed an increase in the number of acts from 3 months to 12 months (IRR=0.15, 95% CI 0.05-0.44) when examining the contrasts. There were no significant differences between the groups in the other sexual behavior outcomes or number of PrEP care visits between groups. The results of the GEE models are visualized in Figure 4.

In terms of acceptability and satisfaction, 89.7% (148/167) of intervention participants rated the quality of the intervention as “good” to “excellent,” and 94.6% (156/167) indicated they were “mostly satisfied” to “very satisfied.” A total of 89.7% (148/167) indicated that they would “probably” or “definitely” refer a friend to receive the intervention. The most common suggestions for improvement to the intervention were to include interaction with an educator or counselor (n=8); make the intervention less prescriptive for low levels of drinking (n=10); and include graphics, video, or music (n=3).

As an exploratory aim, we conducted tests of moderation to determine whether the efficacy of the intervention was modified by the presence of depression or anxiety. The observed effects for anxiety were unstable and varied in direction, magnitude, and statistical significance depending on the analytic approach. When we controlled for baseline imbalances in the outcomes, the intervention appeared to be more efficacious among participants with fewer symptoms of anxiety; however, when we did not control for baseline imbalances, the intervention appeared more efficacious among participants with more anxiety symptoms. The contradictory findings may be due to associations between anxiety and the baseline outcomes, but the result is difficult to interpret and should be considered with caution. No moderation effects were observed for depression in any of the analyses.

In this study, we found no effect of the eSBI intervention on alcohol misuse, but there was evidence of an effect on sexual risk. This study was born from a desire to create a low-cost, scalable, and brief single time-point intervention to address substance misuse in community settings. Prior literature has shown that computerized interventions can successfully reduce substance misuse using an MI approach. For example, Ondersma and colleagues [20] demonstrated that a single 20-minute eSBI intervention, with a more comprehensive MI approach, successfully improved abstinence from drugs at 3 months and 6 months among postpartum drug users. In contrast, our eSBI approach failed to show any major effect of eSBI on drinking behavior in moderate to high-risk drinkers. Although there was an effect at 12 months with significant increase in the number of binge drinking episodes in a peak week for both the eSBI group and our control condition, these differences were not seen at 3 months or 6 months, and the time-by-group effect was not significant. Our intervention’s lack of effect, in contrast to the results by Ondersma et al [20], may be the result of differing populations (HIV at-risk YMSM and YTW vs postpartum women) and the need for cultural tailoring of the intervention for the former. In addition, the trial by Ondersma et al [20] used animation to deliver the intervention, which may have improved engagement with the content.

Since this study was conceived, there has been a number of trials of eSBI that have found no effect [21,22] or some impact in particular populations including college age drinkers [23], nonpregnant women of childbearing age [24], individuals on antiretroviral therapy [25], and graduate students using cannabis [26]. The literature on eSBI can best be described as mixed, and there is evidence that the more efficacious eSBI interventions have strong cultural tailoring, which our intervention did not have due to a desire to scale this intervention broadly. In addition, our MI approach was very brief and included relatively short MI components. It was by design simple, and we were seeking to test a system that might be broadly implemented. However, this brevity may have come with a price, as the intervention may not have had the robustness of a full MI intervention.

Despite the lack of findings for substance misuse, there was evidence of reduction in sexual risk in the intervention group in comparison to the control group for condomless IAS under the influence at 12 months. These analyses should be interpreted with caution, as there were multiple secondary analyses. Nevertheless, this finding is intriguing and should be the subject of additional study.

At the time that this study was conceived in 2014, trials had mixed findings on the impact of brief interventions on AOD. Since that time, observers have raised criticism of the value of screening and brief interventions [27-29]. The Substance Abuse and Mental Health Services Administration (SAMHSA)–promoted SBIRT model has generally shown impact on individuals with high levels of substance use and in primary care settings [27]. It is possible that our intervention was too brief and lacking in boosters or follow-up. Further trials may be merited with these elements in a dynamic trial design that might allow simultaneous comparisons.

Another promising line of inquiry that has shown impact is the use of relational agents (ie, using avatars, conversational bots, or other interfaces) that simulate face-to-face dynamics and allow for a more holistic experience during screening and a brief intervention [30-32]. Relative to this type of relational agent, our intervention was static; designed with simple, scalable elements; and presented in a slide format. The simplistic design was by intention, as we hypothesized that many clinics would not have the infrastructure, computing capabilities, and high bandwidth access that are often required for more computationally complex frameworks. We suspect that our eSBI system lacked enough appealing components to produce a robust attentional response; however, as we did not measure attention to the eSBI system, this is purely speculative.

Although, overall, the intervention received high ratings for acceptability and satisfaction, some participants suggested the addition of more dynamic and engaging components (eg, more graphics, video, and music), commented that the fully automated version was impersonal and would have preferred more interaction with an educator or counselor, or felt the intervention was too prescriptive for perceived low levels of drinking. These are important considerations to inform future iterations in this population.

Finally, since we conceptualized this study in 2014, eSBI approaches have been tested for feasibility among people living with HIV in sub-Saharan Africa, where alcohol misuse is associated with both HIV transmission risk as well as poor HIV treatment outcomes but is often underreported [25,33]. In Uganda and Namibia, an electronic screening approach was designed specifically to increase reporting of alcohol misuse and, in Namibia, was then coupled with brief intervention. In both studies, evidence of feasibility was apparent, although with concerns about confidentiality (ie, providers being aware of drinking levels), which hampered acceptability [25,33]. These types of initiatives will provide important additional tests of effectiveness of eSBI to reduce alcohol misuse in HIV care environments and, given our findings, may also hold promise for reducing the risk for ongoing HIV transmission (ie, via sexual risk reduction).

This study found a lack of longer-term impact of a single time point intervention, eSBI, on alcohol misuse among YMSM and YTW presenting to community-based HIV testing clinics. The intervention showed an effect in reducing one of the sexual risk outcomes (number of condomless IAS acts while under the influence). Given that this intervention was embedded within larger systems of care (ie, placed within HIV testing clinics without the need for staffing), there would appear to be some value in considering further trials in this space. Further adaptation of the eSBI approach might result in greater impact. This eSBI intervention was embedded within an STTR model of care that has been shown to be highly effective in addressing HIV risk and engaging individuals in care. The study showed excellent recruitment and retention of the sample over a 1-year follow-up period. Our eSBI system can be considered a branch of the broader SBIRT model of care. This model has come under some criticism in recent years due to a lack of compelling evidence to show its impact as a primary prevention and intervention approach. Our eSBI system may have failed to achieve the intended goals of alcohol misuse reduction because of the brevity of the intervention, the lack of boosters or follow-up, and the relatively simple design of the intervention. Regardless of the reason, this study falls into alignment with other negative trials of SBIRT. Future research may need to consider relational agents, digital interventions hybridized with human counselors or therapists, and interventions with boosters of some type. In addition, interventions that can address contextual factors (interpersonal, social, and environmental) may help boost the effect of digital interventions like eSBI. Given the instability of results for moderation of the intervention effect by symptoms of anxiety, future studies may need larger samples or be limited to the subpopulation of individuals with higher levels of anxiety to explore these effects more fully.