The impact of substance use on adherence to antiretroviral therapy among HIV-infected women in the United States

Abstract

Research is scant regarding differential effects of specific types of recreational drugs use on antiretroviral therapy adherence among women, particularly to single-tablet regimens (STR). This is increasingly important in the context of marijuana legalization. We examined the effects of self-reported substance use on suboptimal (<95%) adherence in the Women’s Interagency HIV Study, 2003–2014. Among 1,799 HIV-infected women, the most prevalent substance used was marijuana. In multivariable Poisson GEE regression, substance use overall was significantly associated with suboptimal adherence (adjusted prevalence ratio, aPR=1.20, 95% CI: 1.10–1.32), adjusting for STR use, socio-demographic, behavioral, and clinical factors. Among STR users, compared to no drug use, substance use overall remained detrimental to ART adherence (aPR=1.61, 95% CI: 1.24–2.09); specifically, both marijuana (aPR=1.48, 95% CI: 1.11–1.97) and other drug use (aPR=1.87, 95% CI: 1.29–2.70) predicted suboptimal adherence. These findings highlight the need to intervene with drug-using women taking antiretroviral therapy to maintain effective adherence.

INTRODUCTION

Adherence to antiretroviral therapy (ART) is essential among patients living with HIV to maintain satisfactory health outcomes [1,2]. Multiple tablet regimens (MTR) require multiple units or dosing times per day [3]. The advent of once-daily single-tablet ART regimens (STR) in 2006 brought the potential for improvements in ART adherence, due not only to the lower pill burden and simplified dosing schedules, but also to the decreased highly active antiretroviral therapy (HAART) toxicity [4–6].

Four STR co-formulations are currently available: Atripla (Efavirenz + Tenofovir (TDF) + Embricitabine (FTC)); Complera (TDF + FTC + Rilpivirine); Stribild (TDF + FTC + Elvitegravir + Cobicistat); and Triumeq (Dolutegravir + Abacavir + Lamivudine) [7]. Studies of the effects of these STRs on treatment adherence have produced inconsistent findings. A multicenter randomized clinical trial in 2006 found that, patients’ treatment satisfaction improved, while treatment adherence and quality of life were maintained after switching to a STR from multiple tablet ART regimens [8]. More recent observational studies have shown that STRs improve medication adherence and other treatment outcomes [4,9,10]. This includes a longitudinal study within the Women’s Interagency HIV Study (WIHS), which found that STR use was significantly associated with increased ART adherence and virologic suppression [4]. Among homeless and marginally housed people, STR use was associated with increased ART adherence, which further explained the association between STR use and viral suppression [10]. However, the potential therapeutic benefits of STR among active substance users remain unclear. Previous studies have consistently found that active drug use has detrimental impact on the health seeking behaviors among HIV-infected patients, specifically ART adherence [11–16]. In a multicenter longitudinal cohort study (ACTG 362), the use of hard drugs, defined as cocaine, amphetamines, or heroin was significantly associated with increased odds of non-adherence to ART [12]. Another cohort study of 150 HIV-seropositive adults found that active drug users had significantly worse ART adherence than non-drug users, and individuals using stimulants were at greatest risk for suboptimal adherence [13]. Specifically, among injection drug users, the use of cocaine and multiple substances were significantly associated with suboptimal adherence [10]. Cognitive impairment, inconsistent lifestyles and other acute effects of intoxication have been posited to explain the suboptimal adherence among substance users [13,17,18]. Other behavioral factors such as cigarette smoking [19,20], and alcohol use [21,22], as well as demographic and clinical profiles including race/ethnicity [19,23], depression [21,24], CD4+ cell count [19,21], and HIV RNA viral load [21] have been found to be associated with ART adherence level as well. The generalizability of these findings to the context of STR regimens has not been elucidated. As of May 2017, twenty-six states and the District of Columbia have legalized marijuana, either for medical or recreational use [25]. The most commonly reported reasons among HIV patients for marijuana use were relaxation, appetite stimulation, for social situation, and for reductions of HIV symptoms [26,27]. Despite increasing legal access to marijuana, the impact of marijuana use on medication adherence has not been well characterized. A previous WIHS study from 1994 to 2010 found that current marijuana use was significantly associated with 40% lower odds of ART adherence among HIV-infected women [27]. In another longitudinal prospective cohort of HIV-infected drug users in Canada, there was no association between daily marijuana use and suboptimal ART adherence [28]. Studies have shown that marijuana and other drugs are differentially associated with brain functioning and decision making performance [29]. It remains unclear whether these differences will attribute to a disproportionate prevalence of suboptimal ART adherence, by marijuana and other specific drugs.

There has been a gap in the HIV care continuum for women living with HIV; specifically, women are more vulnerable to suboptimal ART adherence than men, which may be explained by different psychological and behavioral profiles [30–32]. The objective of this study was to examine the overall and drug-specific impact of substance use on ART adherence in a cohort of HIV-infected women using single tablet and multiple tablet ART regimens. We hypothesized that substance use overall would negatively impact ART adherence in this cohort of women, and further examined whether this effect differed by specific drug used. Moreover, we investigated whether any association between drug use and ART adherence would differ by regimen (i.e., STR versus MTR), adopting a 2-tailed hypothesis about the relationship between STR and adherence.

METHODS

Study design

Data were collected semiannually from participants in the Women’s Interagency HIV Study between April 1, 2003 and March 31, 2014. The WIHS is the largest ongoing prospective cohort study of HIV among women in the United States. Participants for this analysis were enrolled during three recruitment waves: 1994–95, 2001–02, 2011–12 from six clinical consortia: Brooklyn NY, Bronx NY, Chicago IL, Los Angeles CA, San Francisco CA, and Washington DC (Supplementary Figure 1). The WIHS collects data in semiannual study visits, which includes standardized interviews, specimen collection, and clinical examination. Details of the WIHS have been reported previously [33,34]. In this analysis, the baseline visit was considered as each participant’s initial visit during the study period from 2003 to 2014.

Study Population

In this analysis, we included HIV-infected women who had at least 2 consecutive follow-up visits and reported having been prescribed antiretroviral therapy at any study visit, and who provided information on substance use at their first visit between April 1, 2003 and March 31, 2014. Women who did not have at least 2 consecutive follow-up visits during the study period were excluded.

Substance Use

Women reported substance use over the past 6 months at WIHS study visit. We considered this time-varying exposure in multiple ways. For our first analysis, a binary variable for any substance use was defined as reporting either injection or non-injection of any of the following drugs: marijuana, cocaine (including both crack and powder forms), heroin, methamphetamines (including crank, crystal, tina), club drugs (e.g., ecstasy, ketamine, Gamma Hydroxybutyrate), or non-prescribed prescription drugs (e.g., amphetamines, narcotics, hallucinogens). For our second analysis, we examined the exposure in terms of specific drug used. In order to ensure sufficient sample size within each exposure category, we assessed the type of specific drug categorized as: no substance use, marijuana only, or use of other drugs.

Adherence to Antiretroviral Therapy

Participants self-reported use of ART by indicating the proportion of time ART was taken as prescribed over the previous 6 months, using the following categorical responses: <75% of the time, 75–94%, 95–99%, or 100% of the time. We dichotomized participants’ responses, defining suboptimal adherence as <95%, to be consistent with the literature [1,2,4,28,35]. The construct validity for this outcome definition is supported by a WIHS study by Wilson et al., which found a significant association between suboptimal adherence and subsequent detectable HIV viral load [35].

Covariates

Variables hypothesized a priori to be associated with substance use or with suboptimal ART adherence were examined in the analysis. Continuous calendar time was used as the time scale to adjust for any potential secular trends.

Time-invariant covariates

Fixed covariates assessed at baseline included: race/ethnicity (non-Hispanic Black, non-Hispanic White/other, and Hispanic), education ( high school and >high school), and enrollment cohort (1994–1995, 2001–2002, and 2011–2012).

Time-varying covariates

Before the introduction of STR in 2006, WIHS participants reported their MTR use. Since 2006, WIHS participants have also been asked about their STR use; a binary covariate was constructed to indicate STR use over the past 6 months. All WIHS participants on STR were using one of the three available single tablet formulations during the study period: Atripla, Complera, or Stribild. Symptoms of depression were assessed semiannually using the Center for Epidemiologic Studies Depression scale (CES-D) [36]. Because of the known correlation with somatic symptoms and substance use, CES-D scores ≥23 have been used to indicate depressive symptoms among drug-using populations [37–39]. We therefore used this more stringent cutoff in this analysis. CD4+ cell count was measured semiannually and was categorized as follows: <350, 350–499, and ≥500 cells/mL. The lower limit of quantification (LLQ) for HIV viral load changed from 80 copies/mL in 2003 to 20 copies/mL in 2014. For our analyses, a binary covariate was generated to indicate whether HIV RNA was detectable at each study visit, using LLQ cutoff that was imposed at that time. Other time-varying covariates included: employment status, annual income (≤ $12,000 and >$12,000), alcohol use, and current smoking status. Alcohol use was categorized as abstainer (0 drink/week), moderate (1–7 drinks/week) and heavy (>7 drinks/week) alcohol use [40].

Statistical Analyses

Descriptive statistics were generated to summarize the data. Bivariate associations between substance use and key covariates were evaluated using Chi-square tests for categorical variables and Wilcoxon rank-sum (Mann-Whitney) tests for continuous variables.

Bivariate and multivariable Poisson regression models were constructed, applying generalized estimating equations (GEE) methods with a log link function and a working exchangeable correlation structure, to estimate the relative prevalence over time of suboptimal adherence among those with no reported substance use compared to (a) any substance use and to (b) type of specific drug used. For each analysis, we also analyzed STR and MTR users separately, and tested the interaction between STR use and exposures on suboptimal adherence. Logistic regression models with GEE methods were initially constructed to estimate odds ratios over time, but the model failed to converge to provider parameter estimates. Poisson regression with a robust variance estimator and GEE methods was then applied to estimate prevalence ratios [41]. In each multivariable model, we included covariates found to be significantly (p<0.25) associated with both suboptimal adherence and substance use in bivariate analysis. Variables hypothesized a priori to be potential confounders included age, race, CD4+ cell count, and detectable viral load; these were included in the multivariable models regardless of the bivariate results. GEE methods were chosen to estimate population means and to account for correlations in measurements from repeated observations for each participant [42,43]. The within-subject correlation structure was selected based on the QIC statistics [44,45]. A robust variance estimator was used to obtain a valid standard error of estimates through incorporating both information about selected working within-subject correlation structures and empirical information from the data.

In both bivariate and multivariable analyses, suboptimal adherence was paired with exposure data from the visit immediately prior (Supplementary Figure 2), and observations only from these consecutive visit pairs were included in the analysis. We hypothesized an acute effect of substance use on ART adherence; therefore this analytic design enabled us not only to ensure temporality, but also to assess a more immediate effect of exposure on outcome. In the case of a missed visit, observations from the visit immediately prior were excluded from the analysis. For women who switched ART regimen type (STR versus MTR), observations from the visit immediately prior to the switch were excluded. For women who did not have at least two consecutive follow-up visits, none of the observations from the participant were included in the analysis and therefore these women were excluded from the analysis. Statistical analyses were performed using Stata version 12 software [46].

RESULTS

Study Population Characteristics

Of 1,911 women who met other eligibility criteria, 112 (6%) were excluded from the analysis because they did not have at least 2 consecutive follow-up visits during the study period. A total of 1,799 ART-treated women included in this analysis contributed 23,787 person-visits between April 2003 and March 2014. At the baseline visit, median age was 42 years (interquartile range, IQR: 36–48), 57% were Black or African-American, and 25% had depressive symptoms. Substance use was reported by 396 (22%) women (Table 1). Those identify as White, non-Hispanic as compared to Black or African-American (30% vs 24%, p<0.001), and to be unemployed (27% vs 12%, p<0.001), current smokers (37% vs 11%, p<0.001), and have depressive symptoms (32% vs 18%, p<0.001), and CD4+ cell count <350 as compared to ≥500 (25% vs 18%, p=0.02; Table 1) were more likely to be drug users. Women who reported moderate (28% vs 13%) or heavy (62% vs 13%) alcohol use were also significantly (p<0.001) more likely to be drug users than women who did not drink (Table 1). At baseline, active drug users most frequently used marijuana (77%) or crack (30%). There were 58% of drug users who reported using marijuana only, and 19% reported using marijuana in addition to other drugs. During the study period, participants reported substance use at 4,036 person-visits (17% of 23,787 person-visits), suboptimal ART adherence at 4,869 (20%) person-visits, and single tablet ART regimen use at 3,525 (15%) person-visits. Participants reported changing ART regimens at 570 (2%) person-visits. The prevalence of suboptimal adherence decreased significantly (p<0.001) over the study period, from 22% in 2003 to 18% in 2014 (Figure 1). The prevalence of substance use in 2013 was 20%. Despite some fluctuations over time, by 2014 the prevalence was 23% (p=0.27). The prevalence of STR use increased significantly (p<0.001), from 2% when it first became available in 2006 to 32% by 2014.

Table 1.

Baseline characteristics of HIV-infected WIHS participants, by substance use, 2003–2014

Characteristic Med (IQR) or N (row %)	Total (N=1,799)	Substance usea (N=396)	No substance use (N=1,403)	p-valueb
Age at baseline visit (years)	42 (36–48)	43 (37–49)	41 (35–47)	0.001
Race/Ethnicity
Black, non-Hispanic	1020	244 (24)	776 (76)	<0.001
White, non-Hispanic/Other c	304	92 (30)	212 (70)
Hispanic	475	60 (13)	415 (87)
Education
≤ High school	1202	282 (23)	920 (77)	0.04
> High school	595	114 (19)	481 (81)
Employment status
Unemployed	1196	324 (27)	872 (73)	<0.001
Employed	599	69 (12)	530 (88)
Annual income
≤$12,000	931	247 (27)	684 (73)	<0.001
> $12,000	756	124 (16)	632 (84)
Depressive symptoms
No	1330	243 (18)	1087 (82)	<0.001
Yes	446	144 (32)	302 (68)
Alcohol use (drinks/week)a
Abstainer (0)	984	125 (13)	859 (87)	<0.001
Moderate (1–7)	683	189 (28)	494 (72)
Heavy (>7)	132	82 (62)	50 (38)
Currently smoking
No	1015	108 (11)	907 (89)	<0.001
Yes	783	288 (37)	495 (63)
CD4+ cell count
<350	689	170 (25)	519 (75)	0.02
350–499	392	89 (23)	303 (77)
≥500	687	127 (18)	560 (82)
Detectable HIV viral load
No	978	204 (21)	774 (79)	0.20
Yes	799	187 (23)	612 (77)
Study site
Bronx, NY	319	72 (23)	247 (77)	<0.001
Brooklyn, NY	319	47 (15)	272 (85)
Washington, DC	261	39 (15)	222 (85)
Los Angeles, CA	342	46 (13)	296 (87)
San Francisco, CA	279	126 (45)	153 (55)
Chicago, IL	279	66 (22)	213 (78)
Enrollment cohort
1994–1995	992	226 (23)	766 (77)	<0.001
2001–2002	586	97 (17)	489 (83)
2011–2012	221	73 (33)	148 (67)
Specific drug typea, N (col %)
Marijuana only	–	231 (58)	–	N/A
Crack, no marijuana		71 (18)
Crack and marijuana		45 (11)
Other		49 (12)

IQR: interquartile range

^aIn the 6 months prior to the study visit

^bChi-square test for categorical variables and Wilcoxon rank-sum (Mann-Whitney) test for continuous variables

^cIncluded 24 Asian/Pacific Islander, 9 native American/Alaska, and 31 other race.

Figure 1.

Trends in substance use, suboptimal ART adherence, and single tablet ART regimen use among ART users, WIHS, 2003–2014

Adherence by Substance Use

Throughout the study period, the prevalence of suboptimal adherence to ART among drug users was consistently higher than that among non-drug users (Figure 2). This observed difference between drug users to non-drug users appeared consistent among STR users and MTR users separately (Figure 3). As compared to no use, active substance use was associated with a 20% higher prevalence of suboptimal adherence (adjusted Prevalence Ratio, aPR=1.20, 95% CI: 1.10–1.32; Table 2). Other significant predictors included younger age, Black or African-American race, having depressives symptoms, moderate and heavy alcohol use, current smoking, no single tablet ART regimen use, lower CD4+ cell count, detectable HIV viral load, and being in the original 1994–1995 enrollment cohort (Table 2). This association between substance use and suboptimal adherence appeared similar when restricted to the MRA users only (aPR=1.17, 95% CI: 1.06–1.29; Table 3). When restricting to STR users only, substance use was associated with 61% higher prevalence of suboptimal adherence, after adjusting for calendar year, age, race, employment, depressive symptoms, CD4+ cell count, detectable HIV viral load, and enrollment cohort (aPR=1.61, 95% CI: 1.24–2.09; Table 3). To examine if this impact of substance use on suboptimal ART adherence differed by STR use, an interaction between substance use and STR use was explored, but was non-significant (p_interaction, p_i=0.47).

Figure 2.

Trends in suboptimal ART adherence by any substance use (used in the first analysis) and by specific drug type (used in the second analysis), WIHS, 2003–2014.

Categories by any substance use included (a) any substance use, and (b) no substance use. Categories by specific drug type included (a) marijuana use only, (b) other drug use, and (c) no substance use. Categories as any substance use, marijuana use only, and other drug use are not mutually exclusive.

Figure 3.

Trends in suboptimal ART adherence by any substance use and by ART regimen type (single tablet regimen, STR; multiple tablet regimen, MTR), WIHS, 2006–2014

Table 2.

Bivariate and multivariable associations between substance use with suboptimal ART adherence, WIHS, 2003–2014

Covariate	Unadjusted PR (95% CI)	Adjusted PR a,b (95% CI)
Substance usec	1.32 (1.21–1.45)	1.20 (1.10–1.32)
Specific drug typec
None	1	1
Marijuana use only	1.10 (0.98–1.24)	1.03 (0.93–1.15)
Other	1.64 (1.46–1.84)	1.46 (1.30–1.65)
Calendar year	0.97 (0.96–0.98)	0.99 (0.98–1.00)
Age at baseline visit	0.98 (0.98–0.99)	0.98 (0.97–0.98)
Race/Ethnicity
Black, non-Hispanic	1	1
White, non-Hispanic/Other	0.78 (0.66–0.93)	0.80 (0.68–0.93)
Hispanic	0.73 (0.63–0.83)	0.71 (0.63–0.81)
Education > high school	0.98 (0.86–1.10)	–
Currently employed	1.00 (0.93–1.08)	–
Annual income > $12,000	0.97 (0.90–1.03)	–
Depressive symptoms	1.16 (1.08–1.25)	1.11 (1.03–1.20)
Alcohol use (drinks/week)c
Abstainer (0)	1	1
Moderate (1–7)	1.20 (1.12–1.29)	1.14 (1.06–1.23)
Heavy (>7)	1.37 (1.20–1.55)	1.26 (1.12–1.42)
Current smoking	1.30 (1.19–1.42)	1.18 (1.09–1.29)
Single tablet ART regimen usec	0.64 (0.56–0.73)	0.69 (0.60–0.79)
CD4+ cell count
<350	1	1
350–499	0.79 (0.73–0.86)	0.86 (0.79–0.93)
≥500	0.72 (0.65–0.78)	0.80 (0.73–0.87)
Detectable HIV viral load	1.36 (1.28–1.45)	1.27 (1.19–1.36)

PR, prevalence ratio; CI, confidence interval

^aEstimates were adjusted for enrollment cohort.

^bAdjusted estimates for the exposure were shown by (a) any substance use and (b) specific drug type. Adjusted estimates for all other covariates in the table were derived from a model that included the dichotomous exposure variable.

^cIn the 6 months prior to the study visit

Table 3.

Multivariable associations between substance use with suboptimal ART adherence, among multiple and single tablet users, WIHS, 2003–2014

	Adjusted PR (95% CI)a
Covariate	Multiple tablet users	Single tablet users
Substance useb	1.17 (1.06–1.29)	1.61 (1.24–2.09)
Race/Ethnicity
Black, non-Hispanic	1	1
White, non-Hispanic/Other	0.80 (0.68–0.94)	0.51 (0.28–0.91)
Hispanic	0.73 (0.64–0.84)	0.55 (0.37–0.83)
Depressive symptoms	1.09 (1.01–1.18)	1.53 (1.23–1.91)
Currently employed	–	0.97 (0.74–1.27)
Annual income > $12,000	0.99 (0.92–1.07)	–
Alcohol use (drinks/week)b
Abstainer (0)	1	–
Moderate (1–7)	1.17 (1.09–1.27)
Heavy (>7)	1.29 (1.14–1.46)
Current smoking	1.22 (1.12–1.34)	–
CD4+ cell count
<350	1	1
350–499	0.87 (0.80–0.95)	0.73 (0.55–0.97)
≥500	0.79 (0.72–0.87)	0.75 (0.56–1.01)
Detectable HIV viral load	1.28 (1.19–1.37)	1.22 (0.98–1.53)

PR, prevalence ratio; CI, confidence interval

^aEstimates were adjusted for age at baseline visit, calendar year and enrollment cohort.

^bIn the 6 months prior to the study visit

Adherence by Specific Drug Type

As access to legal marijuana expanded during the study period, we explored the trend and distribution of marijuana use and other specific drugs over time. The prevalence of marijuana use alone fluctuated from 11% in 2003 to 14% in 2014 (Figure 4). The prevalence of crack use and other specific drugs remained relatively stable, and were consistently lower than 8% from 2003 to 2014 (Figure 4). Given that marijuana was the most prevalent substance used (over 61% of total person-visits among active drug users), we analyzed the substance use type as marijuana use only and other drugs as compared to none.

Figure 4.

Trends in substance use by specific drug type among ART users, WIHS, 2003–2014

When we examined the frequency of specific drugs used, we found collinearity between drug type and drug use frequency. Women who used marijuana only were more likely to be daily users (47%; Table 4) than women who used all other types of drugs (26%; p<0.001). Therefore we did not include drug use frequency in our statistical models, in which drug type was our main exposure of interest.

Table 4.

Frequency of use by specific drug type among substance users, WIHS, 2003–2014

Specific drug typea,b	Less than weekly	Weekly, less than daily	At least daily	p-valuec
Marijuana use only (N=2,482)	616 (25%)	695 (28%)	1,171 (47%)	<0.001
Crack, no marijuana use (N=575)	211 (37%)	219 (38%)	145 (25%)
Crack and marijuana use (N=441)	175 (40%)	169 (38%)	97 (22%)
Other (N=519)	217 (42%)	148 (29%)	154 (30%)

^aDrug use in previous six months

^bData shown are N (row %) person-visits.

^cChi-square test

The prevalence of suboptimal ART adherence was consistently higher among marijuana users than non-drug users; women using other drugs had worst adherence throughout the study period (Figure 2). After adjusting for calendar year, age, race, depressive symptoms, alcohol use and current smoking, STR, CD4+ cell count, detectable HIV viral load, and enrollment cohort, there was no significant association between marijuana use and suboptimal adherence (aPR=1.03, 95% CI: 0.93–1.15; Table 2). The use of other drugs was significantly associated with 46% higher prevalence of suboptimal adherence (aPR=1.46, 95% CI: 1.30–1.65), as compared to no drug use (Table 2). When restricted to MTR users, results remained similar to that from overall study population (Table 5). On the contrary, when restricted to STR users, women who used marijuana only (aPR=1.48, 95% CI: 1.11–1.97; Table 5) had significantly higher prevalence of suboptimal ART adherence than non-drug users. Meanwhile, the use of other drugs was significantly associated with 87% higher prevalence of suboptimal adherence (aPR=1.87, 95% CI: 1.29–2.70; Table 5). To examine if these impacts of drug-specific substance use on suboptimal ART adherence differed by STR use, interactions between specific drug types (p_{i marijuana vs none}=0.35, p_{i others vs none}=0.86) and STR use were explored, but were both non-significant. In the subanalysis restricted to marijuana users, we did not observe significant differences in the prevalence of suboptimal ART adherence by the frequency of drug use (results not shown).

Table 5.

Multivariable associations between specific drug type with suboptimal ART adherence, among multiple and single tablet users, WIHS, 2003–2014

	Adjusted PR (95% CI)a
Covariate	Multiple tablet users	Single tablet users
Specific drug typeb
None	1	1
Marijuana use only	1.02 (0.91–1.14)	1.48 (1.11–1.97)
Other	1.40 (1.23–1.60)	1.87 (1.29–2.70)
Race/Ethnicity
Black, non-Hispanic	1	1
White, non-Hispanic/Other	0.80 (0.68–0.94)	0.49 (0.27–0.89)
Hispanic	0.73 (0.64–0.83)	0.55 (0.37–0.83)
Depressive symptoms	1.08 (1.00–1.17)	1.52 (1.22–1.89)
Currently employed	–	0.97 (0.74–1.27)
Annual income > $12,000	0.99 (0.92–1.08)	–
Alcohol use (drinks/week)b
Abstainer (0)	1	–
Moderate (1–7)	1.17 (1.09–1.27)
Heavy (>7)	1.26 (1.11–1.42)
Current smoking	1.22 (1.11–1.33)	–
CD4+ cell count
<350	1	1
350–499	0.87 (0.80–0.95)	0.74 (0.55–0.98)
≥500	0.79 (0.72–0.87)	0.76 (0.57–1.02)
Detectable HIV viral load	1.28 (1.19–1.37)	1.22 (0.98–1.53)

PR, prevalence ratio; CI, confidence interval

^aEstimates were adjusted for age at baseline visit, calendar year, and enrollment cohort.

^bIn the 6 months prior to the study visit

DISCUSSION

In this cohort of HIV-infected women, substance use other than marijuana only was associated with suboptimal adherence to ART, among both STR and MTR users. Marijuana use was significantly associated with suboptimal ART adherence only among STR users. Other specific drugs were significantly associated with suboptimal ART adherence among both STR and MTR users.

The prevalence of single tablet ART regimen use in this study population was slightly lower than the concurrent prevalence (24%) among Medicaid enrollees in the U.S, 2006–2009 [9]. Meanwhile, higher ART adherence with STR use in our study is consistent with the recent evidence suggesting that simplified ART regimens improve adherence [4–6,9,47]. Factors such as perceived ease of the regimen, patient preference, reported quality of life, and HIV symptoms might explain the observed difference in ART adherence between STR users and MTR users [8,48].

The overall detrimental effect of substance use observed in this study is consistent with findings from previous studies [11–13,15,16,18,49–52]. In our study, other behavioral factors including moderate and heavy alcohol use, as well as current smoking, were significantly associated with suboptimal ART adherence, consistent with prior research findings [19–22]. Independent of these behavioral factors, we observed a significant association between substance use and suboptimal ART adherence. Importantly, our study is the first to demonstrate that substance use is detrimental to ART adherence after adjusting for STR use. When we assessed potential effect modification, we did not observe a significant interaction between substance use and STR use on ART adherence. According to the current ART use guidelines, providers may prioritize STR to drug users who are deemed to have low likelihood of ART adherence [53]. Our data suggest that, in practice, substance use may continue to have a strong detrimental impact on women’s health seeking behavior, whether they are prescribed STR or MTR regimens.

We found that, among STR users, marijuana use only was significantly associated with 48% higher prevalence of suboptimal adherence as compared with no drug use. However, such detrimental effect was not observed among MTR users. The existing evidence regarding the longitudinal impact of marijuana on medication adherence is limited and inconsistent. In a previous WIHS study, current marijuana use was found to be associated with 40% lower odds of ART adherence, although this observed association was not accounted for other specific drug use [27]. In another longitudinal cohort study, daily marijuana use was not associated with suboptimal ART adherence [28]. These differences might be explained by different exposure definitions, e.g., classification by diagnostic criteria (e.g., dependence). More specifically, another study found that marijuana dependent individuals have significantly lower ART adherence than non-dependent marijuana users [54]. Of the existing studies of the association between marijuana use and ART adherence, few have compared whether this association varies by the purpose of marijuana use. A study of 252 HIV-infected patients found that marijuana use was beneficial for ART adherence among those with moderate to severe nausea, while no significant association was observed among those with mild or no nausea. In the current study, we were unable to differentiate between medical and recreational marijuana use. With increasing legal access to marijuana, special attention should be given to the therapeutic value of marijuana use, and how that may be offset by potential adverse effects on ART adherence. Empirical data are needed in the context of health outcomes and health seeking behaviors in order to elucidate these relationships.

Interpretation of these findings warrants consideration of several limitations. Although it has been suggested that self-reported adherence is consistently higher than estimates obtained via other approaches (e.g., electronic monitoring devices), the validity of self-reported adherence data has been widely established [55–58]. There was a limited sample size when we restricted to the STR users only. In addition, we did not have the data to evaluate whether the observed association was mediated or confounded by participation in drug treatment programs. We were also unable to determine the temporality between exposure and outcome data because of the semiannual data collection structure, and therefore we conducted the analysis on paired consecutive visits to ensure outcome data were collected after the exposure data (Supplementary Figure 2). Consistent with other published papers, WIHS as an interval cohort, has data from the structured semiannual visits, and it is not feasible to have more frequent data collection [4,21,27]. Finally, less than 95% adherence as a cutoff defining suboptimal adherence was based on older studies of unboosted protease inhibitor therapy [2]. It might be a conservative cutoff in the current ART era, as adherence levels as low as 80%–84% may be sufficient for newer regimens to achieve viral load suppression [59]. Our study period began in 2003, thus a conservative cutoff is preferred with the assumption of non-differential measurement error of adherence, resulting in a conservative study finding.

Despite these limitations, there are several strengths of our study. Our detailed individual-level data on drug type and ART regimen type enabled us to investigate differential effects of substance use on adherence among US women taking STR and MTR. Drug use behavior differs between men and women, hence our findings extend the existing knowledge and inference to female drug users, a population with limited resources available [60]. Moreover, a study population of 1,799 over an 11-year study period provides robust evidence for possible causal interpretation of study findings.

CONCLUSIONS

Current guidelines for the use of ART among drug users recommend regimens with simple dosing schedules in order to enhance medication adherence [53]. Drug users, however, might continue to face adherence challenges when prescribed single tablet regimens as compared to multiple tablet regimens. Our study findings thus highlight the continuing need to monitor active drug use among HIV-infected patients, particularly with increasing access to legal marijuana, to maintain effective ART adherence. Future research should focus on recreational and medical marijuana use, among HIV-infected individuals, particularly those prescribed single tablet ART regimens.

Supplementary Material

10461_2017_1808_MOESM1_ESM

Supplementary Figure 1. Flowchart of the study visits, WIHS, 2003-2014

Semiannual study visits from 2003 to 2014 in the dashed box were included in the analysis. Calendar year was listed as last two digits (i.e. 1994: 94).

10461_2017_1808_MOESM2_ESM

Supplementary Figure 2. Study visit of exposure, covariate, and outcome data included in the analysis

Exposure and covariate data (i.e., from visit X) were paired with outcome data (i.e., from visit X+1 in the analysis.