Daily Associations Between Alcohol Consumption and Antiretroviral Therapy (ART) Adherence Among HIV-Positive Men Who Have Sex With Men

Abstract

Alcohol consumption is one of the strongest predictors of suboptimal adherence to antiretroviral therapy (ART), however, there is little research that has investigated both within- and between-person associations of alcohol consumption and ART adherence at the event-level. In this secondary data-analysis, (N = 22) HIV-positive MSM prospectively reported daily alcohol consumption and ART adherence for 42-days. Multilevel models demonstrated 1) days in which participants reported consuming any alcohol was associated with 2.48 increased odds of ART non-adherence, compared to days in which participants reported no alcohol consumption, and 2) there was a non-significant trend indicating days in which participants reported consuming greater than their own average levels of alcohol was associated with increased odds of ART non-adherence. Findings highlight the importance of combining intervention efforts that address alcohol consumption and suboptimal ART adherence, and indicate a need for future research to investigate the mechanisms by which alcohol influences ART adherence.

Introduction

There are approximately 1.1 million people with HIV (PWH) in the United States (U.S., 1). Although men who have sex with men (MSM) represent only 2% of the U.S. population, they account for 70% of all HIV diagnoses (1). National goals for HIV treatment include linkage to care within one month of diagnosis for 85% of PWH, retention in care for 90%, and viral suppression for 80% (1). Treatment goals are based on evidence that PWH are 96% less likely to transmit HIV when virally suppressed on antiretroviral therapy (ART; 2–5). Current estimates suggest that only 49% of PWH are retained in care, and of those, only 53% are virally suppressed (1). Among MSM with HIV, these estimates are similar – 48% are retained in care, and 52% are virally suppressed (1). Interventions to improve ART adherence and viral suppression are thus essential for curbing HIV incidence.

The optimal adherence level for achieving viral suppression has conventionally been considered 95% (6), however, newer ART regimens are more robust and 80%-90% adherence is sufficient for sustained viral suppression (6,7). One of the strongest predictors of suboptimal ART adherence is alcohol and other substance use (8,9). Indeed, PWH who are at-risk drinkers, compared to those who abstain, experience a significantly increased risk for suboptimal adherence to ART (9,10) and lack of viral suppression (11). Compared to the general population, MSM engage in higher rates of heavy episodic drinking (≥ 5 drinks on a single day at least once per week; 12), and among PWH, MSM engage in the highest rates of heavy drinking compared to other groups (13,14). Therefore, MSM who consume alcohol are at high risk for suboptimal ART adherence, and merit additional research attention.

Based on a systematic review of the literature, Woolf-King et al. (9) proposed an integrated conceptual model of the relationship between alcohol consumption and ART adherence grounded in constructs from the Information-Motivation-Behavioral Skills model and the acute pharmacological effects of alcohol intoxication. This integrated model categorizes ART non-adherence as either intentional (i.e., purposely skipping doses of ART based on alcohol-ART interactive toxicity beliefs) or unintentional (i.e., prospective memory impairment resulting from acute alcohol intoxication). Moreover, risk of non-adherence is exacerbated as the quantity of alcohol consumed during a drinking event increases, enhancing cognitive impairment and increasing the likelihood of a prospective memory failure (9). As such, HIV-positive MSM are more likely to miss a dose of ART on days that they consume alcohol, either as a result of intentionally skipping a dose to avoid mixing alcohol and ART, or as a result of alcohol-impaired cognitive functioning that interferes with medication-taking behaviors. While this model is grounded in well-established health behavior theory, empirical research with the methodological rigor necessary to examine these event-level drinking relationships is sparse.

Woolf-King et al. (9) concluded that the majority of published studies in the literature on alcohol consumption and ART adherence have focused on the global or situational association between the two variables. Global association studies can be defined as those that correlate aggregated alcohol use with ART adherence over some duration of time (e.g., past 30-day alcohol consumption and past 3-day adherence). There are several limitations of global association studies. First, global association studies do not provide information on the temporal relationship between alcohol consumption and suboptimal adherence (i.e., drinking and non-adherence co-occurring on the same day). Second, global association studies rely on retrospective self-reported estimates of perceived ART adherence over extended periods of time (e.g., previous three months), which are vulnerable to social desirability and recall biases and may result in overestimates of true levels of adherence (50). Last, correlational findings can be confounded by other factors that are related to both alcohol consumption and suboptimal adherence (e.g., trait impulsivity, stressful life environments). Compared to global association studies, situational association studies are more specific in assessing typical adherence to ART while under the influence of alcohol (e.g., “I skip taking my HIV medications if I will be drinking alcohol;” 15). However, these studies also rely on retrospective reporting and are unable to assess alcohol consumption and medication-taking during specific drinking events. To maximize the effectiveness of ART adherence interventions that also address alcohol use, the temporal sequencing of the association between the two behaviors must be established (16,17). As such, event-level studies are designed to examine the temporal co-occurrence of alcohol use and ART adherence on a specific occasion. For example, a daily diary study can prospectively assess adherence on drinking and non-drinking days, over a period of weeks. There is thus a need for event-level studies that can investigate the co-occurrence of alcohol use and ART adherence.

Our review of the literature identified only six event-level studies that have examined the co-occurrence of alcohol consumption and ART adherence at the event-level (i.e., day). These studies demonstrated that on average, ART non-adherence is 2-4 times more likely on days in which any amount of alcohol is consumed compared to days in which it is not consumed (9). Notably, only two event-level studies reported results in samples comprised of MSM (18,19). Further, only one used a prospective event-level assessment methodology (i.e., ecological momentary assessment; 19), and the other is the only study in which both a temporal and dose-response relationship between alcohol and ART non-adherence has been examined (18).

Parsons et al. (18) used a retrospective event-level assessment procedure with a mixed-gender sample (N = 272; 78% male) who reported hazardous alcohol consumption (≥ 8 on the Alcohol Use Disorder Identification Test [AUDIT]) and enrolled in an alcohol reduction and ART adherence randomized clinical trial. Data from the 14-days prior to the baseline assessment were used to test the same-day association between alcohol use and ART adherence. Results indicated that non-adherence was approximately nine times (aOR = 8.78, 95% CI [7.16, 10.77]) more likely on days during which participants consumed alcohol (i.e., temporal), and each additional drink was associated with a statistically significant small increase (aOR = 1.2, 95% CI [1.18, 1.22]) in odds of non-adherence (i.e., dose-response). Analyses were not conducted separately for MSM participants, however. Additionally, the study relied on retrospective self-reports over a 14-day period, which may be an insufficient timeframe to representatively capture within-person fluctuations of the two behaviors.

Turner et al. (19) conducted a secondary data-analysis examining the association between same-day alcohol use and ART non-adherence. Data were collected prospectively over a 30-day period as part of a digital HIV care navigation intervention designed to improve HIV care outcomes in 113 young (i.e., ages 18-34) MSM and transgender women (20). Results demonstrated participants were approximately two times (aOR = 1.89, 95% CI [1.14, 3.15]) more likely to miss a dose of ART on days during which they consumed alcohol. Notably, there was no test of whether the quantity of alcohol consumed (i.e., dose-response) was associated with ART non-adherence.

Both Turner et al. (19) and Parsons et al. (18) collected data from participants who were enrolled in studies designed to improve HIV-related health outcomes, and the intervention tested in Parsons et al. (18), was specifically designed to target alcohol use and ART adherence. As such, it is possible that their findings may not be representative of alcohol consumption and medication-taking behavior under typical circumstances, such as without discussions regarding alcohol consumption modification, or developing strategies to improve adherence. It is also possible that enrolling in an HIV care navigation intervention enhanced participants’ ART adherence throughout the course of the data-collection period. Nevertheless, both studies demonstrated an inverse alcohol and ART-adherence relationship. Yet, it remains plausible that the magnitude of the association could be even greater in a non-intervention context. In an effort to extend these findings, the sample in this study differed from the sample recruited by Turner et al. (19), in that we analyzed data with MSM who were not enrolled in an intervention designed to facilitate linkage to, engagement in, and retention in HIV care.

Using prior research (18,19) as a guide, this study performed secondary data analyses to examine the within- and between-person event-level associations between same-day alcohol consumption and ART adherence among a sample of HIV-positive MSM. Based on definitions used by Woolf-King et al. (9), we operationalized the temporal association to be the co-occurrence of alcohol consumption (regardless of quantity) and ART non-adherence on the same day. In contrast, dose-response associations were defined as the relationship between the number of alcoholic beverages consumed and the likelihood of ART non-adherence. Of note, this level of specificity is a step removed from examining the medication-taking event, in which the temporal sequencing between alcohol consumption preceding ART non-adherence could be established. Nonetheless, examining same-day associations is an improvement over global and situational association studies. We hypothesized that: 1) days in which participants reported consuming any alcohol would be associated with increased odds of ART non-adherence, compared to days in which participants reported no alcohol consumption, and 2) days in which participants reported consuming greater than their own average level of alcohol consumption would be associated with increased odds of ART non-adherence, compared to days in which participants reported consuming average/below-average levels of their own alcohol use.

Method

Study Procedures

This is a secondary analysis of data collected from a daily diary study designed to examine event-level substance use and sexual behavior among HIV-positive MSM (N = 101). Participants were recruited via in-person methods (e.g., flyers, infectious disease clinics), participant referrals, and web-based social media methods (e.g., Grindr), from San Francisco, California and Syracuse, New York. Additional papers reporting findings from the parent study have been published elsewhere (47–49). Briefly, participants attended two study visits in the laboratory (i.e., baseline and six-week follow-up), and were assigned to either a six-week daily diary condition (Interactive Voice Response [IVR] condition) or a no-contact control condition (Control condition). During the baseline session, interviewer-administered self-report questionnaires were completed, biospecimen samples were collected (i.e., dried blood spots for phosphatidylethanol analysis), and for those assigned to the IVR condition, the interviewer provided instructions for completing a daily diary phone session and a practice session was completed. Participants assigned to both conditions returned for a follow-up session six-weeks post-baseline and the procedures and questionnaire battery of this session were identical to the baseline, excluding the daily diary training.

Inclusion criteria were: (i) 18-65 years of age, (ii) self-identified as a cis/transgender man, (iii) sexually active (i.e., reported having >1 instance of receptive or insertive anal sex with a man in the last 6 weeks), (iv) inconsistent condom use (i.e., using condoms more than 0% of the time but less than 100% of the time during sexual intercourse in the previous 6 weeks), (v) consumption of more than one drink containing alcohol in a typical month, (vi) owned a cellular device, (vii) did not demonstrate evidence of active psychosis (i.e., judged by trained research staff), and (viii) ability to understand spoken English. All procedures were approved by the Institutional Review Boards at Syracuse University and University of California, San Francisco. Daily ART adherence was assessed only at the Syracuse site, therefore, participants assigned to the IVR condition recruited at this site (N = 22) were included in this secondary data-analysis.

Participants assigned to the IVR condition received a daily phone call to their cell phone via an automated IVR system. The IVR system was programmed to call the participant every 15 minutes for two hours, at a time of the participant’s choosing, or until the participant completed the assessment. For privacy purposes, each participant entered a unique personal identification number (PIN) prior to beginning the assessment. An automated audio recording administered a set of questions each day, and participants replied using their cell phone keypad. Participants also had the ability to place an outgoing call to the IVR system to initiate/complete the assessment. Daily diary assessments that were not completed during the 2-hour data-collection window were considered “missed days.” Research staff attempted to contact participants who missed two consecutive daily diary assessments and offer technical support as necessary. Participants had the ability to modify the time of their 2-hour data-collection window at any point of the six-week data-collection period. Additionally, participants received a weekly text message that contained the total amount of compensation they had accrued to date. Participants received $2 for each daily diary that was completed and a $6 bonus was awarded for completing all assessments throughout each week. A total of $20 could be earned each week for a maximum total of $120 for completing all 42 daily diary assessments. Participants were also compensated with $30 and $40 for completing the baseline and follow-up sessions, respectively.

Measures

The measures described next were used in this secondary data-analysis. An extensive battery of self-report questionnaires with additional instruments was administered at both the baseline and follow-up sessions.

Demographics.

Demographic variables included in this study were age, race, personal income, years since HIV diagnosis, and highest level of education completed.

Alcohol Use Disorder Identification Test.

The Alcohol Use Disorder Identification Test (AUDIT; 21) was used to characterize the extent to which participants engaged in alcohol consumption that placed them at-risk for developing AUD. A cutoff score of eight is indicative of high-risk for AUD (22). In this sample, scores ranged from 2 – 31 (M = 11.59, SD = 6.21), and the scale demonstrated good reliability (α = 0.75).

30-Day Aggregated Antiretroviral Therapy Adherence.

Retrospective aggregated estimates of ART adherence over the previous 30-days were self-reported at the baseline session, prior to beginning the daily diary assessments. Participants were asked to estimate the number of days they missed at least one dose of their HIV medications in the last 30 days, which was then converted to an adherence percentage. ART adherence percentages ranged from 0%-100% (M = 92.86, SD = 23.79).

Interactive Voice Response Daily Diary.

Participants completed daily diaries using the automated IVR phone system. In addition to reporting daily alcohol use and ART adherence, participants answered questions related to physical pain, sexual activity (e.g., condomless anal intercourse), other substance use during sexual activity, sexual partner characteristics, and sexual partner substance use. The average duration to complete a daily diary assessment session was 2.9 minutes (SD = 1.14).

Daily Alcohol Use.

The IVR system asked participants to report the number of standard alcoholic beverages they consumed during the previous 24-hours with the following question: “Approximately how many standard drinks did you consume in the last 24 hours?” During the baseline session, the interviewer reviewed the definition of a standard drink and provided instructions for calculating standard drink estimates. Responses ranged from 0 – 20 standard drinks. Responses to this question were used to categorize a response day as a dichotomous drinking day (i.e., 1 = drinking day, 0 = non-drinking day) and to record the number of standard drinks consumed on each drinking day.

Daily ART Adherence.

The IVR system asked participants whether they adhered to their ART regimen using the following question: “We would be surprised if most people take 100% of their medications. If 0% means you have taken none of your HIV medications in the last 24 hours, 50% means you have taken half of your HIV medications in the last 24 hours, and 100% means you have taken every single dose over the past 24 hours. What percent of your HIV medications did you take on a scale of 0-100%?” (23,24). Given that the number of pills comprising each participant’s ART regimen varied, a conservative approach was taken, coding any response <100% as 1 = non-adherent, and 100% as 0 = adherent.

Data Analysis Plan

Data were analyzed using the Statistical Package for Social Sciences (SPSS) version 26 and Mplus version 7.4 (25). The criterion for statistical significance was set to an α level of 0.05. Descriptive statistics were used to summarize age, race, personal income, years since HIV diagnosis, highest level of education attained, and AUDIT scores. For continuous variables, means, medians, standard deviations, percentiles, and ranges were generated; frequencies and proportions were used for categorical variables.

Multilevel modeling was used to analyze the daily diary data to examine the daily-associations between alcohol consumption and ART adherence. Multilevel modeling procedures allow for analyzing data that have a nested multilevel structure, which in the case of these analyses constitutes assessment days (Level-1) nested within participants (Level-2). The within-person relationship between alcohol consumption and ART adherence is modeled for each participant individually (Level-1) and the average relationship across all participants is modeled separately (Level-2). Due to the dichotomous-nature of the daily ART adherence outcome variable, multilevel logistic regression models were estimated using a full information maximum likelihood estimation with robust standard errors to account for missing data and any variables with non-normal distributions. Fixed and random effects were modeled such that the intercepts and slopes were modeled first as random effects in logistic regressions. Consistent with other research in this area (26), if residual variances for the slopes were determined to not vary across Level-2 variables, they were set as fixed effects. Unstandardized coefficient estimates, standard errors (SEs), adjusted odds ratios (aORs) and 95% Confidence Intervals (95% CIs) are reported for model estimates.

The parent study was not originally designed to investigate the alcohol and ART-adherence relationship. Further, the ART-adherence item was added to only one site after the initiation of data-collection, and only 22 participants were assigned to the IVR condition prior to the conclusion of the study. Although at the between-person level (i.e., number of participants) the sample-size was relatively small compared to traditional sampling standards, each participant had the opportunity to provide data across the 42-day assessment period, yielding a maximum of 924 data points across the sample at the within-person level. To estimate the sample-size required to detect the hypothesized within-person effect of number of standard drinks on ART non-adherence, a power analysis was conducted using the Power In Two-Level Designs (PINT) v.212 software (27,28). PINT estimates standard errors of regression coefficients to conduct power analyses for multilevel models. We entered the observed effect size and covariance terms into the program to calculate the Level-1 and Level-2 sample-sizes required to detect the within-person effect of number of standard drinks on ART non-adherence. The results of this power analysis demonstrated that with α = 0.05 and ß = .80, a sample size that satisfied a standard error ≤ .06 would be required to yield sufficient statistical power to detect a “small” effect (Cohen’s d = .15). Additionally, the power analysis demonstrated one combination to detect a “small” effect of within-person number of standard drinks on ART non-adherence, a Level-2 sample of N = 22 and a Level-1 sample of n = 42 would yield a standard error = .07. Another power analysis was conducted to account for missing data at Level-1. This analysis demonstrated that with an average of n = 30 reporting days at Level-1, per N = 22 participants at Level-2, a standard error = .09, would yield sufficient statistical power to detect a “small” effect of Cohen’s d = .22. Thus, the results of this power analysis provided evidence to support sufficient statistical power was achieved to model “small” effects (d = .15 – .22) of within-person fluctuations in alcohol use and ART adherence, despite the relatively small Level-2 sample size (28).

To test Hypotheses 1 and 2, two separate multilevel models were estimated. Level-2 was defined by participant (N = 22) and Level-1 was defined by study days (n = 42) nested within participants. In both multilevel models, between-person differences (sample-mean centered) were partitioned from within-person fluctuations in daily predictors (person-mean centered) to control for between-person trends. Two separate models were specified to test (1) the temporal relationship between any alcohol consumption and daily ART adherence, and (2) a dose-response relationship between the amount of alcohol consumed and daily ART adherence. The two models were identical except for differing Level-1 predictors. To test Hypothesis 1, that days on which participants consumed any amount of alcohol would be associated with a greater likelihood of ART non-adherence, the unique Level-1 predictor was a dichotomous drinking day variable (1 = yes, 0 = no). To test Hypothesis 2, that days on which participants consumed higher than their own average levels of alcohol would be associated with a greater likelihood of ART non-adherence, the unique Level-1 predictor was a person-centered number of standard drinks. Additional fixed effects at Level-1 in both models included a mean-centered, linear time trend scaled to units of weeks (i.e., 42 study days represented as values ranging from −2.93 – 2.93); and a dichotomized weekend variable (i.e., Friday and Saturday versus weekdays). These variables were included as Level-1 predictors based on research demonstrating that drinking patterns and ART adherence can differ based on weekend versus weekday routines (29,30). Fixed effects at Level-2 included: age (sample-centered); years since HIV diagnosis (sample-centered); baseline ART adherence (sample-centered); and an average number of standard drink estimate (sample-centered). These variables were included as Level-2 predictors because of evidence suggesting that ART adherence changes over time (31,32), improves with age (33), and to control for between-person variability in recent adherence rates. The outcome for both multilevel models was daily ART non-adherence (i.e., dichotomized; 1 = non-adherent, 0 = adherent).

Results

Sample Characteristics

Participants included in these analyses were 22 HIV-positive MSM who were assigned to the IVR condition (Table 1). The average age of participants was 35.50 years (SD = 9.30), with an average of 6.37 years (SD = 4.79) since first being diagnosed with HIV. Most participants had completed a college degree or above (90.92%), 31.82% were Black, 54.55% were White, and 13.64% identified as mixed race or another race/ethnicity. The average score for the AUDIT was 11.59 (indicating at-risk for AUD; SD = 6.21).

Table 1.

Characteristics of HIV-positive MSM (N = 22) and Daily Diary Assessments (n = 924 days)

Participant Characteristics	M (SD)
Age (years)	35.50 (9.30)
AUDIT	11.59 (6.21)
Years since HIV diagnosis	6.37 (4.79)
<100% ART adherence (days)	3.82 (5.37)
Drinks per assessment day	1.04 (0.78)
Drinks per drinking day	3.21 (1.72)
Daily Diary Characteristics	n (%)a
Alcohol used and ART adherence reported	641 (69.37%)
Alcohol used and <100% ART adherence	26 (2.81%)
≥ 1 missed dose of ART	84 (9.09%)
≥ 1 standard drink consumed	184 (19.91%)
Heavy drinking dayb	59 (6.39%)

Note. AUDIT = Alcohol Use Disorder Identification Test, M = Mean, SD = Standard Deviation

^anumber of days endorsed during the daily diary assessment

^b≥ 5 standard drinks.

Daily Diary Assessment Characteristics

Out of a potential 924 person days, participants reported their alcohol use and ART adherence on 641 days (69.37%). Participants reported a total of 184 days (19.91%) on which they consumed at least one standard drink of alcohol and 84 days (9.09%) on which they missed at least one dose of ART medication. There was an average report of 1.04 (SD = 0.78) standard drinks per assessment day, 3.21 (SD = 1.72) standard drinks per drinking day, and a total of 59 days (6.39%) characterized as heavy drinking days (i.e., ≥ 5 standard drinks). Participants reported <100% ART adherence on an average of 3.82 days (SD = 5.37), and 10 participants (45%) reported perfect ART adherence during the 42-day assessment period. Co-occurring alcohol consumption and <100% ART adherence was reported on 26 (2.81%) days, with an average report of 1.65 (SD = 2.67) standard drinks consumed on non-adherent days (Table 1).

Multilevel Models

An unconditional model (i.e., without predictors) was estimated to calculate the amount of variance in daily ART non-adherence attributable to between-person differences. This model demonstrated an Intraclass Correlation of 0.37, indicating that approximately 63% of the variance in daily ART non-adherence can be attributed to within-person factors. Daily diary assessment day (i.e., number of days enrolled in the study) was not significantly related to daily ART non-adherence (γ = −0.03, OR = 1.04, 95% CI [0.82, 1.30], p = .77), indicating there was not a linear relationship between time and ART non-adherence. This suggests that there was no participant reactivity to the daily diary assessments (26). Nonetheless, based on recommendations to include random time coefficients when analyzing intensive longitudinal data (34), this random slope coefficient was retained in both models. Similarly, the relationships between the random slopes for the dichotomous drinking day predictor and number of standard drinks per day predictor with daily ART non-adherence were non-significant, and thus were set as fixed effects.

Drinking Day on ART Adherence.

A multilevel model was constructed to test the hypothesis that a day on which any amount of alcohol consumed would be associated with a higher likelihood of missing a dose of ART on that same day. The results of this analysis supported this hypothesis, in that on days in which participants consumed any amount of alcohol, they were almost 2.5 times more likely to miss a dose of ART on that same day (aOR = 2.48, 95% CI [1.18, 5.18], p = .02). Drinking day was a stronger predictor of ART non-adherence than whether the day was on a weekend, which was associated with 75% increased odds of non-adherence (aOR = 1.75, 95% CI [1.01, 3.05], p = .016). In terms of the between-person effects on ART non-adherence, results demonstrated that higher levels of ART adherence at baseline were associated with significantly reduced odds of non-adherence (aOR = 0.87, 95% CI [0.79, 0.96], p = .01). Additionally, the longer a participant was diagnosed with HIV (aOR = 0.65, 95% CI [0.51, 0.82], p < .001), and the older a participant was (aOR = 0.82, 95% CI [0.70, 0.97], p = .02), the less likely they were to be non-adherent. The participant’s average number of standard drinks consumed during the daily diary assessment period was not significantly associated with ART non-adherence.

Number of Drinks on ART Adherence.

A multilevel model was constructed to test the hypothesis that days on which a participant consumed a greater amount of alcohol than their own average would be associated with a higher likelihood of missing a dose of ART on that same day. The results of this analysis did not support this hypothesis, although the trend was in the predicted direction and approached statistical significance (aOR = 1.15, 95% CI [0.96, 1.37], p = .12). The within-person effect of weekend day was associated with two times increased odds of ART non-adherence on that day (aOR = 2.00, 95% CI [1.09, 3.64], p = .02). In terms of the between-person effects on ART non-adherence, results were largely consistent with the prior model. Specifically, the longer a participant was diagnosed with HIV (aOR = 0.67, 95% CI [0.50, 0.89], p = .01), and the older a participant was (aOR = 0.84, 95% CI [0.71, 0.99], p = .04), the less likely they were to be non-adherent. Notably, in contrast to the prior model, baseline levels of ART adherence were not significantly associated with ART non-adherence.

Discussion

This study sought to address gaps in the literature examining the relationship between alcohol consumption and ART adherence. We investigated within- and between-person associations of same-day alcohol use and ART adherence among MSM. Results supported the hypothesis that consuming any amount of alcohol is associated with increased odds of ART non-adherence on the same day. This same-day co-occurrence association remained statistically significant even after adjusting for well-established correlates of ART adherence, including age, time since HIV diagnosis, weekend versus weekday, average levels of alcohol consumption, and baseline levels of ART adherence. These findings extend the broader event-level literature by establishing a prospective within-person effect of alcohol consumption on ART adherence to a non-treatment seeking MSM sample. However, the results from this study did not wholly support findings from prior research that has demonstrated a dose-response relationship between the quantity of alcohol consumed and increased odds of ART non-adherence.

Turner and colleagues (19) conducted the only other prospective examination of the event-level association between alcohol consumption and ART non-adherence among MSM. Because participants in their study enrolled in an intervention designed to improve HIV care treatment outcomes (e.g., viral suppression), it is plausible that the magnitude of the within-person effects of alcohol on ART adherence would be attenuated. Comparatively, it may be expected in a sample of non-treatment seeking HIV-positive MSM, alcohol use under naturalistic conditions may exhibit a stronger influence over ART adherence. We thus anticipated that the magnitude of the within-person effects in this sample could have been greater than that found in Turner et al. (19). Instead, findings from this study indicated only a slightly stronger effect. Specifically, daily alcohol use was associated with approximately 2.5 increased odds in same-day ART non-adherence in our study compared to 1.9 times increased odds reported by Turner et al. (19). One potential explanation for the similarity in results may be due to the statistical analyses conducted by Turner et al. (19). Their analysis did not differentiate the within-person fluctuations in daily alcohol consumption from between-person average levels of alcohol use, nor were well-established correlates of alcohol use and ART adherence included, such as baseline levels of ART adherence at Level-2 and day of the week at Level-1. Thus, it is possible that the same-day association between alcohol and ART non-adherence demonstrated by Turner et al. (19) may be partially attributable to between-person factors that were not estimated independently from those at the within-person level (e.g., average levels of ART adherence). Alternatively, because the Turner et al. (19) intervention did not contain an explicit focus on alcohol reduction, the strength of the association between alcohol and ART adherence may not have changed as a function of the intervention. This highlights the need for future work to further explicate proximal factors influencing ART non-adherence that can explain the underlying mechanisms of the alcohol-ART adherence relationship.

Of note, this study was not able to test the potential mechanisms by which alcohol use affects ART adherence. As outlined in the integrated model by Woolf-King et al. (9), there are distinct pathways based on whether non-adherence is intentional or unintentional. For example, missing a dose of ART could be unintentional, and attributable to “forgetting” resulting from acute alcohol intoxication. The model also suggests alcohol-related ART avoidance behaviors, stemming from alcohol-ART interactive toxicity beliefs (35), as an alternative mechanism through which alcohol consumption is associated with intentional ART non-adherence. Indeed, there is a growing body of literature supporting this pathway (35), however, to our knowledge no event-level study has distinguished alcohol-ART interactive toxicity belief-related intentional non-adherence from unintentional non-adherence. Instead, this area of research has relied on situational association studies (e.g., 15) that assess participants’ typical ART adherence behavior in the context of alcohol consumption, as opposed to assessing participants’ reasons for missing a dose of ART (intentional vs. unintentional) on specific days. Consequently, it is plausible that participants were more likely to be non-adherent to ART regimens on days in which they consumed alcohol due to other factors unrelated to the acute cognitive effects of alcohol intoxication. For instance, drinking is often a social activity, and research has demonstrated that PWH may intentionally skip doses while in the presence of others in an effort to avoid potential unwanted disclosure of HIV-status and anticipated stigma associated with HIV-positive serostatus (36,37). Additional research is needed to characterize the factors that underlie the relationship between alcohol and ART non-adherence, which can be used to inform the design of effective interventions.

Contrary to our prediction, there was only partial evidence of the hypothesized within-person dose-response relationship between drinking at above average levels and ART non-adherence. Nonetheless, findings from the multilevel model showed that the pattern of results was in the predicted direction, and the magnitude of the effect was similar to that demonstrated by Parsons et al. (18). Given participants completed an average of 29 daily diaries out of a potential 42, this null finding may be due to inadequate statistical power required to detect the within-person effect, resulting in a Type II error. Moreover, heavy drinking was relatively infrequent, comprising only 6% of the assessment days, another potential limitation in an ability to detect a dose-response effect. To illustrate, weekends were associated with a two times greater odds of ART non-adherence, even after adjusting for number of standard drinks consumed. Thus, any potential covariation between above-average drinking on weekends could have accounted for an overlapping proportion of within-person variance in ART non-adherence. Another potential explanation may be related to MSM developing compensatory strategies to mitigate negative consequences associated with drinking, such as ART non-adherence (38). Indeed, given that this sample of MSM endorsed episodes of heavy drinking that were on average consistent with national estimates (13,14), they may have developed “behavioral tolerance” demonstrated by hazardous and non-hazardous drinkers (38) that buffered against unintentional non-adherence. As such, even alcohol consumption above a person’s average may not have produced sufficient prospective memory failure for unintentional ART non-adherence, rather, moderate-heavy doses that impair psychomotor functioning may be required to influence medication-taking. Future research should consider incorporating heavy drinking episode assessments as an additional within-person predictor of ART-nonadherence to complement the variable of drinking above one’s own average offered by multilevel modeling.

Strengths, Limitations, and Directions for Future Research

This study adds to event-level research aiming to develop a nuanced understanding of the within-person fluctuations that influence the relationship between alcohol consumption and ART adherence. The use of a prospective design with 42 daily diary assessments allowed for the separation of within-person effects of alcohol consumption from between-person factors typically associated with ART adherence, while mitigating potential errors related to recall bias. Incorporating multiple well-established between-person predictors of ART non-adherence allowed for a conservative estimate of the within-person effect of alcohol consumption. Moreover, this study demonstrated that after accounting for daily fluctuations in drinking at the within-person level, average levels of drinking at the between-person level were not a significant predictor of ART non-adherence—providing further evidence that event-level designs offer an advanced understanding of this complex relationship beyond what is afforded by traditional global association research. An additional strength was the use of a strong theoretical foundation to test two facets of alcohol consumption (i.e., temporal versus dose-response) to further elucidate its link with ART non-adherence among a sample of HIV-positive MSM.

The results of the study should also be considered in the context of its limitations. First, although there is no “gold standard” of measuring ART adherence, adherence was assessed using self-report questionnaires, which are vulnerable to social desirability biases and can yield overestimates of true adherence (39). Indeed, participants reported perfect adherence on 91% of assessment days, which is substantially greater than national estimates, potentially limiting the ability to detect the anticipated relationships due to ceiling effects. Future research should use objective measures of ART adherence (e.g., pharmacy refills) and/or biomarkers of ART to improve measurement accuracy (40). Another measurement limitation was the absence of adherence assessments that differentiate intentional non-adherence from unintentional non-adherence, thus precluding our ability to determine whether missing doses of ART on alcohol use days was attributable to cognitive impairment or ART avoidance behaviors. Additionally, other contextual factors that often co-occur with alcohol consumption and have been linked to non-adherence (e.g., the presence of others, deviation from medication-taking routines) were omitted. Further, although the focus of this study was designed to investigate within-person associations, the small number of participants reduced the confidence in the interpretation of between-persons effects. Though results of a power-analysis indicated sample-sizes of N = 22 participants and n = 42 assessment days would yield sufficient power to detect a within-person effect of average number of standard drinks on ART non-adherence, participants only reported on a total of 642 assessment days out a potential 924, and n = 10 of the participants reported perfect ART adherence during the 42-day assessment period. This potentially restricted the statistical power necessary to detect within-person effects. Therefore, it remains a possibility that with a larger sample size, there would have been sufficient evidence in support of the dose-response hypothesis. Last, the majority of participants in this study were engaged in HIV care, and reported generally high rates of ART adherence, which may limit the generalizability of these findings to MSM who are at higher risk for experiencing suboptimal adherence. Future research can address these limitations by recruiting large samples of HIV-positive MSM across the continuum of care, with a wide-range of drinking patterns, and employing prospective intensive longitudinal assessment methodologies to investigate the relationship between alcohol consumption and subsequent intentional/unintentional non-adherence with objective assessment tools.

Clinical Implications

Findings from this study can be used to enhance clinical intervention efforts. Specifically, interventions designed to reduce alcohol consumption and improve ART adherence can benefit from incorporating content that targets improving ART adherence by addressing the relationship between the two behaviors. A recent review of behavioral interventions targeting alcohol use among PWH determined that existing interventions have limited efficacy for reducing alcohol consumption and HIV-related secondary outcomes (41), and only identified three trials that consisted of MSM samples (42–44). Of these three, only one (42) reported HIV-treatment outcomes (i.e., viral load, CD4 count). Kahler et al. (42) demonstrated that, compared to the control condition, participants assigned to the MI treatment condition reported decreased frequency and quantity of alcohol consumption at six- and 12-months post-baseline. However, there were no significant treatment effects on HIV-related outcomes. Accordingly, future interventions can consider drawing on the within-person temporal relationship between alcohol and ART adherence and integrate content on this topic into treatment. For instance, delivering psychoeducation about the frequent co-occurrence of the two behaviors and dispelling erroneous alcohol-ART interactive toxicity beliefs may provide an efficient path for simultaneously reducing alcohol consumption and improving ART adherence. Another promising intervention is illustrated by Hasin et al. (45). The HealthCall mobile smartphone application was designed to augment brief drinking-reduction interventions (e.g., MI) delivered to alcohol-dependent HIV-positive patients receiving treatment in a primary care setting. HealthCall utilizes self-monitoring and daily self-reports of drinking and other health behaviors (e.g., ART adherence) in combination with personalized normative feedback to promote treatment engagement. Results of an RCT found, compared to a control condition, HealthCall to be associated with the greatest reductions in alcohol consumption, as well as preliminary evidence of improved ART adherence (45). Finally, research should consider tailoring intervention programming to meet the specific needs of HIV-positive MSM drinkers given the varied and unique multilevel obstacles that impede optimal ART adherence as compared to other HIV-positive populations (46).

Conclusion

Approximately 43% of MSM with HIV are not virally suppressed, and alcohol use remains one of the strongest predictors of ART non-adherence (8,9). This event-level examination of the daily associations between alcohol consumption and ART adherence in a sample of MSM demonstrated that consuming any amount of alcohol was associated with increased odds of ART non-adherence; however, drinking at above average levels was not. These findings highlight the importance of combining intervention efforts that address alcohol consumption and suboptimal ART adherence. Future research investigating the mechanisms by which alcohol consumption is related to intentional and unintentional ART non-adherence is warranted.

Table 2.

Multilevel Model of Drinking Day Predicting Daily ART Adherence

Daily ART Adherence
	Estimate (SE)	Odds Ratio	95% CI	p-value
Fixed Effects
Level 2 (Between-Person)
Intercept	3.94 (0.60)	---	[2.76, 5.12]	< .001
Years since HIV Diagnosis	−0.43 (0.12)	0.65	[0.51, 0.82]	< .001
Age	−0.19 (0.08)	0.82	[0.70, 0.97]	.02
Baseline ART Adherence	−0.14 (0.05)	0.87	[0.79, 0.96]	.005
Average Number of Drinks	−0.03 (0.79)	0.97	[0.21, 4.56]	.97
Level 1 (Within-Person)
Drinking Day (Dichotomous)	0.91 (0.38)	2.48	[1.18, 5.18]	.016
Study Day	0.08 (0.11)	1.09	[0.88, 1.34]	.45
Weekend (Friday/Saturday)	0.91 (0.38)	1.75	[1.01, 3.05]	.048
Random Effects
Level 2 (Between-Person)
Intercept	3.70 (0.57)	---	[2.57, 4.81]	< .001
Level 1 (Within-Person)
Study Day	0.06 (0.14)	1.06	[0.78, 1.35]	.68
Residual	3.18 (1.91)	---	[−0.56, 6.92]	.10

Note. N = 22. Alcohol Use Predictor highlighted in bold font.

SE = Standard Error, 95% CI = 95% Confidence Interval.

All Level-2 variables are sample-centered; all continuous Level-1 variables are within-person centered.

Table 3.

Multilevel Model of Number of Standard Drinks Predicting Daily ART Adherence

Daily ART Adherence
	Estimate (SE)	Odds Ratio	95% CI	p-value
Fixed Effects
Level 2 (Between-Person)
Intercept	3.73 (0.60)	---	[2.56, 4.90]	< .001
Years since HIV Diagnosis	−0.40 (0.15)	0.67	[0.50, 0.89]	.006
Age	−0.18 (0.08)	0.84	[0.71, 0.99]	.04
Baseline ART Adherence	−0.11 (0.06)	0.90	[0.79, 1.02]	.09
Average Number of Drinks	0.18 (0.86)	1.20	[0.22, 6.46]	.83
Level 1 (Within-Person)
Number of Standard Drinks	0.14 (0.09)	1.15	[0.96, 1.37]	0.12
Study Day	0.08 (0.12)	1.09	[0.86, 1.38]	.49
Weekend (Friday/Saturday)	0.69 (0.31)	2.00	[1.09, 3.64]	.02
Random Effects
Level 2 (Between-Person)
Intercept	3.42 (0.56)	---	[2.32, 4.52]	< .001
Level 1 (Within-Person)
Study Day	0.03 (0.17)	1.03	[.73, 1.43]	0.89
Residual	3.57 (2.49)	---	[−1.30, 8.45]	.15

Note. N = 22. Alcohol Use Predictor highlighted in bold font.

SE = Standard Error, 95% CI = 95% Confidence Interval.

All Level-2 variables are sample-centered; all continuous Level-1 variables are within-person centered.