Alcohol interactive toxicity beliefs and ART non-adherence among HIV-infected current drinkers in Mbarara, Uganda

Abstract

Interactive toxicity beliefs regarding mixing alcohol and antiretroviral therapy (ART) may influence ART adherence. HIV-infected patients in Uganda completed quarterly visits for one year, or one visit at 6 months, depending on study randomization. Past month ART non-adherence was less than daily or <100% on a visual analog scale. Participants were asked if people who take alcohol should stop taking their medications (belief) and whether they occasionally stopped taking their medications in anticipation of drinking (behavior). Visits with self-reported alcohol use and ART use for ≥30 days were included. We used logistic regression to examine correlates of the interactive toxicity belief and behavior, and to determine associations with ART non-adherence. 134 participants contributed 258 study visits. The toxicity belief was endorsed at 24%, the behavior at 15%, and any non-adherence at 35% of visits. In multivariable analysis, the odds of non-adherence were higher for those endorsing the toxicity behavior (adjusted odds ratio (AOR): 2.06; 95% confidence interval (CI): 0.97–4.36) but not the toxicity belief (AOR: 0.63; 95%CI: 0.32–1.26). Clear messaging about maintaining adherence, even if drinking, could benefit patients.

Introduction

Antiretroviral therapy (ART) can dramatically improve the health and lives of those infected with HIV and lead to viral suppression. However, the efficacy of ART depends upon treatment adherence; the minimum level of ART adherence needed to achieve HIV viral suppression varies by ART class (1). Various factors are recognized to be barriers to ART adherence, including lack of social support, being unmarried, depression, HIV stigma, food insecurity, transportation cost and time to clinic, and alcohol use (2–9). Alcohol use in particular has been found to be associated with missed doses and treatment interruptions (10–12). One possible explanation for the consistent association observed between alcohol use and poor adherence may be explained by alcohol myopia theory, the idea that acute alcohol use causes short-sightedness, and can cause users to focus on and respond primarily to their immediate environment (13), leading to unintentional non-adherence. However, a phenomenon termed interactive toxicity beliefs may be another important cause of reduced ART adherence among drinkers. Interactive toxicity beliefs are the beliefs that mixing alcohol and ART is toxic, and endorsing such beliefs may lead patients to purposefully alter their medication adherence while they are drinking alcohol, or when they plan to drink.

A small number of studies have described alcohol toxicity beliefs and their impact on HIV outcomes among people living with HIV in the United States. In a pilot study in Florida, 20% of participants reported “weekending” (drinking more alcohol on weekends, and intentionally skipping one or two days of ART due to drinking plans) (14). Sankar et al. found that most (85%) participants believed that alcohol and ART shouldn’t be mixed; 51% reported that they wouldn’t take their medications if they had been drinking alcohol (15). Similarly, Kalichman et al. found interactive toxicity beliefs to be common among HIV-positive drinkers and non-drinkers in Georgia (16). They found that stopping ART while drinking alcohol was associated with non-adherence, after adjusting for scores on the Alcohol Use Disorders Identification Test (AUDIT) and demographics. In a longitudinal study of HIV infected drinkers, half of the participants reported skipping or stopping their HIV medications while drinking. Those that endorsed interactive toxicity beliefs were more likely to have poor adherence on drinking days (17).

Interactive toxicity beliefs in settings other than the US have been less studied, thus far. In a qualitative study among HIV-positive adults new to HIV care in Uganda, however, it was common for participants to describe being told by clinic staff that alcohol and ART “don’t mix”, and that alcohol weakens the effect of ART (18). In a qualitative study of determinants of adherence among patients on ART for at least 6 months in Tanzania (19), alcohol use was commonly reported as a reason for both unintentional and intentional non-adherence. Among HIV infected drinkers in South Africa, several patterns of taking ART were reported when drinking, including taking ART early, taking ART as scheduled, and skipping ART, with about half of the participants reporting a combination of these ART use patterns on drinking days (20). Reasons for the different adherence patterns, however, were not assessed. As such, it appears that the patterns of ART use among HIV-positive adults in sub-Saharan Africa (SSA) may be altered, both unintentionally and intentionally, by drinking. Additionally, there is some evidence from the two qualitative studies noted above that alcohol interactive toxicity beliefs may be common. However, the specific associations between participant alcohol interactive toxicity beliefs and medication adherence have yet to be examined in SSA.

The purpose of this study was to: (1) determine the prevalence of interactive toxicity beliefs and behaviors, (2) describe the correlates of these, and (3) examine whether they were associated with self-reported ART non-adherence, in a longitudinal study of HIV-infected adult drinkers new to HIV care in Uganda. We hypothesized that endorsement of the interactive toxicity belief and behavior would be associated with ART non-adherence among our study participants.

Methods

Study population

The Biomarker Research on Ethanol Among Those with HIV (BREATH) Study was a mixed methods study of changes in alcohol consumption during the first year of HIV care. New adult patients attending the Immune Suppression Syndrome (ISS) Clinic in Mbarara, Uganda, were invited to participate. Eligibility criteria included: age ≥18 years old, fluency in English or Runyakole (the local language), residence within 60km of the clinic, being new to the ISS Clinic and HIV care, and either self-reporting alcohol consumption within the past year, or being suspected of recent alcohol consumption by the clinic counselor (<1% of participants).

Participants were randomized to one of two study arms: the main cohort study arm, or a minimally assessed study arm designed to examine assessment reactivity (21). Those randomized to the main cohort arm completed structured quantitative interviews quarterly for one year; a sub-set of these participants also completed qualitative interviews every six months during this year (18). Participants randomized to the minimally assessed arm were interviewed only once, six months following study enrollment. At each study visit, participants completed an interviewer-administered structured interview, and underwent breath alcohol concentration testing and phlebotomy. There is no systematic protocol in place for alcohol interventions in this setting; however, if a study participant scored ≥20 on the AUDIT (indicative of probable dependent alcohol use) (22), requested additional help decreasing their alcohol consumption, endorsed suicidal ideation on the survey (“thoughts of ending your life”), or seemed especially distressed during their study interview, they were referred to a mental health counselor in the Psychiatry Department of Mbarara Regional Referral Hospital for care. Study participants engaged in HIV clinical care, including initiating ART, independently of study activities. All study procedures were approved by Institutional Review Boards at the University of California San Francisco, the Mbarara University of Science and Technology (MUST), and the Uganda National Council for Science and Technology.

Laboratory measurements

Specimens were collected for HIV viral load concurrently with the first study interview (Bayer System 340 bDNA analyzer; Bayer Healthcare Corporation, Whippany, NJ), and for CD4+ cell count at baseline, 6-, and 12-month visits (Coulter Epics XL.MCL Cytometer; Beckman Coulter, Brea, CA). CD4 count and viral load testing were conducted at the MUST Clinical and Research Laboratory.

Specimens were also collected for phosphatidylethanol (PEth) testing at each study visit. PEth is a phospholipid which forms only in the presence of alcohol; it has been shown to be highly sensitive and specific for any and heavy alcohol use among persons with HIV in Uganda (23). Venous blood samples were collected from participants by clinic staff at each study visit, and transferred to dried blood spot (DBS) cards by laboratory staff. DBS cards were stored at −80°C at the MUST Clinical and Research Laboratory until they were shipped to the United States Drug Testing Laboratories, Inc. (USDTL), where they were tested for PEth using liquid chromatography-tandem mass spectrometry (LC-MS/MS) following extraction into methanol (24). The lower limit of detection was 8 ng/ml, and the most common PEth homologue (16:0/18:1) was detected.

Variables

ART non-adherence

There is currently no gold standard for measuring or defining ART non-adherence (25). In our study, any ART non-adherence was defined using self-report. Self-reported adherence is thought to be an over-report of actual adherence (26–28), thus we chose to use a conservative definition of any ART non-adherence. As such, any self-reported ART non-adherence in the past 30 days was defined using a composite variable made from the following questions regarding ART adherence, asked at each study visit: 1. “In the past 30 days, did you take all your anti-HIV pills every day?”; 2.“In the past 30 days, how many days, in total, have you not taken your anti-HIV medication pills?”; 3. Using a visual analog adherence scale, participants were asked, on a scale from 0–100 (100 = all doses), to indicate how many of their ART doses they took in the past 30 days (29). Participants who reported not taking their ART on every day, or reported <100 on the adherence scale, were considered to be non-adherent in the past 30 days at that visit. We also asked participants if they were unable to get their ART in the past three months because the pharmacy was out of pills; participants that replied yes to this question (n=5) were excluded from our measure of non-adherence (set to missing).

Interactive toxicity items

At each visit, participants were asked questions regarding mixing alcohol use and medications. We modified questions from Kalichman et al (16), instructing participants to consider co-trimoxazole, multivitamins, and ART as medications, and simplifying the language for ease of translation into the local language. For this analysis, we focused on two of these questions, one an interactive toxicity “belief”: “I think that people who take alcohol should stop taking their medications because they should not mix them”, and the other a “behavior”: “I occasionally stop taking any medications I am on if I think I will be drinking alcohol”. Response options were true, false, or not applicable.

Covariates

We included participant demographics, as well as factors found to be associated with ART adherence or interactive toxicity beliefs, in this analysis.

Demographics

We examined demographic characteristics including sex, age, religion/denomination, education, marital status, and social support, and household characteristics including household assets, food insecurity, and travel time to the clinic. Perceived social support was measured using a modified version of the Duke-UNC Functional Social Support Scale (30) (Cronbach’s alpha = 0.89); a mean score <3 was considered “low social support”. The household asset index was created using principal components analysis, and grouped participants’ households based on ownership of durable goods, housing quality, and available energy sources (31, 32). The bottom 40% was considered low, the middle 40% as middle, and the top 20% as high. The Household Food Insecurity Access Scale (HFIAS) was used to assess food insecurity (33, 34) (Cronbach’s alpha = 0.94); participants were categorized into three groups (food secure, mildly/moderately insecure, severely insecure) for analysis due to small numbers in the mildly insecure food access group. Participants were asked how long it takes to travel from their home to the ISS Clinic; we created four groups (0–20 minutes, 21–40 minutes, 41–60 minutes, >60 minutes).

Mental and physical health status and clinical contact

General health status was assessed using the first question of the Medical Outcomes Study HIV Health Survey (MOS-HIV) at each study visit (35, 36). While the MOS-HIV has been shown to be a valid measure of health in Uganda, the first question (“In general, would you say your health is excellent, very good, good, fair or poor?”) has not been validated among people with HIV. However, this single item has been shown to predict mortality and healthcare utilization as well as multi-item scales among Veterans Administration outpatients in the US (37). As such, we chose to use this single item as an indicator of general health status, and dichotomized responses as excellent/very good/good versus fair/poor. Depressive symptoms were assessed using a modified 16-item version of the Hopkins Symptom Checklist that has been validated for use in persons with HIV in Uganda (38–40) (Cronbach’s alpha = 0.90). As inclusion of somatic symptoms has been shown to inflate depression scores among HIV-infected people, we chose to exclude the four somatic items (“feeling low in energy, slowed down,” “feeling fidgety,” “poor appetite,” and “having difficulty falling or staying asleep”) for our analyses (41–43). Cronbach’s alpha for the 12-item version of this scale in our study was 0.87. The total 12-item score was then averaged; participants with a score of 1.75 or more were classified as having symptoms of probable depression. We created a variable to indicate the number of clinic visits in the past three months, retrieved from the MUST ISS Clinic electronic medical records.

Alcohol use

Alcohol use was assessed at each visit, using both self-report and PEth. We used the Alcohol Use Disorders Identification Test – Consumption (AUDIT-C), a 3-item scale with a total score ranging from 0–12 (44, 45), to classify participants as self-reported non-drinkers (AUDIT-C=0), self-reported low risk drinkers (AUDIT-C=1–3 for men, 1–2 for women), or self-reported unhealthy drinkers (“AUDIT-C positive”: AUDIT-C ≥4 for men, ≥3 for women) in the past three months. Because we have previously observed under-reporting of alcohol use in this population (46–48), we created a composite variable to better capture any unhealthy use using both self-report and PEth. Combining two measures with high specificity improves the sensitivity of the measure, compared to using either measure alone (49). As such, unhealthy alcohol use was defined as a positive AUDIT-C score, or PEth ≥50 ng/ml. This PEth cutoff for unhealthy alcohol use was highly sensitive (93%) and reasonably specific (83%) for detecting average daily drinking of at least 2 drinks per day in a study of 222 patients with liver disease (S. Stewart, personal communication); another study among 80 reproductive age women showed a cutoff of 45 ng/ml had 61% sensitivity and 95% specificity for the same level of drinking (50).

Analysis

All analyses were limited to visits at which the participants reported taking ART for at least 30 days. We also limited the analyses to those visits in which the participant reported consuming any alcohol in the prior three months, because (1) we felt that the responses about adherence and interactive toxicity would be more accurate among those reporting recent alcohol use in a population in which we have observed socially desirable reporting, and (2) perceptions of the toxicity of mixing ART and drinking are most relevant to current drinkers. To describe the participants included in our analyses, we calculated frequency distributions for categorical variables and medians and inter-quartile ranges (IQR) for continuous variables.

We described endorsement (response choice = true) of the interactive toxicity items at each study visit using bivariate generalized estimating equation (GEE) logistic regression models for both items separately, with robust standard errors and exchangeable working correlations.

To analyze ART non-adherence, we conducted bivariate and multivariable GEE logistic regression models of non-adherence, using robust standard errors and exchangeable working correlations. We used a purposeful selection approach to create the multivariable model (51); both interactive toxicity items were forced into the multivariable model. Covariates were initially included if they were associated with ART non-adherence, or either of the interactive toxicity items, in bivariate analyses at a p-value ≤0.25. They were then excluded in a backwards stepwise manner, until all remaining variables were associated at p≤0.10. Next, any covariates initially excluded based on the p-value ≤0.25 cut-off were individually added into the model, and their significance re-assessed; they were retained if they were associated at p ≤0.10. This process continued iteratively until no new covariates were added or removed.

Results

Participant characteristics

A total of 205 participants were enrolled in the main BREATH Study cohort, and 141 participants were enrolled in the minimally assessed comparison arm and completed their study interview at 6 months, for a total of 346. Of these participants, 215 were on ART for 30 days during at least one study interview and 134 of those admitted to any alcohol consumption in the prior three months. These 134 participants were on ART and reported any alcohol use in 258 follow-up visits, which were included in these analyses. 53% of participants completed 1 visit, 16% completed 2 visits, 16% completed 3 visits, and 15% completed 4 visits.

Forty-nine of the participants (37%) were female, slightly more than half were married (55%), and 38% had more than a primary education (Table 1). The median age was 31 years (IQR: 26–37).

Table 1.

Characteristics of BREATH Study participants who initiated ART within the first year of HIV care and self-reported any alcohol use in the past 3 months, Mbarara, Uganda (n = 134).

		N (%)
Baseline characteristics
Sex
Male		85 (63.4)
Female		49 (36.6)
Marital status
Married		73 (54.5)
Not married		61 (45.5)
Age (years) (median (IQR))		31 (26–37)
18–25		33 (24.6)
26–35		58 (43.3)
>35		43 (32.1)
Education
Primary education or less		83 (61.9)
More than a primary education		51 (38.1)
Religion
Protestant		83 (61.9)
Catholic		45 (33.6)
Other		6 (4.5)
Household assets
Low		50 (37.3)
Middle		60 (44.8)
Rich		24 (17.9)
Household Food Security
Food secure		47 (35.1)
Mildly/moderately food insecure		50 (37.3)
Severely food insecure		37 (27.6)
Low social support?
No		111 (84.1)
Yes		21 (15.9)
Time from ISS Clinic (minutes)
0–20		35 (26.1)
21–40		37 (27.6)
41–60		38 (28.4)
>60		24 (17.9)
Study arm
Main cohort arm		92 (68.7)
Minimally assessed arm		42 (31.3)
At the first study visita
Unhealthy alcohol use, past 3 months (self-report hazardous or PEth ≥ 50)
Yes		85 (63.4)
No		49 (36.6)
General health status
Good/very good/excellent		116 (86.6)
Fair/poor		18 (13.4)
Affective symptoms of depression, past 3 months? (Hopkins Symptom Checklist average > 1.75)
Yes		12 (9.0)
No		122 (91.0)
Summary over all study visits
Number of months on ART during the study (median (IQR))		7 (5–11)
Any non-complete ART adherence, past 30 days? (each visit)
Yes		88 (34.8)
No		165 (65.2)
Number of pills missed, past 30 days (each visit) (median (IQR))		0 (0-0)
Interactive toxicity items (each visit)
“I think that people who take alcohol should stop taking their medications because they should not mix them” (belief)
True		62 (24.0)
False		192 (74.4)
Don’t know		4 (1.6)
“I occasionally stop taking any medications I am on if I think I will be drinking alcohol” (behavior)
True		38 (14.7)
False		218 (84.5)
Not applicable		2 (0.8)

^aFirst visit included in this analysis was first visit on ART for at least 30 days and self-reported alcohol consumption

At the first study interview following ART initiation, 9% screened positive for symptoms of depression in the past three months, and 13% reported their health status as fair or poor. Eighty-five (63%) were positive for unhealthy alcohol use.

Interactive Toxicity Items

At approximately one-fourth (24%) of visits, participants endorsed the interactive toxicity belief; participants endorsed the interactive toxicity behavior at 15% of visits. Participants endorsed both items at 8% of visits. In bivariate analyses, the odds of endorsement for each item were higher with increasing age, although the association with the belief did not reach statistical significance (Wald X² = 4.74; p-value = 0.09) (Table 2). The odds of endorsing the interactive toxicity belief were increased for those with a higher number of clinic visits in the past three months (Wald X² = 5.02; p-value = 0.03), and were higher for participants with low perceived social support (Wald X² = 4.74; p-value = 0.03). The odds of endorsing the interactive toxicity behavior were significantly higher among those with a higher household asset index (Wald X² = 6.16; p-value = 0.05), and for Catholic participants (compared to Protestants) (Wald X² = 4.17; p-value = 0.04), and lower for those with longer travel times to the clinic (Wald X² = 8.10; p-value = 0.04). There were no statistically significant associations between any of the other covariates included and endorsement of these interactive toxicity items in our study.

Table 2.

Bivariate odds ratios (ORs) and 95% confidence intervals (CI) for endorsing alcohol interactive toxicity items, BREATH Study, Mbarara, Uganda (n=134 participants, n = 258 observations).

		Interactive Toxicity Belief: “I think that people who take alcohol should stop taking their medications because they should not mix them”		Interactive Toxicity Behavior: “I occasionally stop taking any medications I am on if I think I will be drinking alcohol”
		OR (95% CI)	Wald X2 (p-value)	OR (95% CI)	Wald X2 (p-value)
Demographics at baseline
Sex			0.13 (0.72)		0.01 (0.94)
Female		1.00		1.00
Male		1.13 (0.57, 2.24)		0.97 (0.43, 2.18)
Marital status			0.01 (0.93)		0.74 (0.39)
Married		1.00		1.00
Not married		0.97 (0.50, 1.88)		0.69 (0.30, 1.59)
Age (years)			4.74 (0.09)		7.40 (0.02)
18–25		1.00		1.00
26–35		2.79 (1.09, 7.14)		3.68 (1.09, 12.44)
>35		1.86 (0.70, 4.96)		1.17 (0.27, 5.03)
Education			1.11 (0.29)		0.81 (0.37)
Primary education or less		1.00		1.00
More than primary education		0.69 (0.35, 1.37)		1.45 (0.64, 3.28)
Religion			0.49 (0.78)		4.17 (0.12)
Protestant		1.00		1.00
Catholic		1.00 (0.50, 1.98)		2.38 (1.04, 5.49)
Other		0.46 (0.05, 4.15)		1.46 (0.16, 13.66)
Household assets			2.96 (0.23)		6.16 (0.05)
Low		1.00		1.00
Middle		1.70 (0.84, 3.43)		2.44 (0.97, 6.15)
Rich		0.88 (0.34, 2.29)		3.66 (1.23, 10.89)
Household Food Security			4.39 (0.11)		1.01 (0.60)
Food secure		1.00		1.00
Mildly/moderately food insecure		0.74 (0.33, 1.67)		0.62 (0.24, 1.59)
Severely food insecure		1.66 (0.73, 3.79)		0.74 (0.27, 2.04)
Low social support?			4.74 (0.03)		2.38 (0.12)
No		1.00		1.00
Yes		2.41 (1.09, 5.33)		2.08 (0.82, 5.29)
General health status, past 3 months			0.07 (0.80)		0.02 (0.88)
Good/very good/excellent		1.00		1.00
Fair/poor		1.13 (0.45, 2.81)		1.08 (0.40, 2.92)
Affective symptoms of depression, past 3 months?			0.11 (0.74)		0.03 (0.87)
No		1.00		1.00
Yes		1.22 (0.39, 3.80)		1.12 (0.31, 4.04)
Unhealthy alcohol use, past 3 months (self-report hazardous or PEth ≥ 50)			0.10 (0.76)		0.45 (0.50)
No		1.00		1.00
Yes		0.90 (0.47, 1.74)		0.78 (0.38, 1.61)
Study Arm			0.91 (0.34)		0.09 (0.77)
Main cohort arm		1.00		1.00
Minimally assessed arm		1.45 (0.68, 3.08)		1.15 (0.45, 2.94)
Time to clinic (minutes)			1.53 (0.68)		8.10 (0.04)
0–20		1.00		1.00
21–40		1.20 (0.49, 2.89)		0.21 (0.05, 0.81)
41–60		0.70 (0.31, 1.60)		0.43 (0.16, 1.11)
>60		0.83 (0.31, 2.22)		0.27 (0.08, 0.92)
Months on ART at this visit		0.94 (0.86, 1.02)	2.00 (0.16)	1.01 (0.92, 1.10)	0.01 (0.91)
Number of clinic visits, past 3 months		1.28 (1.03, 1.59)	5.02 (0.03)	0.94 (0.69, 1.29)	0.15 (0.69)

ART non-adherence

Over all visits included here (n = 258 visits), any ART non-adherence was reported at 88 visits (35%). Among visits where participants reported missing their ART on at least one day in the past 30 days (n = 62), the median number of days not taking ART was 2 (IQR: 1–3).

In bivariate analysis, the odds of any ART non-adherence were higher for those endorsing the interactive toxicity behavior (OR: 1.60; 95%CI: 0.84–3.02), and lower for those endorsing the interactive toxicity belief (OR: 0.76; 95%CI: 0.42–1.39); these associations were not statistically significant (Table 3). The odds of non-adherence were higher for those with recent unhealthy alcohol use (OR: 1.92; 95%CI: 1.10–3.33), for unmarried participants (OR: 2.04; 95%CI: 1.14–3.66), and for those with symptoms of depression (OR: 2.13; 95%CI: 0.81–5.63) (not statistically significant). No other variables were associated with non-adherence in bivariate analysis.

Table 3.

Bivariate and multivariable odds ratios (OR) and 95% confidence intervals (CI) for any ART non-adherence in the prior 30 days, BREATH Study, Mbarara, Uganda (n=134 participants, n=253 observations).

		Bivariate		Multivariable
		OR (95% CI)	Wald X2 (p-value)	OR (95% CI)	Wald X2 (p-value)
Interactive toxicity items, each visit
“I think that people who take alcohol should stop taking their medications because they should not mix them” (belief)			0.79 (0.38)		1.72 (0.19)
False		1.00		1.00
True		0.76 (0.42, 1.39)		0.63 (0.32, 1.26)
“I occasionally stop taking any medications I am on if I think I will be drinking alcohol” (behavior)			2.06 (0.15)		3.55 (0.06)
False		1.00		1.00
True		1.60 (0.84, 3.02)		2.06 (0.97, 4.36)
Baseline demographics
Sex			2.10 (0.15)
Female		1.00		-
Male		0.64 (0.35, 1.17)		-
Marital status			5.75 (0.02)		5.62 (0.02)
Married		1.00		1.00
Not married		2.04 (1.14, 3.66)		2.04 (1.13, 3.69)
Age (years)			1.12 (0.57)
18–25		1.00		-
26–35		0.86 (0.42, 1.76)		-
>35		0.67 (0.31, 1.43)	-
Education			1.37 (0.24)
Primary education or less		1.00		-
More than a primary education		0.69 (0.38, 1.28)		-
Religion			0.98 (0.61)
Protestant		1.00		-
Catholic		1.32 (0.71, 2.43)		-
Other		0.73 (0.13, 4.27)		-
Household assets			0.59 (0.74)
Low		1.00		-
Middle		0.97 (0.51, 1.82)		-
Rich		0.71 (0.29, 1.76)		-
Household Food Security			0.16 (0.92)
Food secure		1.00		-
Mildly/moderately food insecure		0.90 (0.46, 1.77)		-
Severely food insecure		1.03 (0.48, 2.19)		-
Low social support?			0.40 (0.53)
No		1.00		-
Yes		1.28 (0.59, 2.78)		-
Time to clinic visit (minutes)			4.33 (0.23)
0–20		1.00		-
21–40		0.48 (0.23, 1.04)		-
41–60		0.60 (0.27, 1.32)		-
>60		0.90 (0.42, 1.97)		-
Study arm			0.57 (0.45)
Main cohort arm		1.00		-
Minimally assessed arm		1.31 (0.65, 2.64)		-
At each visit
General health status, past 3 months			1.34 (0.25)
Good/very good/excellent		1.00		-
Fair/poor		1.49 (0.76, 2.94)		-
Affective symptoms of depression, past 3 months?			2.33 (0.13)
No		1.00		-
Yes		2.13 (0.81, 5.63)		-
Unhealthy alcohol use, past 3 months (self-report hazardous or PEth ≥ 50)			5.31 (0.02)		3.72 (0.05)
No		1.00		1.00
Yes		1.92 (1.10, 3.33)		1.75 (0.99, 3.08)
Months on ART at this visit		1.00 (0.93, 1.08)	0.00 (0.98)	-
Number of clinic visits, past 3 months		0.97 (0.77, 1.22)	0.08 (0.78)	-

In multivariable analysis, participants who endorsed the interactive toxicity behavior had increased odds of any ART non-adherence at that visit, compared to those who did not (adjusted OR (AOR): 2.06; 95%CI: 0.97–4.36) (Table 3). The odds of any ART non-adherence were lower for those endorsing the interactive toxicity belief (AOR: 0.63; 95%CI: 0.32–1.26), and higher for participants with unhealthy alcohol use in the past three months (AOR: 1.75; 95%CI: 0.99–3.08) and for unmarried compared to married participants (AOR: 2.04; 95%CI: 1.13–3.69). No other covariates were retained in the final multivariable model.

Discussion

We found endorsement of the interactive toxicity items to be relatively common among persons with HIV who are new to HIV care and ART. Current self-reported drinkers agreed that people should stop taking their ART if they will be drinking at 24% of their study visits. A lower proportion reported skipping ART themselves when they thought they would be drinking (15% of study visits). The proportion believing that alcohol should not be mixed with ART is similar to the percent of alcohol users in a US study, who reported that they believed people should stop their ART while drinking (25%) (16); however, the proportion reporting they stopped taking ART when they were drinking is somewhat lower than in other studies among current drinkers in the US, in which the prevalence of various interactive toxicity behaviors reported ranged from approximately 20–45% (14, 17). The clinical implication is that HIV clinic staff in similar settings should be aware that these beliefs are not rare, and that sometimes patients do stop taking their medications when they are drinking alcohol. Non-nucleoside reverse transcriptase inhibitors (NNRTIs), the most common backbone of antiretrovirals used in SSA, are quite safe with little hepatotoxicity (52–54), and patients should not stop their ART when drinking. It was common for health care providers to stress abstinence from alcohol in the qualitative sub-study of the BREATH Study, and that alcohol and ART don’t mix (18). Participants who had attended the clinic more frequently in the three months prior to the study interview had increased odds of endorsing the interactive toxicity belief at that visit; frequent receipt of this messaging at clinic visits may have led participants to be more likely to retain and believe this idea that alcohol and ART don’t mix. More discussion and advice around decreasing alcohol use (rather than exclusively focusing on abstinence), as well as tools for maintaining high levels of adherence even if drinking, may be warranted and beneficial for patients.

Any ART non-adherence in our study was common using our conservative measure of any non-adherence; non-adherence was reported at 35% of visits. However, the overall number of days missed was low; among visits in which ART was reported missing for at least 1 day in the past 30 days, the median number of days missed was 2 (IQR: 1–3). Reporting the interactive toxicity behavior was associated with an increased odds of any ART non-adherence (Wald X² = 3.55; p-value = 0.06). This suggests that beliefs about toxicity could impact ART adherence.

Similar to other studies (2, 5, 6), unhealthy alcohol use was also associated with ART non-adherence in our multivariable model, although it did not reach statistical significance (Wald X² = 3.72; p-value = 0.054). This independent effect suggests that alcohol use may impact adherence in multiple ways, i.e. via alcohol myopia theory, and via interactive toxicity beliefs. Our unhealthy alcohol use variable consisted of self-report supplemented by a biomarker of recent alcohol use, making it an objective measure of recent unhealthy alcohol use.

Also similar to other studies (7, 8, 55), we observed a strong independent association between marital status and non-adherence; married participants were less likely to report non-adherence. While we did not find an association with our more formal social support measure, spouses may provide support and encouragement for their partner, leading to improved adherence.

This study had some limitations. First, with the exception of unhealthy alcohol use, all the behavioral variables were elicited by self-report. As described earlier, self-reported adherence is thought to be an over-report of true adherence (26–28). Due to this, and based on our previous experience detecting socially desirable under-reporting of alcohol use in this population, our definition of any ART non-adherence in the past 30 days was very conservative. We used this definition in an attempt to minimize the effect of over-reporting adherence; however, patients new to ART are likely to be highly motivated to improve their health, and thus adherence may truly be quite high. In addition, we limited our analyses to only those participants who self-reported recent alcohol use, and thus we may have excluded an important group of participants – those drinkers who do not admit to recent use. Among the visits excluded because participants denied drinking (n = 216 visits), 42% were PEth-positive (PEth ≥ 8 ng/ml), indicative of recent use. When these 90 visits were included in the model of non-adherence, the interactive toxicity results did not change substantially. However, we believed that participants who under-report alcohol use might not honestly report on their interactive toxicity beliefs or their ART adherence. Our finding that interactive toxicity beliefs were reported at a higher number of visits (24%) than those where participants reported themselves ceasing medications while drinking (15%) suggests that these behaviors may have been under-reported. As mentioned earlier, in the qualitative sub-study of the BREATH Study, it was common for participants to describe medical providers telling them that ART and alcohol “don’t mix” (18); as such, participants may have been hesitant to report stopping their ART to consume alcohol themselves. An additional limitation of the study was inclusion of scales not validated for use in our particular study population; however, we felt it was appropriate and necessary to make scales culturally relevant for our study setting. The internal consistency of these scales was high, indicating good scale reliability. Similarly, using single questions to assess the interactive toxicity items, and rate general health status, may have oversimplified these measures. However, other studies have assessed interactive toxicity beliefs as single items (16), and the general health question has been shown to perform as well as multi-item scales (37).

In summary, we found beliefs regarding the interactive toxicity of alcohol use and ART, as well as the prevalence of incomplete ART adherence, to be relatively common among self-reported drinkers who were on ART for at least one month. We found endorsement of the interactive toxicity behavior item, as well as unhealthy alcohol use, to be related to non-adherence. It would be beneficial for patients to receive clear messaging and education about the harms associated with alcohol use, including how to decrease unhealthy levels of drinking, as well as how to maintain adherence even when drinking.