Efficacy of Behavioral Intervention, Text Messaging, and Extended Intervention to Address Alcohol Misuse in Sexual Minority Men with HIV: A Factorial Randomized Clinical Trial

Abstract

This clinical trial examined the individual and combined effects of three different approaches to reducing alcohol misuse among a sample of sexual minority men (SMM) with HIV. Specifically, we used a 2 × 2 × 2 randomized factorial design to compare: (a) behavioral intervention based in motivational interviewing (MI) vs brief intervention (BI), (b) interactive text messaging (ITM) for alcohol use vs. no ITM, and (c) extended intervention (EI) length of nine months vs. a one-month intervention duration. Participants (N = 188) were SMM with HIV and alcohol misuse recruited in Miami, FL, and Boston, MA. Participants were randomized to one of eight intervention combinations and assessed at 6- and 12-month follow-ups. Large reductions of over 50% in drinks per week and heavy drinking days were observed in all conditions at follow-up. Those who received ITM, compared to those who did not, reported significantly lower drinks consumed per week at 6 and 12 months (incidence rate ratios = 0.73 [95% CI = 0.57, 0.90] and 0.72 [95% CI = 0.56, 0.87], respectively), and increased odds of cessation of alcohol misuse at 12 months, odds ratio = 1.46, 95% CI = 1.03, 2.08. Results provided no evidence of better alcohol use outcomes for either MI or EI relative to their comparison conditions, and no specific combination of intervention components demonstrated a notable benefit. This study suggests a two-session BI can effectuate substantial reductions in alcohol use in SMM with HIV and that adding one month of ITM can yield further improvements.

Clinical Trials Number: NCT02709759

INTRODUCTION

People with HIV (PWH) are highly impacted by alcohol and its negative health consequences. PWH self-report high rates of alcohol misuse and heavy episodic drinking (4+/5+ drinks in a day for cisgender women/men respectively) [1,2]. Sexual minority men (SMM), including gay, bisexual, and other men who have sex with men, constitute the majority of PWH in the United States [3] and report relatively high rates of alcohol misuse compared to other PWH [1]. Alcohol misuse among PWH is associated with worse retention in HIV care [4–6], worse antiretroviral therapy (ART) adherence [7–10], decreased odds of HIV viral suppression [5,8,10], greater HIV disease severity [11], and increased risk of mortality [12,13], as well as a variety of long-term comorbidities including cancer, cardiovascular disease, liver disease, and neurocognitive dysfunction [14]. Furthermore, alcohol misuse in PWH is associated with more frequent condomless sex with HIV-serodifferent partners [15,16], which combined with its association with decreased HIV viral suppression, could contribute to greater HIV transmission risk when serodifferent partners are not using HIV pre-exposure prophylaxis. Effective alcohol interventions for PWH are essential for improving HIV care continuum outcomes and reducing comorbid conditions.

A variety of behavioral interventions have shown efficacy among PWH in reducing drinking, improving ART adherence, and reducing plasma HIV RNA (i.e., HIV viral load [VL]) [17]. Many of these interventions have utilized principles of Motivational Interviewing (MI), a non-confrontational, directive, client-centered counseling approach that has been shown to facilitate change across a range of health behaviors [18–20]. Despite some promising results for MI-based interventions for alcohol misuse in PWH, including robust effects seen for an intervention specifically tailored for SMM with HIV [21], it is unclear whether MI produces stronger effects than brief interventions (BI) [22–24], which typically require less technical skill and supervision to deliver. BIs are short (5–30 minutes), structured counseling intended to encourage patients to reduce their alcohol use by providing them feedback about their drinking levels and information about the health risks involved [25]. Although some studies have found suggestive evidence that MI-based interventions may reduce drinking more than BI in PWH [26,27], meta-analyses have not found a consistent difference across the literature [17,28,29].

In addition to questions about whether MI provides greater benefit than BI for PWH, the optimal dose of behavioral interventions is also unclear. Durations of behavioral interventions that report alcohol outcomes examined in PWH have ranged across trials from one to sixteen sessions [17], and no study has systematically examined whether providing more sessions of a given intervention produces stronger effects (e.g., standard vs. extended duration). Two studies that systemically addressed this question—not conducted specifically with PWH—did not find that including additional “booster” sessions beyond an initial brief intervention improved outcomes [30,31]. There have, however, been a handful of studies with PWH that suggest combining behavioral interventions for alcohol use with technology-based interventions—which can extend intervention outside the counseling sessions—may improve outcomes beyond counseling alone [27,32,33]; however, the role of technology interventions in improving outcomes is not well established.

To inform optimization of behavioral interventions for alcohol misuse in SMM with HIV, the present study utilized a factorial design consistent with a Multiphase Optimization Strategy (MOST) [34,35]. Our goal was to examine which intervention components showed the strongest effects while simultaneously examining potential optimal combinations of components. Specifically, the present study tested (a) whether MI tailored for SMM with HIV [21] outperformed BI; (b) whether interactive text messaging (ITM) provided additional benefits beyond behavioral counseling; (c) whether extending interventions (i.e., BI, MI, and/or ITM) for nine months provided additional benefits relative to interventions lasting only one month. Both MI and BI were delivered remotely via videoconferencing to maximize efficiency of delivery and potential population reach.

We hypothesized that MI would result in greater reductions in alcohol use than BI. Second, we hypothesized that participants who received ITM would have greater reductions in drinking than those who did not. Third, we hypothesized that extended intervention (EI) would result in greater reductions in drinking than interventions delivered for only one month. Secondary outcomes included condomless sex with non-steady partners, ART adherence, and HIV viral suppression.

METHODS

Study Design

Details about the methods in this trial have been previously published [36]. The trial utilized a 2 (BI vs. MI) × 2 (no EI vs. EI) × 2 (no ITM vs. ITM) randomized factorial design. Outcomes were assessed at 6 and 12 months after baseline, and study condition was masked to those completing follow-up assessments. The intervention sessions for both BI and MI occurred over a HIPAA-compliant videoconferencing platform immediately following baseline assessment and one-month post baseline via telephone. Those in EI spoke with their study interventionist again via telephone three months post baseline, via videoconference immediately after their 6-month follow-up assessment, and a final time via telephone nine months post baseline. The study was approved by the Institutional Review Boards at Fenway Health, Florida International University, and Brown University and registered with the Clinical Trials Registry (NCT02709759).

Sample Size and Power

We conducted a power analysis for the primary aims based on the desire to detect a significant effect of a given experimental factor on alcohol use at .80 power and a significance level of α = 0.05. In our previous trial, we found effect sizes ranging from d = 0.33 to 0.50[21] with an assessment-only control. Therefore, we decided to power the trial to detect a small to medium effect size (~d = 0.35) for any of the three experimental factors in the factorial design. Assuming that differences in the levels of baseline drinking account for about 20% of the variance in drinking at follow-ups, and would be included in the model, power analyses revealed that a sample size of 204 would be needed to detect an effect of d = 0.35 at any one follow-up with power of .80. Consistent with recommendations for optimization trials, we did not power the study specifically to test for interactions among conditions[34,37]. To allow for loss to follow-up of 10%, we sought to recruit and randomize 224 participants.

Participants

Eligible participants (1) were ≥ 18 years of age; (2) reported alcohol misuse, defined as an average of one or more heavy drinking (≥ 5 drinks) episodes per month or drinking more than 14 standard drinks per week during the previous 3 months; (3) were receiving care for HIV; (4) identified as a cisgender man; (5) had an HIV viral load test completed within the past 6 months; and (6) reported having sex (i.e., oral or anal) with a male partner in the past 12 months and/or identified as gay or bisexual. Participants were excluded if they: (1) reported injection drug use within the past 3 months; (2) displayed psychosis, active suicidal ideation, and/or mania; (3) received treatment for an HIV-related opportunistic infection within the past 3 months; (4) were currently receiving treatment for an alcohol or drug use disorder; or (5) could not commit to the study’s 12-month duration. We also excluded participants with a past 12-month history of severe alcohol withdrawal symptoms such as hallucinations, seizures, or delirium tremens.

Recruitment and Study Procedures

Participants were recruited from HIV primary care clinics and the local community around Florida International University in Miami, FL, and Fenway Health in Boston, MA, from August 2016 to March 2020. Participants were recruited through passive (e.g., fliers posted at HIV clinics and relevant community sites and online advertisements) and proactive (e.g., in-clinic screening and recruitment by research staff when patients arrived for routine HIV care appointments) recruitment strategies. Patients who expressed interest in participating completed a brief screening to assess inclusion criteria. Participants who met criteria were scheduled for a baseline interview to confirm eligibility where they completed written informed consent. At baseline and each follow-up visit, participants completed structured interviews, questionnaires, and neurocognitive tests. Participants were compensated $100 for the baseline interview, $50 for the 6-month interview, and $60 for the 12-month interview. For more details, see our previously published work[36].

Randomization

At the conclusion of the initial study visit, participants were randomized by computer to an assigned study condition. Using block sizes of 16 (two assignments of each of the eight potential combinations of treatment conditions), the software ensured that all conditions and combinations of conditions remained balanced for each study site[36].

Measures

The primary outcomes were alcohol consumption as operationalized by (1) average number of standard alcohol drinks consumed per week and (2) frequency of heavy drinking days in the past 30 days as assessed by the Timeline Followback[38] at 6- and 12-month follow-ups. We also formed a composite from these two variables to indicate whether participants no longer met criteria for alcohol misuse at those follow-ups, i.e., cessation of alcohol misuse. There were three secondary outcomes: (1) the number of days engaging in condomless sex with a non-steady partner in the past 30 days, (2) past 30-day self-reported ART adherence (i.e., percentage of days taking all prescribed ART doses), and (3) HIV viral suppression (i.e., <20 copies/ml). The former two outcomes were assessed via the Timeline Followback[38,39], while the latter was assessed via blood test (see below).

Clinical Assays.

At baseline, participants provided documentation of a plasma HIV VL test within the past 6 months and their most recent CD4 cell count; for those who were patients at Fenway Health, these data were extracted directly from the medical record. At 12 months, we again obtained VL and CD4 counts from medical records and if those were not available from within the past 6 months, we obtained the necessary blood sample to complete the assay.

Intervention details

The behavioral interventions were delivered by five master’s- or doctoral-level project staff with previous experience working with SMM, PWH, and/or people with substance use disorders in clinical/research settings. These interventionists received approximately 20 hours of training in MI, including reading and participating in role-playing exercises. Interventionists also attended weekly clinical supervision meetings with a doctoral-level trainer and supervisor to discuss current cases and receive feedback on sessions. All interventionists received detailed treatment manuals for both MI and BI. All intervention sessions were audio recorded. Recordings were reviewed in supervision sessions and were coded to evaluate MI integrity and fidelity using the Motivational Interviewing Skill Code (MISC) Version 2.5 [40]

Motivational Intervention (MI)

The MI condition closely mirrored the intervention deployed in our previous clinical trial [21], which drew heavily from techniques of Motivational Interviewing [41] to facilitate behavior change through support and guidance. Baseline sessions lasted approximately 60 minutes. Interventionists explored participants’ perceived benefits and drawbacks of drinking and how drinking related to their HIV treatment, while reflecting and amplifying statements supporting behavior change. They provided personalized feedback on participants’ drinking compared to a national sample of SMM [42], as well as feedback on the health impact of alcohol on ART adherence, liver functioning, VL/CD4 counts, cognitive functioning, sexual risk taking, and other substance use including tobacco. For participants receptive to change, interventionists worked with them on a change plan, including goal setting and problem-solving potential barriers to change. For participants who were not ready to change drinking, interventionists explored potential motivations for change in the future. Follow up session(s) lasting 20 to 30 minutes focused on participants’ drinking since the previous session. For participants who successfully reduced their drinking, the interventionist affirmed their success and discussed strategies for maintaining change. For participants who did not reduce drinking, interventionists elicited and explored reasons for participants’ lack of change and strategized future harm reduction goals.

Brief Intervention (BI)

The BI was a scripted intervention session lasting approximately 10 minutes. Participants were informed that their drinking was medically unsafe, that alcohol use had particularly harmful effects on health for PWH, and that continuing to drink at their current level could lead to health problems. They were asked what changes, if any, they wanted to make to their drinking. For those participants ready to change, the interventionist helped them decide on a specific goal and discussed steps they could take towards that goal. For participants not ready to change, the interventionist reiterated advice regarding lower risk drinking levels while affirming that the choice of whether to change was theirs. Follow up session(s) lasting approximately 10 minutes focused on participants’ drinking since the previous session. Depending on participants’ current drinking, interventionists discussed either maintaining current reductions or considering and planning for future change.

Interactive text messaging (ITM)

Participants randomized to ITM received daily texts during the first 30 days of their participation. Texts asked participants to report (1) the number of drinks they consumed the previous day; (2) what aspects of the previous day’s drinking they disliked (e.g., having a hangover today); and (3) whether they expected to drink on the current day. Automated customized messages were sent in response reflecting individual drinking goals. In addition, the system provided participants with weekly feedback on the number of drinks they had consumed with a tailored message based on their previous responses. Participants assigned to EI continued to receive ITMs for eight additional months but at a reduced frequency. Simple static messages were sent twice per week, and once per week an ITM interaction was initiated to assess past week drinking, to provide feedback, and to set goals around drinking.

Extended intervention (EI)

The EI condition entailed extending any intervention a participant received (i.e., BI or MI and ITM) through to nine months.

Intervention integrity

We used the Motivational Interviewing Skill Code Version 2.5 (MISC [40]) to evaluate MI integrity and the discriminability in counseling style between the MI and BI conditions. Five independent raters trained on the MISC coded 163 of the 188 initial counseling sessions. Double coding was conducted on approximately 10% of sessions (n = 18) to monitor reliability of coding and allow for review of coding discrepancies during coding supervision. Here we report on the MISC global scores, which assess the extent to which interventionists demonstrated key principles of MI counseling style using single items with 1 = low to 5 = high scoring, where 3 is considered adequate. We averaged ratings on Evocation, Collaboration, and Autonomy Support to form the MISC MI Spirit score for each session, which had good internal consistency, α = .80, and also examined the Empathy scale by itself as a second indicator of MI style. Results of t-tests indicated that interventionists demonstrated greater MI Spirit when delivering MI (M = 3.91, SD = 0.46), compared to when delivering BI (M = 3.27, SD = 0.46), t(161) = 8.90, p < .0001), although counseling in both conditions exceeded the threshold for adequate MI counseling style. Likewise, interventionists demonstrated greater Empathy when delivering MI vs. BI: M = 4.00, SD = 0.52 vs. M = 3.36, SD = 0.61, respectively, t(161) = 7.63, p < .0001. As would be expected, they provided similar levels of Direction, which is not a specific indicator of better MI counseling per se, in MI vs BI, (M = 3.94, SD = 0.59 vs. M = 4.05, SD = 0.67, respectively, t(161) = −1.12, p = .27), indicating that they kept an appropriate focus on alcohol use and related concerns in both conditions.

COVID-19

Participants were recruited prior to public health restrictions from the COVID-19 pandemic in March 2020. Those still within the intervention or follow-up window as of March 13, 2020, switched to telephone or videoconference counseling and assessment for all future interactions, resulting in increased missing data. Follow-up rates for the 6-month and 12-month follow-ups dropped from 78.8% and 77.8%, respectively, prior to the pandemic to 73.9% and 55.2% after the pandemic.

Analysis Plan

We summarized study enrollment, randomization, and follow-ups by CONSORT. Study participants were characterized by baseline data stratified by intervention condition: BI vs. MI, no ITM vs. ITM, and standard (SI) vs. EI, respectively.

For primary outcomes—(a) drinks per week and (b) number of heavy drinking days at 6- and 12-months follow-ups—we fit marginalized zero-inflated negative binomial (MZI-NB) models to test the main and interactive effects of each of the three factors in the study design:

$$log\left\{E\right(Y\left)\right\}={\beta }_0+{\beta }_{1}A+{\beta }_{2}B+{\beta }_{3}C+{\beta }_{12}AB+{\beta }_{23}BC+{\beta }_{13}AC+{\beta }_{123}ABC+\gamma X.$$ (Eq1)

with a similar logit model describing the excess zeros as in the traditional zero-inflation model. The terms A, B, and C in Eq1 are dummy variables for BI vs. MI, no ITM vs. ITM, and no EI vs. EI; they were coded orthogonally so they could be evaluated concurrently with the three 2-way interactions and one 3-way interaction among these factors. The model parameters β₁, β₂, and β₃ capture the main effects, and β₁₂, …, β₁₂₃ capture the two- and three-way interaction effects. The models were adjusted for baseline covariates X, including the baseline value of the dependent variable, recruitment site, and indicators for interventionists. Zero-inflated NB models were chosen over Poisson or negative binomial regression models due to evident lack of fit of the latter two models to the data given excessive zeros in the outcomes and overdispersion. We chose the marginalized model (MZI-NB) for its direct description of the population marginal mean counts; in contrast, standard zero-inflated models assume latent classes and hence do not directly estimate the overall exposure effects [43,44]. Following intent-to-treat principles, we conducted regression analyses on all participants who were randomized to an intervention condition regardless of the number of sessions of intervention they completed.

Secondary outcomes were: (1) the number of days engaging in condomless sex with a non-steady partner in the past 30 days at 6 and 12 months, (2) percentage of the past 30 days taking all prescribed ART doses at 6 and 12 months, and (3) HIV viral suppression (i.e., <20 copies/ml) at 12 months. Due to limited variation in condomless sex and ART adherence outcomes, we dichotomized these, respectively, as: (a) 0 day vs ≤ 1 day of condomless sex with non-steady partner and (b) ≤ 95% ART adherent [i.e., perfect or near-perfect adherence [45]] vs < 95% ART adherence. For secondary outcomes as well as the composite outcome of cessation of alcohol misuse, logistic regression models, similar to the model above but replacing the left-hand side of Eq1 by h{E(Y)} with h(.) being the logit link function, were used to test the main and interactive effects of the three study factors. All data analyses were conducted in R (version 4.1.3).

Missing data.

We conducted our primary analyses using complete data. We also re-ran analyses with imputed data as a sensitivity analysis. Imputation of missing data was carried out using the chained-equation method (R MICE package) assuming that data were missing at random (MAR). Specifically, we assumed that conditional on the participant’s baseline characteristics and the data at the current and previous visit(s), the study outcomes were missing randomly independent of the observed values. The MAR assumption is untestable but can be plausible for our study given that we collected a rich set of baseline characteristics about the study participants. We maintained the longitudinal structure of the data and built imputation models that imputed missing values at 6 months only using baseline data and observed 6-month data, and imputed the missing values at the 12 months using baseline and imputed 6-month data and observed 12-month data. Imputations were iterated until the distribution of imputed values stabilized. Five complete datasets were created and analyzed as described above. The results were then pooled using Rubin’s method [46,47].

RESULTS

Study Participation

Figure 1 shows the flow of participants through each milestone in the study. A total of 141 randomized participants came from the Florida International University site and 47 from the Fenway Health site. Among these 188 randomized participants, study retention rates were 78.2% and 70.7% respectively; retention was not significantly associated with any of the variables in Table 1 (see Supplemental Table 1 for analyses). Figure 1 shows the detailed study retention for each intervention condition.

Figure 1:

CONSORT diagram showing the flow of participants through each stage of the trial

Table 1.

Demographic and clinical characteristics of the study participants (N=188)

Variable		Counseling Condition		Text Messaging		Intervention Duration
	All	BI	MI	No ITM	ITM	SI	EI
N	188	94	94	96	92	94	94
Age (M, SD)	49.3 (3.3)	49.2 (3.3)	49.5 (3.2)	49.2 (3.1)	49.4 (3.4)	47.9 (3.3)	50.8 (3.2)
Education (n, %)
Less than high school	38 (20.2)	18 (19.1)	20 (21.3)	20 (20.8)	18 (19.6)	16 (17.0)	22 (23.4)
High school diploma or GED only	53 (28.2)	26 (27.7)	27 (28.7)	29 (30.2)	24 (26.1)	27 (28.7)	26 (27.7)
At least some college	97 (51.6)	50 (53.2)	47 (50.0)	47 (49.0)	50 (54.3)	51 (54.3)	46 (48.9)
Race (n, %)1
AI or AN	4 (2.1)	3 (3.2)	1 (1.1)	2 (2.1)	2 (2.2)	3 (3.2)	1 (1.1)
Asian	3 (1.6)	1 (1.1)	2 (2.1)	2 (2.1)	1 (1.1)	1 (1.1)	2 (2.1)
Black or African American	117 (62.2)	58 (61.7)	59 (62.8)	55 (57.3)	62 (67.4)	59 (62.8)	58 (61.7)
Pacific Islander	1 (0.5)	0 (0)	1 (1.1)	1 (1.0)	0 (0)	1 (1.1)	0 (0)
White	69 (36.7)	36 (38.3)	33 (35.1)	39 (40.6)	30 (32.6)	34 (36.2)	35 (37.2)
Ethnicity (n, %)
Hispanic or Latino	47 (25.0)	25 (26.6)	22 (23.4)	24 (25.0)	23 (25.0)	25 (26.6)	22 (23.4)
Not Hispanic or Latino	141 (75.0)	69 (73.4)	72 (76.6)	72 (75.0)	69 (75.0)	69 (73.4)	72 (76.6)
Sexual Identity (n, %)
Gay/Homosexual	101 (53.7)	55 (58.5)	46 (48.9)	51 (53.1)	50 (54.3)	49 (52.1)	52 (55.3)
Bisexual	64 (34.0)	28 (29.8)	36 (38.3)	33 (34.4)	31 (33.7)	36 (38.3)	28 (29.8)
Straight/Heterosexual	18 (9.6)	9 (9.6)	9 (9.6)	10 (10.4)	8 (8.7)	7 (7.4)	11 (11.7)
Queer, uncertain, or other	5 (2.7)	2 (2.1)	3 (3.2)	2 (2.1)	3 (3.3)	2 (2.1)	3 (3.2)
Employ Status (n, %)
Unemployed	99 (52.7)	45 (47.9)	54 (57.4)	53 (55.2)	46 (50.0)	57 (60.6)	42 (44.7)
Employed Part-time	21 (11.2)	13 (13.8)	8 (8.5)	6 (6.2)	15 (16.3)	9 (9.6)	12 (12.8)
Employed Full-time	29 (15.4)	12 (12.8)	17 (18.1)	14 (14.6)	15 (16.3)	11 (11.7)	18 (19.1)
Full-time/Part-time Student	6 (3.2)	3 (3.2)	3 (3.2)	4 (4.2)	2 (2.2)	4 (4.3)	2 (2.1)
Retired	33 (17.6)	21 (22.3)	12 (12.8)	19 (19.8)	14 (15.2)	13 (13.8)	20 (21.3)
Total Family/Household Income (n, %)
Less than $10,000	75 (39.9)	42 (44.7)	33 (35.1)	42 (43.8)	33 (35.9)	40 (42.6)	35 (37.2)
$10,000 to $50, 000	95 (50.5)	44 (46.8)	51 (54.3)	46 (47.9)	49 (53.3)	47 (50.0)	48 (51.1)
$50,000 or more	18 (9.6)	8 (8.5)	10 (10.6)	8 (8.3)	10 (10.9)	7 (7.4)	11 (11.7)
Times run out of money for basic necessities (n, %)
Never	62 (33.0)	28 (29.8)	34 (36.2)	34 (35.4)	28 (30.4)	26 (27.7)	36 (38.3)
Once	21 (11.2)	10 (10.6)	11 (11.7)	10 (10.4)	11 (12.0)	13 (13.8)	8 (8.5)
Twice	37 (19.7)	18 (19.1)	19 (20.2)	18 (18.8)	19 (20.7)	12 (12.8)	25 (26.6)
3 or more	66 (35.1)	37 (39.4)	29 (30.9)	34 (35.4)	32 (34.8)	42 (44.7)	24 (25.5)
Relationship (n, %)
Currently in relationship	86 (45.7)	42 (44.7)	44 (46.8)	46 (47.9)	40 (43.5)	39 (41.5)	47 (50.0)

Note:

¹Participants could select multiple racial identities, so the numbers add up to more than 188. Seven participants reported multiple racial identities; six with two identities and one with three identities endorsed. MI = motivational interviewing. BI = Brief Intervention. ITM = interactive text messaging. SI = standard intervention duration. EI = extended intervention duration. AI = American Indian. AN = Alaskan Native.

Table 1 presents the baseline characteristics of the overall study sample and subdivided by the three experimental factors. Results suggest study participants were generally comparable across study conditions.

Intervention Engagement

All participants received an initial session of either BI or MI. Among participants in the SI condition, 28 of 47 (59.6%) assigned to BI completed the second and final counseling session compared to 21 of 47 (44.7%) assigned to MI. In the EI condition, those assigned to BI completed an average of 1.7 (SD = 1.1) of the remaining four possible sessions with 41 of 47 (87.2%) completing at least one follow-up session and five (10.6%) completing all sessions. Those randomized to both EI and MI condition completed an average of 1.9 (SD = 1.4) sessions with 39 of 47 (83.0%) completing at least one session and eight (17.0%) completing all sessions.

Of those assigned to ITM (n = 92), 66.4% (n = 61) enrolled in the program by responding to an initial text, and these participants responded on an average of 20.8 days (SD = 10.2) during the 30-day daily texting period. Of the 46 participants assigned to extended ITM, 52.2% (n = 24) continued to respond after 30 days with an average of 20.0 days (SD = 11.1) of responding during the once weekly texting period that lasted 32 weeks in total.

Adverse Events

There were no serious adverse events that were deemed possibly or likely due to study participation.

Primary Outcomes

The top half of Table 2 provides a summary of primary study outcomes at each study visit by intervention condition and the cessation of alcohol misuse composite variable. Across conditions, large reductions in drinks per week and heavy drinking days were seen relative to baseline, which were similar at 6 and 12 months.

Table 2.

Summary of study outcomes at baseline & follow-up

Variable	Counseling Condition		Text Messaging		Intervention Duration
	BI (n=94)	MI (n=94)	No ITM (n=96)	ITM (n=92)	SI (n=94)	EI (n=94)
Primary Outcomes
# of standard alcohol drinks per week 1
Baseline	24.4 (21.5) 17.5 (8.2–34.5)	27.1 (23.6) 22.0 (10.0–35.8)	28.2 (24.7) 20 (9–41)	23.2 (19.9) 19.0 (9.8–28.0)	23.6 (18.5) 19.0 (9.0–34.8)	27.9 (25.9) 20.5 (10.0–35.0)
6-mo Follow up	8.8 (12.2) 4.5 (0.0–11.2) NA=26	12.1 (15.3) 7.5 (2.0–16.0) NA=16	14.0 (17.0) 8.0 (2.8–18.5) NA=24	7.2 (9.2) 5 (0–9) NA=18	10.0 (12.8) 6 (1–14) NA=13	11.2 (15.4) 6 (1–14) NA=29
12-mo Follow up	8.5 (11.7) 4 (1–13) NA=27	7.6 (9.3) 5.5 (0.2–9.8) NA=28	10.6 (12.5) 6 (2–15) NA=29	5.4 (7.3) 3 (0–7) NA=26	8.0 (10.9) 6 (1–9) NA=26	8.0 (10.3) 3 (0–13) NA=29
# of heavy drinking days in the past 30 days 2
Baseline	9.6 5 (2–16) 94, 100%	10.5 7 (3–16) 93, 98.9%	10.7 5.5 (2.0–18.5) 95, 99%	9.3 5 (2–13) 92, 100%	9.6 5.0 (2.0–15.8) 94, 100%	10.5 6.0 (2.2–16.8) 93, 98.9%
6-mo Follow up	2.9 0 (0–3) 32, 47.1% NA=26	3.9 1 (0–5) 46, 59.0% NA=16	4.5 1 (0–5) 41, 56.9% NA=24	2.3 0.5 0.0–2.8) 37, 50.0% NA=18	3.1 1 (0–4) 46, 56.8% NA=13	3.8 0 (0–4) 32, 49.2% NA=29
12-mo Follow up	2.9 1.0 (0.0–2.5) 34, 50.7% NA=27	2.6 0 (0–3) 31, 47.0% NA=28	3.5 1 (0–4) 38, 56.7% NA=29	2.0 0 (0–2) 27, 40.9% NA=26	2.6 0 (0–3) 33, 48.5% NA=26	2.9 0 (0–3) 32, 49.2% NA=29
# of participants drinking <15 drinks/wk and no heavy drinking days during the past 30 days 3
6-mo Follow up	36, 52.9% NA = 26	30, 38.5% NA = 16	30, 41.7% NA = 24	36, 48.6% NA = 18	34, 42.0% NA = 13	32, 49.2% NA = 29
12-mo Follow up	33, 49.3% NA = 27	33, 50.0% NA = 28	27, 40.3% NA = 29	39, 59.1% NA = 26	35, 51.5% NA = 26	31, 47.7% NA = 29
Secondary Outcomes
# of days >0 engaging in condomless sex with a non-steady partner 4
Baseline	21, 22.3% NA=0	22, 23.4% NA=0	24, 25% NA=0	19, 20.7% NA=0	18, 19.1% NA=0	25, 26.6% NA=0
6-mo Follow-up	10, 14.5% NA=25	9, 11.5% NA=16	11, 15.1% NA=23	8, 10.8% NA=18	8, 9.9% NA=13	11, 16.7% NA=28
12-mo Follow-up	6, 9.0% NA=27	11, 16.7% NA=28	9, 13.4% NA=29	8, 12.1% NA=26	8, 11.8% NA=26	9, 13.8% NA=29
Percentage of days taking all prescribed ART doses 5
Baseline	94.4 (93.3–100)	91.0 (93.3–100)	92.4 (93.3–100)	93.1 (93.3–100)	91.0 (90.8–100)	94.5 (93.3–100)
6-mo Follow-up	91.7 (93.3–100) NA=25	93.9 (93.3–100) NA=16	93.6 (93.3–100) NA=23	92.2 (93.3–100) NA=18	94.5 (93.3–100) NA=13	90.9 (93.3–100) NA=28
12-mo Follow-up	92.7 (93.3–100) NA=27	95.0 (96.7–100) NA=28	93.1 (93.3–100) NA=29	94.5 (93.3–100) NA=26	94.2 (93.3–100) NA=26	93.5 (93.3–100) NA=29
HIV viral suppression 6
Baseline	66, 71.0% NA=1	62, 66.7% NA=1	61, 64.9% NA=2	67, 72.8% NA=0	63, 68.5% NA=2	65, 69.1% NA=0
12-mo Follow-up	37, 64.9% NA=37	37, 59.7% NA=32	32, 55.2% NA=38	42, 68.9% NA=31	41, 66.1% NA=32	33, 57.9% NA=37

Note: MI = motivational interviewing. BI = Brief Intervention. ITM = interactive text messaging. SI = standard intervention duration. EI = extended intervention duration. NA = data not available.

¹Statistics are mean, standard deviation, median, IQR, the number of missing values.

²Statistics are mean; median; and interquartile range (IQR); N of “>0”; % of “>0”; Number of NA.

³Statistics are count; %; Number of NA.

⁴Statistics are #>0; %; Number of NA.

⁵Statistics are mean, IQR, # of NA.

⁶Statistics are #<20, % <20, Number of NA.

Table 3 presents results of regression analyses using MZI-NB for drinks per week. The main effect estimates indicate that ITM significantly reduced drinks per week at both 6 months and 12 months compared to no ITM, both before and after adjusting for baseline alcohol use, study site, and interventionist. No significant main effects were observed for MI vs. BI or for EI vs. SI. The adjusted two-way interaction between ITM and EI was significant, suggesting that the effect of ITM was weakened in the presence of EI. Analyses with imputed datasets yielded consistent results, although ITM × EI interaction was no longer significant nor was the effect of ITM at 12 months (see Supplemental Table 2).

Table 3.

MZI-NB model for average number of standard alcohol drinks consumed per week

Variable	6 month follow-up				12 month follow-up
	Unadjusted		Adjusted		Unadjusted		Adjusted
	IRR (95% CI)	P-value	IRR (95% CI)	P-value	IRR (95% CI)	P-value	IRR (95% CI)	P-value
Intercept	9.84 (7.6, 12.03)	---	11.59 (3.36, 19.83)	---	7.33 (5.70, 8.95)	---	4.60 (−0.12, 9.32)	---
MI	1.17 (0.91, 1.43)	0.172	1.16 (0.96, 1.37)	0.095	0.94 (0.73, 1.15)	0.591	0.88 (0.70, 1.06)	0.229
ITM	0.73 (0.57, 0.90)	0.007**	0.81 (0.66, 0.96)	0.024*	0.72 (0.56, 0.87)	0.003**	0.71 (0.56, 0.86)	0.002**
EI	1.01 (0.79, 1.23)	0.931	0.86 (0.70, 1.01)	0.092	1.07 (0.83, 1.30)	0.563	1.01 (0.80, 1.22)	0.918
MI × ITM	1.00 (0.78, 1.23)	0.971	1.06 (0.87, 1.25)	0.552	0.98 (0.76, 1.19)	0.820	1.03 (0.81, 1.25)	0.755
MI × EI	1.03 (0.80, 1.26)	0.773	1.09 (0.89, 1.29)	0.344	1.08 (0.84, 1.32)	0.503	1.10 (0.88, 1.32)	0.358
ITM × EI	1.21 (0.94, 1.47)	0.100	1.25 (1.02, 1.47)	0.016*	1.23 (0.96, 1.50)	0.066	1.26 (1.01, 1.52)	0.025*
MI × ITM × EI	0.85 (0.66, 1.04)	0.162	0.97 (0.80, 1.15)	0.769	0.79 (0.62, 0.97)	0.038	0.86 (0.68, 1.04)	0.164

Note: MI = motivational interviewing. ITM = interactive text messaging. EI = extended intervention. IRR= incidence rate ratio.

Table 4 shows results of the MZI-NB models for heavy drinking days. As with drinks per week, ITM was associated with greater reductions in heavy drinking days compared to no ITM, but these effects were nonsignificant, and the combination of ITM and EI was associated with more heavy drinking days at 6 months with a significant ITM × EI interaction. The ITM × EI interaction was no longer significant when imputed datasets were analyzed; see Supplemental Table 3. With imputed data, the effect of ITM in reducing heavy drinking days at 6 months became significant, and a significant effect of EI increasing heavy drinking days emerged.

Table 4.

MZI-NB model for the number of heavy drinking days (>= 5 standard drinks per day)/month

Variable	6 month follow-up				12 month follow-up
	unadjusted		adjusted		unadjusted	adjusted
	IRR (95% CI)	P-value	IRR (95% CI)	P-value	IRR (95% CI)	P-value	IRR (95% CI)	P-value
Intercept	3.12 (2.19, 4.04)	---	3.24 (−0.18, 6.67)	---	2.57 (1.75, 3.39)	---	0.82 (−0.37, 2.00)	---
MI	1.19 (0.83, 1.55)	0.240	1.24 (0.90, 1.57)	0.128	1.01 (0.69, 1.34)	0.941	1.04 (0.74, 1.34)	0.777
ITM	0.75 (0.52, 0.97)	0.052	0.81 (0.59, 1.02)	0.116	0.80 (0.54, 1.05)	0.167	0.78 (0.54, 1.01)	0.102
EI	1.12 (0.79, 1.45)	0.457	0.96 (0.69, 1.22)	0.746	1.09 (0.74, 1.44)	0.581	1.07 (0.76, 1.37)	0.664
MI × ITM	0.95 (0.67, 1.23)	0.710	0.93 (0.68, 1.19)	0.615	1.07 (0.73, 1.41)	0.687	1.12 (0.79, 1.45)	0.457
MI × EI	1.06 (0.74, 1.37)	0.722	1.10 (0.81, 1.40)	0.464	1.17 (0.80, 1.55)	0.326	1.14 (0.81, 1.46)	0.370
ITM × EI	1.45 (1.02, 1.88)	0.015*	1.46 (1.05, 1.86)	0.009**	1.23 (0.84, 1.63)	0.198	1.29 (0.92, 1.65)	0.081
MI × ITM × EI	0.79 (0.56, 1.02)	0.119	0.86 (0.63, 1.10)	0.281	0.75 (0.51, 0.98)	0.072	0.78 (0.55, 1.00)	0.087

Note: MI = motivational interviewing. ITM = interactive text messaging. EI = extended intervention. IRR = incidence rate ratio.

Table 5 presents logistic regression analyses for the composite outcome of alcohol misuse cessation, i.e., drinking fewer than 15 drinks per week and reporting no heavy drinking days in the past 30 days. At 12 months, ITM compared to no ITM resulted in significantly greater odds of this positive drinking outcome, both with and without adjustment for other variables. No other main effects or two-way or three-way interactions were significant. Results did not differ using imputed data (Supplemental Table 4).

Table 5.

Logistic regression for the composite outcome of drinking less than 15 drinks per week and no heavy drinking days in the past 30 days

Variable	6-month follow-up				12 month follow-up
	unadjusted		adjusted		unadjusted		adjusted
	OR (95% CI)	P-value	OR (95% CI)	P-value	OR (95% CI)	P-value	OR (95% CI)	P-value
Intercept	0.86 (0.61, 1.20)	---	0.75 (0.17, 3.45)	---	0.98 (0.69, 1.39)	---	1.43 (0.35, 6.18)	---
MI	0.74 (0.52, 1.03)	0.073	0.75 (0.52, 1.07)	0.117	1.02 (0.72, 1.45)	0.899	1.01 (0.70, 1.46)	0.941
ITM	1.12 (0.80, 1.57)	0.518	1.19 (0.83, 1.71)	0.353	1.46 (1.03, 2.08)	0.033*	1.47 (1.02, 2.13)	0.039 *
EI	1.19 (0.85, 1.67)	0.313	1.22 (0.85, 1.74)	0.283	0.95 (0.67, 1.35)	0.770	0.92 (0.64, 1.32)	0.659
MI × ITM	1.01 (0.72, 1.42)	0.945	0.95 (0.66, 1.36)	0.777	0.99 (0.70, 1.40)	0.951	0.97 (0.68, 1.40)	0.879
MI × EI	0.85 (0.61, 1.19)	0.350	0.91 (0.63, 1.30)	0.601	1.09 (0.77, 1.54)	0.642	1.07 (0.75, 1.54)	0.702
ITM × EI	0.95 (0.67, 1.33)	0.747	0.96 (0.67, 1.38)	0.838	0.85 (0.60, 1.20)	0.361	0.84 (0.58, 1.20)	0.340
MI × ITM × EI	1.02(0.73, 1.44)	0.891	1.03 (0.72, 1.47)	0.881	1.05 (0.74, 1.49)	0.781	1.01 (0.70, 1.45)	0.972

Note: MI = motivational interviewing. ITM = interactive text messaging. EI = extended intervention. OR = odds ratio

Secondary Outcomes

The bottom half of Table 2 presents secondary study outcomes. Condomless sex with non-steady partners was rare throughout the study period, with a continuous decrease from baseline to the 6-month and 12-month follow-ups. No significant main or interaction effects were observed in logistic regression analyses of this outcome (see Supplemental Tables 5 and 8).

At baseline, 6, and 12 months, 67.6%, 66.0%, and 72.2%, respectively, reported perfect or near perfect (≥ 95%) ART adherence. No significant main or interaction effects were observed in logistic regression analyses of this outcome (see Supplemental Tables 6 and 9).

At baseline, 68.8% of participants with available data had suppressed HIV viral load (<20 copies/mL); at 12 months, only 62.1% of participants with available data had suppressed HIV viral load. At 12 months, a significant interaction between ITM and EI indicated an unexpected negative effect of combining ITM and EI, OR = 0.59, 95% CI (0.39, 0.86); Supplemental Table 7. However, this effect was no longer significant when imputed datasets were analyzed (Supplemental Table 10).

DISCUSSION

Results of this randomized controlled trial did not yield any evidence that MI produced superior alcohol consumption or secondary HIV-specific outcomes, compared to BI. These results are consistent with a handful of prior studies that have also examined this question in PWH [22,23,33,48]. To date, superior outcomes for MI for alcohol use have only been seen when compared to a control condition involving assessment only and HIV care-as-usual [21,49]. A recent study suggested that MI may have benefits over BI in those lower in motivation to change their drinking or those who use other drugs [24]. Potential moderators of MI efficacy in the present trial will be addressed in future secondary analyses. However, at present, evidence suggests that conducting a high-quality brief intervention for alcohol misuse in PWH is likely to result in substantial reductions in drinking that are maintained over one year and are equal to those outcomes that would be achieved by a more intensive intervention based in MI. This finding has substantial implications for the allocations of resources in HIV care settings since training and staff resources needed for delivering BI are considerably less than those needed for delivering MI. Finally, both counseling conditions in the present study were delivered by telehealth, further establishing that both MI and BI can be delivered remotely with fidelity and good clinical outcomes.

The second major finding of the study was that ITM, compared to no ITM, resulted in about a 25% greater reduction in drinks per week at both 6 and 12 months and about a 50% greater odds of no longer meeting criteria for alcohol misuse at 12 months. These significant results are particularly remarkable in that they occurred in the context of substantial reductions in alcohol use in the sample as a whole. This was the first clinical trial to test an alcohol-focused ITM in PWH. Results are consistent with promising results from trials in PWH using interactive voice response and app-based interventions to extend intervention beyond a counseling session [27,33]. Such technology-delivered enhancements to brief interventions are likely to be low-cost and scalable, and greater adoption of these technologies in HIV care settings appears warranted. Notably, although only about two-thirds of participants enrolled fully in the ITM program following their first counseling session, this rate of engagement was higher than engagement in the second session of counseling for those in SI, where only about half of participants engaged in the second session. The immediacy of ITM may facilitate continued engagement around alcohol use compared to waiting for one month before a second counseling session, further arguing for the utility of ITM.

The final factor in the present clinical trial was extending the counseling and, if applicable, the ITM received for a period of 9 months compared to a 1-month intervention protocol. Overall, there was no evidence of benefit for EI, and in fact, there was evidence of potential iatrogenic effects of extending ITM, especially when combined with brief intervention. As with the effect of MI, future secondary analyses will examine moderators of EI to determine whether there are indications of benefit to extending EI for particular subgroups of PWH with alcohol misuse. However, in the absence of evidence for moderators, results support providing brief intervention with a 1-month follow-up contact to assess and reinforce any reductions in alcohol use. Such a protocol is resource-efficient and still allows for providing further assistance at the time of the follow-up contact, if desired, such as referral to other treatment. The only advantage seen for EI vs. SI was regarding greater completion of at least one follow-up session of counseling beyond the initial session. This likely reflects the fact that participants in EI could complete a counseling session immediately after completing their in-person 6-month follow-up visit. Accordingly, leveraging regular HIV care visits to re-engage people with HIV in alcohol counseling may be a promising strategy to facilitate greater receipt of follow-up sessions of both BI and MI. In the current study, participants were not seeking treatment, nor did they have to be interested in changing their drinking. It may be that engagement in follow-up sessions would be higher, and that EI would show some benefit, if it were not offered uniformly but based on patient interest.

Strengths and Limitations

The study had a number of substantial strengths including its diverse sample, rigorous factorial design, and strong internal validity. Specifically, the study sample was diverse with regard to race and ethnicity, and participants with socioeconomic disadvantage were well represented—many of whom had a current detectable viral load at study entry. Thus, the sample represented a population of SMM who are of particular importance for addressing Ending the HIV Epidemic goals [50].

By using a factorial design, the present trial tested three different intervention components efficiently. The counseling interventions were delivered through videoconferencing with high fidelity by highly trained and continuously supervised interventionists who delivered the interventions in both conditions. Further, the counseling conditions differed meaningfully in the extent to which they were consistent with the principles of Motivational Interviewing. In this regard, the study had high internal validity. However, the outcomes also represent delivery of interventions under optimal conditions, and results should be generalized with that understanding. In particular, by following a manualized protocol over two sessions with regular supervision and fidelity monitoring, the BI condition may have been more extensive and impactful compared to BI typically delivered in primary care (e.g., [51]).

The study also had some notable limitations. The number of participants randomized represented only 83.9% of our goal when recruitment was forced to end early at the start of the COVID-19 pandemic. In addition, our 70.7% follow-up rate at 12 months was less than our targeted rate of 90%. This differential was partially due to disruptions in staffing and the availability of in-person follow-ups during the pandemic. Therefore, statistical power to test our primary hypotheses was lower than what was planned, and our capacity to test the effects of combinations of factors was limited due to small cell sizes. Thus, conclusions about lack of differences between conditions should be made cautiously.

The study excluded those currently receiving alcohol or other substance use disorder treatment, and therefore, the sample may not have included those with the most severe use disorders. Future studies are needed to determine whether BI, MI, or ITM have any benefit for those already engaged in treatment who are continuing to misuse alcohol.

The study relied on self-reported alcohol use for our primary outcome, and it is possible that participants under-reported their alcohol use at follow-up given the demand characteristics of the study in which all participants received advice to reduce drinking regardless of intervention condition. Participants also completed multiple assessments of alcohol use over time, which can result in assessment reactivity whereby greater awareness of one’s drinking prompts reductions in alcohol use regardless of intervention content. These reactivity effects can make it harder to detect intervention effects.

Conclusions

Results of this factorial randomized controlled trial suggest that MI, in the present context, did not have advantages over BI, nor were there advantages of extending any interventions beyond one month. Results also indicate that delivering one month of ITM, a low-resource technology that allows participants to track their alcohol use and that reinforces behavior change, has substantial promise as an add-on to either MI or BI for SMM with HIV. Future studies should examine how ITM can be implemented in HIV care settings where BI can be readily delivered. In addition, future research should replicate findings using biochemical measures of alcohol use.

Data Availability

Data and code are available upon request to corresponding author.