Periods of Homelessness Linked to Higher VACS Index Among HIV-Positive People Who Use Drugs

Abstract

We sought to evaluate the impact of homelessness on HIV disease progression among people who use unregulated drugs (PWUD) living with HIV and test if this association was mediated by adherence to antiretroviral therapy (ART). We applied general linear mixed-effects modeling to estimate the longitudinal relationship between homelessness and the Veterans Aging Cohort Study (VACS) Index, a validated measure of HIV disease progression that predicts all-cause mortality, among a prospective cohort of PWUD. In a longitudinal model adjusted for ART adherence, homelessness was significantly associated with increased VACS Index scores and 16% of the association was mediated by ART adherence. These findings indicate that homelessness was a significant risk factor for HIV disease progression and this association was marginally mediated by ART adherence. Future studies are needed to quantify the other mechanisms (e.g., food insecurity, mental health) by which homelessness increases mortality risk among PWUD living with HIV.

Introduction

Advancements in combination antiretroviral therapy (ART) have dramatically reduced HIV-associated morbidity and mortality worldwide [1–4]. Improvements in ART potency, tolerability, and convenience have contributed to declines in rates of HIV transmission and the development of AIDS as well as increasing life expectancy among members of many key affected populations [5–7]. Life expectancy among people living with HIV engaged in ART has improved over time, and these improvements are most significant among high-income countries [1, 2]. In spite of these improvements, HIV-associated pathologies (e.g., infections) and AIDS-related deaths remain common among many people living with HIV [4]. People living with HIV also have higher rates of non-infectious chronic diseases due to increased inflammation and immune activation that persists despite engagement in effective ART [8, 9]. Among people living with HIV, biomarkers of inflammation and immune activation are strongly associated with morbidity and mortality [10, 11].

In response to this improved understanding of the sources of HIV-related pathogenesis, the Veterans Aging Cohort Study (VACS) Index was developed as a predictor of all-cause mortality among people living with HIV [12–16]. The VACS Index includes measures of HIV mortality risk (age, CD4 cell count, HIV-1 RNA plasma viral load) and indicators of comorbidity and organ system dysfunction (hemoglobin, platelet count, creatinine levels, aspartate aminotransferase [AST], alanine aminotransferase [ALT]) and hepatitis C virus (HCV) serostatus [17]. The VACS Index is highly correlated with measures of systemic inflammation, provides a more comprehensive assessment of physiologic injury among people living with HIV on ART and is a more discriminative predictor of all-cause mortality than the traditional Restricted Index (i.e., age, CD4 cell count and HIV viral load) [13, 18]. Since the component measures of the VACS Index are routinely collected as a part of regular clinical care, in accordance with current HIV treatment guidelines, this prognostic tool is available to most clinical centers. The VACS Index has been validated in several large cohort studies of people living with HIV, including people who use drugs, and is highly responsive to ART initiation, adherence and interruption [13–15, 19].

A key affected population who continue to experience elevated HIV-related morbidity and mortality are people who use drugs (PWUD). Although the effectiveness of ART among HIV-positive PWUD has been demonstrated at the individual and community level [20, 21], many PWUD living with HIV continue to experience barriers to optimal ART treatment resulting in suboptimal virological outcomes and elevated rates of viral transmission [5, 22–24]. Among PWUD, sub-optimal engagement in HIV treatment and care has been linked to both individual-level factors (e.g., high-intensity substance use and comorbid mental health concerns) as well as contextual factors such as homelessness [2, 25–28]. Point prevalence estimates of homelessness among PWUD range from 30 to 40%, and homelessness has been independently associated with a reduced likelihood of ART initiation, decreased ART adherence and lower rates of HIV viral suppression [29, 30]. Homelessness has also been identified as an independent risk factor for mortality among PWUD [31]. Age-adjusted mortality rates among people experiencing homelessness are as much as nine times that of the general population and, among PWUD, drug overdose and AIDS have been identified as the leading causes of death [32–34].

Despite the HIV disease-related risks associated with homelessness, we are unaware of any study that has specifically examined the association between homelessness and HIV disease progression using the VACS Index among PWUD. Evaluating the VACS Index among homeless PWUD may provide a useful summary measure of HIV disease progression and mortality risk that could be applied in research and clinical practice. People who are homeless also experience high rates of psychiatric and other medical comorbidities (e.g., cirrhosis and cardiovascular disease) that may impede HIV care and exacerbate HIV disease progression [35–37]. In light of these knowledge gaps and persistent inequities in life expectancy among people living with HIV who use drugs, we sought to: (1) evaluate the impact of homelessness on the VACS Index among a prospective cohort of people living with HIV who use unregulated drugs; (2) examine if the association between homelessness and the VACS Index was mediated by adherence to ART.

Methods

Study Participants

The data for this investigation were collected from the AIDS Care Cohort to Evaluate exposure to Survival Services (ACCESS) [38]. ACCESS is an open and ongoing community-recruited prospective cohort of people living with HIV and who use drugs. Participants have been recruited through community-based methods, including street out-reach and self-referral, since 1996 in the Downtown Eastside neighborhood of Vancouver, Canada, the site of an explosive outbreak of HIV infection [39]. Individuals were eligible for ACCESS if they were: aged ≥ 18 years, HIV seropositive, report using unregulated drugs other than or in addition to cannabis, resided in the Greater Vancouver Regional District and provided informed consent at study enrolment.

At baseline and every 6 months thereafter, participants completed an interviewer-administered questionnaire that elicited socio-demographic information, substance use patterns, engagement with health and social services, and other related exposures. Every 6 months, participants provided blood samples for HCV serological analysis and HIV clinical monitoring (e.g., CD4 cell counts, HIV viral load). At each study visit participants received $40 CAD as remuneration for their time. The University of British Columbia/Providence Health Care Research Ethics Board approved the ACCESS study (H05-50233).

Measures

Canada’s universal health care coverage provides no-cost health care to all individuals. In the province of British Columbia, ART and HIV clinical monitoring are provided free of charge to all people living with HIV, delivered through the Drug Treatment Program of the British Columbia Centre for Excellence in HIV/AIDS, the province-wide ART dispensary and clinical monitoring registry. A confidential data linkage with this program was used to access comprehensive retrospective and prospective data on HIV treatment measures (e.g., HIV-1 RNA plasma viral loads, CD4+ cell counts), records of ART dispensation, as well as the laboratory measures necessary to calculate the VACS Index for ACCESS participants (e.g., hemoglobin, platelets, AST and ALT, creatinine, and HCV) [40]. Blood samples drawn to analyze these measures were collected through the ACCESS study or as a part of regular clinical care.

The analytical sample for this analysis included all ACCESS participants receiving ART who had at least one CD4+ cell count measurement and at least one HIV-1 viral load measurement within 180 days of their first HIV-positive study interview. All ACCESS participants who met these criteria between December 2005 and November 2017 were included in this analysis. The primary outcome of interest was the VACS Index. The VACS Index combines indicators of HIV-specific mortality (age, CD4 cell count, HIV-1 RNA plasma viral load), with composite measures of liver and renal injury (FIB-4, estimated glomerular filtration rate (eGFR)) and HCV. FIB-4 is based on age alanine aminotransferase, aspartate aminotransferase, and platelet count, and eGFR is based on creatinine, age, sex and ethnicity. VACS Index scores were calculated in accordance with previous studies [15]. Possible scores could range from 0 to 164 with higher scores indicating an increased risk of mortality [13]. Previous studies have shown that each five-unit increase in the VACS Index score is associated with a 20% increase in five-year all-cause mortality risk [14]. Resources to calculate the VACS Index for clinical and research purposes are publicly available online (https://medicine.yale.edu/intmed/vacs/cohorts/vacsresources/vacsindexinfo/, www.vacohort.org). A single imputation regression approach based on multilayer perception was used to impute missing values for the components of the VACS Index [41]. This method applies artificial neural networks to model and predict missing values for monotonic patterns of missing data and has been used in previous analyses of the VACS Index [19, 41].

The primary explanatory variable of interest was homelessness. As in previous studies, this variable was defined as living on the street with no fixed address at any time in the 6-month period preceding the follow-up interview [42]. We also included a range of socio-demographic, substance use, structural and clinical factors in the analysis hypothesized to confound the association between homelessness and the VACS Index. These exposures included: sex at birth (male vs. female); age (per year older); ethnicity (white vs. non-white); employment (having a regular, temporary, or self-employed work vs. none); cocaine use (≥ daily vs. < daily); crack cocaine use (≥ daily vs. < daily); heroin use (≥ daily vs. < daily); crystal methamphetamine use (≥ daily vs. < daily); non-medical prescription opioid use (≥ daily vs. < daily); alcohol use (≥ daily vs. < daily); cannabis use (≥ daily vs. < daily); injection drug use (yes vs. no); mental illness, defined as being diagnosed with a mental illness in the past 6 months (yes vs. no); receipt of opioid agonist therapy (OAT, yes vs. no); engagement with alcohol or drug treatment programs (other than OAT) (yes vs. no); incarceration (yes vs. no); community supervision (probation, parole, bail, conditional sentence or diversion vs. none); and time since baseline visit (per year longer). All behavioral variables refer to the 6-month period preceding the most recent follow-up visit and variable definitions are consistent with previous studies [38, 43].

We also included as an explanatory variable a measure of adherence to prescribed antiretroviral therapy, defined as the number of days in the previous 180 days in which ART had been dispensed and picked up, whether in community, clinical or carceral settings divided by the number of days since ART initiation, capped at 180 days. We dichotomized this variable at 95%, consistent with previous studies. This pharmacy-derived measure of adherence has been shown to predict HIV viral suppression and survival [40, 44].

Statistical Analyses

First, we analyzed the baseline characteristics of the study sample using the Wilcoxon’s rank sum test for continuous variables and the Chi-square test for binary variables. Second, the crude associations between each of the explanatory variables and the VACS Index at each follow-up visit were analyzed longitudinally using general linear mixed-effects models (GLMM). As a next step, we applied an a priori model-building protocol to fit the multivariable model. Beginning with a full model including homelessness and all covariates significantly associated with the VACS Index in the crude analysis, reduced models were fit by removing one covariate at a time that produced the smallest relative change in the homelessness coefficient. This process was repeated in a stepwise manner until the minimum change in the homelessness coefficient exceeded 5%. The objective of this method is to retain the covariates with the largest relative influence on the association between the primary exposure of interest and the outcome variable [45].

In a sensitivity analysis, we also built a multivariable model including all variables that were significantly associated with the VACS Index in the crude analysis at a standard cut-off of P < 0.05. The final model in this analysis was selected based on removing the variable with the largest P-value in an iterative process and selecting a multivariable model with the lowest Akaike information criterion value. The Sobel test was used to analyze if the association between homelessness and the VACS Index was mediated by ART adherence and we performed a causal mediation analysis to derive the natural direct and indirect effects of homelessness on the VACS Index scores [46, 47]. Finally, we applied inverse probability of censoring weights to account for potential biases caused by differential loss to follow-up among homeless participants [48]. All statistical analyses were conducted using SAS version 9.4 (SAS, Cary, NC, USA) and all P-values were two-sided with a significance threshold of 0.05.

Results

Between December 2005 and November 2017, 802 participants completed at least one study visit, were engaged in ART, provided at least one CD4+ cell count and HIV-1 viral load measurement at baseline and were included in the present study. At baseline, the median age of participants was 44 (interquartile range [IQR] 37–49) years, 256 (31.8%) were female, 437 (54.8%) self-reported white ancestry and 224 (27.9%) participants were homeless. The median VACS Index score at baseline was 18 (IQR 12–29) and the median years of follow-up completed by the study sample was 6.6 (IQR 3.4–8.7). Participants who were homeless at baseline were more likely to report daily use of cocaine (χ² = 16.56, P < 0.001), crack (χ² = 9.69, P = 0.002) heroin (χ² = 45.67, P < 0.001), injection drug use (χ² = 20.67, P < 0.001), incarceration (χ² = 29.62, P < 0.001) and community supervision (χ² = 13.69, P < 0.001) in the past 6 months. Participants who reported being homeless at baseline were less likely to be adherent to ART (χ² = 12.17, P = 0.001) in the past 6 months. The baseline characteristics of the study sample stratified by homelessness are presented in Table 1.

Table 1.

Baseline characteristics stratified by homelessness in the past 6 months among the ACCESS cohort, an ongoing prospective cohort of HIV-positive people who use drugs in Vancouver, Canada (n = 802)

Characteristic	Total n (%)	Homelessness
		Yes 224 (27.9%) n (%)	No 578 (72.1%) n (%)	Chi-square (χ2) value	P-value
VACS Indexb
Median	18	21.5	18	1.12	0.291
IQR	(12–29)	(15–30)	(11–29)
Ageb
Median	44	41	44	15.96	< 0.001
IQR	(37–49)	(35–47)	(38–50)
Sex
Female	256 (31.8)	73 (32.6)	182 (31.5)	0.09	0.764
Male	549 (68.2)	151 (67.4)	396 (68.5)
White ancestry
Yes	437 (54.8)	126 (56.8)	310 (54.1)	0.46	0.500
No	361 (45.2)	96 (43.2)	263 (45.1)
Employmenta
Yes	136 (16.9)	32 (14.3)	104 (18.0)	1.58	0.209
No	669 (83.1)	192 (85.7)	474 (82.0)
Cocaine usea
≥ Daily	54 (6.7)	28 (12.6)	26 (4.5)	16.56	< 0.001
< Daily	749 (93.3)	195 (87.4)	551 (95.5)
Crack usea
≥ Daily	239 (29.9)	84 (38.0)	154 (26.7)	9.69	0.002
< Daily	561 (70.1)	137 (62.0)	422 (73.3)
Heroin usea
≥ Daily	107 (13.3)	59 (26.5)	48 (8.3)	45.67	< 0.001
< Daily	696 (86.7)	164 (73.5)	529 (91.7)
Methamphetamine usea
≥ Daily	54 (6.7)	18 (8.1)	36 (6.2)	0.86	0.354
< Daily	749 (93.3)	205 (91.9)	541 (93.8)
Prescription opioid usea
≥ Daily	35 (4.4)	13 (5.8)	22 (3.8)	1.56	0.211
< Daily	768 (95.6)	210 (93.8)	555 (96.2)
Alcohol usea
≥ Daily	49 (6.1)	18 (8.0)	31 (5.4)	2.01	0.156
< Daily	756 (93.9)	206 (92.0)	547 (94.6)
Cannabis usea
≥ Daily	203 (25.2)	59 (26.3)	143 (24.7)	0.22	0.640
< Daily	602 (74.8)	165 (73.7)	435 (75.3)
Injection drug usea
Yes	569 (70.9)	184 (82.5)	382 (66.2)	20.67	< 0.001
No	234 (29.1)	39 (17.5)	195 (33.8)
Mental illnessa
Yes	441 (55.3)	122 (55.2)	316 (55.1)	0.01	0.969
No	357 (44.7)	99 (44.8)	258 (45.0)
Opioid agonist therapya
Yes	360 (45.0)	99 (44.4)	260 (45.3)	0.05	0.818
No	440 (55.0)	124 (54.6)	314 (54.7)
Drug or alcohol treatmenta
Yes	75 (9.4)	26 (11.7)	48 (8.4)	2.05	0.152
No	724 (90.6)	197 (88.3)	525 (91.6)
Incarcerationa
Yes	86 (10.7)	45 (20.2)	40 (6.9)	29.62	< 0.001
No	716 (89.3)	178 (79.8)	536 (93.1)
Community supervisiona
Yes	114 (14.4)	48 (21.7)	65 (11.4)	13.69	< 0.001
No	678 (85.6)	173 (78.3)	503 (88.6)
ART adherence
≥ 95%	466 (57.9%)	108 (48.2%)	357 (61.8%)	12.17	0.001
< 95%	339 (42.1%)	116 (51.8%)	221 (38.2%)

Bold text refers to P-values < 0.05

^aRefers to activities in the 6 months prior to the follow-up interview, IQR interquartile range, Not all cells may add up to 802 as participants may choose not to answer sensitive questions

^bRefers to continuous variables where the Wilcoxon’s rank sum test was used to compare group distributions above versus below the median value

The bivariable and multivariable GLMM analyses of the VACS Index are presented in Table 2. Factors significantly and positively associated with the VACS Index in the bivariable analysis included age (β = 0.59, 95% Confidence Interval [CI] 0.53–0.64, P < 0.001), white ethnicity (β = 2.44, 95% CI 0.66–4.22, P = 0.007), years since baseline interview (β = 0.53, 95% CI 0.46–0.60, P < 0.001), mental illness (β = 1.55, 95% CI 0.57–2.52, P = 0.002) and homelessness (β = 0.65, 95% CI 0.01–1.31, P = 0.049). Being on community supervision (β = −1.00, 95% CI − 1.76 to − 0.23, P = 0.011) and optimal ART adherence (β = − 2.76, 95% CI − 3.21 to − 2.31, P < 0.001) were negatively associated with VACS Index. In the multivariable analysis, homelessness was significantly associated with increased VACS Index scores (β = 1.43, 95% CI 0.77–2.08, P < 0.001). With ART adherence included in the model, homelessness remained significantly associated with increased VACS Index scores and the effect size was moderately attenuated (β = 1.20, 95% CI 0.55–1.84, P < 0.001). ART adherence (β = − 3.13, 95% CI − 3.58 to − 2.68, P < 0.001) and years since baseline (β = 0.60, 95% CI 0.53–0.67, P < 0.001) were also significantly associated the VACS Index. The mediation analysis indicated that 16.2% of the association between homelessness and the VACS Index was mediated by ART adherence (Sobel test statistic = 4.10, P < 0.001). The total effect of the association between homelessness and the VACS Index was β = 1.43 (95% CI 1.40–1.73, P < 0.001), which was partitioned into a direct effect of β = 1.20 (95% CI 1.13–1.49, P = 0.002) and an indirect effect of β = 0.23 (95% CI 0.22–0.28, P = 0.003). When applying inverse probability of censoring weights to the analysis, the same variables remained significantly associated with the VACS Index although the effect sizes of homelessness (β = 1.03, 95% CI 0.26–1.79, P = 0.009) and ART adherence (β = − 2.89, 95% CI − 3.34 to − 2.44, P < 0.001) were moderately attenuated (Table 3). The effect size of homelessness on the VACS Index increased marginally in the sensitivity analysis (β = 1.27, 95% CI 0.62–1.92, P < 0.001) and other variables included in the final multivariable included white ethnicity (β = 2.80, 95% CI 1.05–2.54, P = 0.002), methamphetamine use (β = − 1.83, 95% CI − 2.88 to − 0.78, P = 0.001), ART adherence (β = − 3.09, 95% CI − 3.54 to − 2.64, P < 0.001) and years since baseline (β = 0.61, 95% CI 0.54–0.68, P < 0.001) (Table 3).

Table 2.

Bivariable and multivariable generalized linear mixed-effects analysis of factors associated with the VACS Index among 802 participants

Characteristic	Unadjusted		Final adjusted
	β (95% CI)	p-value	β (95% CI)	P-value
Homelessness (yes vs. no)	0.65 (0.01, 1.31)	0.049	1.20 (0.55, 1.84)	< 0.001
Age (OR per year older)	0.59 (0.53, 0.64)	< 0.001
Gender (male vs. female)	0.87 (− 1.03, 2.77)	0.372
White ethnicity (yes vs. no)	2.44 (0.66, 4.22)	0.007
Employmenta (yes vs. no)	− 0.47 (− 1.06, 0.11)	0.112
Crack usea (≥ daily vs. < daily)	0.05 (− 0.56, 0.65)	0.880
Cocaine usea (≥ daily vs. < daily)	− 0.27 (− 1.21, 0.67)	0.571
Heroin usea (≥ daily vs. < daily)	0.17 (− 0.61, 0.94)	0.677
Methamphetamine usea (≥ daily vs. < daily)	− 1.01 (− 2.08, 0.05)	0.062
Prescription opioid usea (≥ daily vs. < daily)	− 0.37 (− 1.47, 0.74)	0.514
Alcohol usea (≥ daily vs. < daily)	0.94 (− 0.05, 1.93)	0.063
Cannabis usea (≥ daily vs. < daily)	− 0.09 (− 0.75, 0.58)	0.800
Injection drug usea (≥ daily vs. < daily)	− 0.44 (− 0.99, 0.12)	0.122
Mental illnessa (yes vs. no)	1.55 (0.57, 2.52)	0.002
Opioid agonist therapya (yes vs. no)	− 0.14 (− 0.86, 0.59)	0.714
Drug or alcohol treatmenta (yes vs. no)	− 0.46 (− 1.26, 0.35)	0.268
Incarcerationa (yes vs. no)	− 0.51 (− 1.48, 0.45)	0.298
Community supervisiona (yes vs. no)	− 1.00 (− 1.76, − 0.23)	0.011
ART adherence (≥ 95% vs. < 95%)	− 2.76 (− 3.21, − 2.31)	< 0.001	− 3.13 (− 3.58, − 2.68)	< 0.001
Time since baseline (per year longer)	0.53 (0.46, 0.60)	< 0.001	0.60 (0.53, 0.67)	< 0.001

CI confidence interval, ART antiretroviral therapy

^aRefers to activities in the 6 months prior to the follow-up interview, bold text refers to P-values < 0.05

Table 3.

Multivariable sensitivity analyses of factors associated with the VACS Index and multivariable analysis applying inverse probability of censoring weights

Characteristic	Sensitivity analysis		Inverse probability of censoring weights
	Final adjusted		Final adjusted
	β (95% CI)	P-value	β (95% CI)	P-value
Homelessness (yes vs. no)	1.27 (0.62, 1.92)	< 0.001	1.03 (0.26, 1.79)	0.009
Age (OR per year older)
Gender (male vs. female)
White ethnicity (yes vs. no)	2.80 (1.05, 4.54)	0.002
Employmenta (yes vs. no)
Crack usea (≥ daily vs. < daily)
Cocaine usea (≥ daily vs. < daily)
Heroin usea (≥ daily vs. < daily)
Methamphetamine usea (≥ daily vs. < daily)	− 1.83 (− 2.88, − 0.78)	0.001
Prescription opioid usea (≥ daily vs. < daily)
Alcohol usea (≥ daily vs. < daily)
Cannabis usea (≥ daily vs. < daily)
Injection drug usea (≥ daily vs. < daily)
Mental illnessa (yes vs. no)
Opioid agonist therapya (yes vs. no)
Drug or alcohol treatmenta (yes vs. no)
Incarcerationa (yes vs. no)
Community supervisiona (yes vs. no)
ART adherence (≥ 95% vs. < 95%)	− 3.09 (− 3.54, − 2.64)	< 0.001	− 2.89 (− 3.34, − 2.44)	< 0.001
Time since baseline (per year longer)	0.61 (0.54, 0.68)	< 0.001	0.67 (0.60, 0.74)	< 0.001

CI confidence interval, ART antiretroviral therapy

^aRefers to activities in the 6 months prior to the follow-up interview, bold text refers to P-values < 0.05

Discussion

The objectives of the present study were to evaluate the impact of homelessness on HIV disease progression among PWUD, and analyze if this association was mediated by adherence to ART. We observed a significant association between homelessness and HIV disease progression, as measured by the VACS Index, over a 12-year follow-up period. This association remained significant after adjustment for a range of potential demographic, behavioral and socio-structural confounders, including ART adherence. The mediation analysis revealed that ART adherence mediated 16% of the association between homelessness and the VACS Index.

Although the impact of homelessness on HIV treatment and care has been studied extensively [29, 30, 49–51], this is one of the first studies to longitudinally evaluate the association between periods of homelessness and HIV disease progression, and the first to our knowledge to employ the VACS Index, a validated measure of HIV disease progression and risk of all-cause mortality among PWUD [13, 15, 19]. A previous empirically-based simulation study estimated that eliminating homelessness could increase community levels of viral suppression by 82% among homeless people who use drugs [52]. Real-world data have shown how Housing First initiatives can serve as critical interventions within comprehensive HIV prevention and treatment programs [53]. While our results are in line with these findings, we also found that only 16% of the association between homelessness and the VACS Index was mediated by ART adherence. This suggests that the majority of the association between homelessness and HIV disease progression is mediated by exposures unrelated to exposure to HIV pharmaco-therapies. Therefore, the findings from this study highlight the need to identify and address other potential mechanisms that link homelessness with HIV disease progression and all-cause mortality among this population.

One possible mechanism explaining these findings may be that people who are homeless have more severe psychiatric comorbidities and greater substance use rates that limit their ability to engage in healthcare [37, 54]. Psychiatric problems, and substance use in particular, have been associated with increased risk of becoming homeless and reduced likelihood of exiting homelessness [36, 55]. Homelessness itself has also been linked to increased psychological distress and substance use [36, 56]. These conditions may translate to increased VACS Index scores not just through disruptions in ART adherence, but comorbid substance use and other mental health disorders are associated with medical complications (e.g., cirrhosis, infections, cardiovascular disease) and decreased engagement with the healthcare services to treat these conditions [35, 57]. While substance use was not significantly associated with VACS Index levels in the GLMM analysis, we did observe increased injection drug use and greater daily use of cocaine, crack and heroin among participants who were homeless at baseline. It is possible that people who were homeless at baseline may have had a more extreme history of substance use prior to enrollment in this study. The morbidity associated with a more intense history of substance use (e.g., organ damage, cardiovascular disease,) may have contributed to increased VACS Index scores among participants who were homeless, even though substance use was not significantly associated with the VACS Index over follow-up in the present study [58]. Intersecting stigma associated with homelessness, substance use and HIV also compound the barriers to healthcare access among this population and often lead to delayed clinical presentation, overreliance on emergency services, and increased hospitalization for conditions that are often preventable [35, 59–62].

Another potential explanation for these findings may involve other prevalent health challenges linked to homelessness, in particular food insecurity. Approximately 50% of homeless people receiving ART are estimated to be food insecure, defined as “the limited or uncertain availability of nutritionally adequate, safe foods or the inability to acquire personally acceptable foods in socially acceptable ways” [63, 64]. Food insecurity has been associated with increased rates of morbidity including hypertension, diabetes and heart disease [65–68]. Among HIV-positive people engaged in ART, food insecurity was associated with a two-fold increase in mortality risk among those who were underweight after adjustment for socio-demographic factors, substance use, baseline CD4+ cell count and ART adherence [69]. In addition to the direct effects of food insecurity on malnutrition, food insecurity has been linked to incomplete viral suppression and lower CD4+ cell counts among homeless and marginally housed HIV-positive individuals, and these associations were only marginally mediated by ART adherence [70]. Other studies have found that the odds of viral suppression were 70% lower among people who were food insecure after adjusting for ART adherence, and the plasma concentration of protease-inhibitor antiretrovirals such as darunavir is decreased by 30% when taken without food. The bioavailability of other protease inhibitors has been shown to increase by 35–700% when taken with food compared to a fasted state [71–73]. Given the high prevalence of food insecurity among homeless people living with HIV, and the impact of food insecurity on HIV clinical outcomes, these findings may explain the association between homelessness and the VACS Index that we observed, as well as the limited role of ART adherence in mediating this association.

People living with HIV also have higher rates of non-infectious chronic diseases due to increased inflammation and immune activation that persists despite engagement in effective ART [8, 9]. A major contributor to chronic inflammation and immune activation is gut mucosal dysfunction, which is characterized by epithelial permeability, translocation of microbial products into circulation and microbiome dysbiosis [74, 75]. The gastrointestinal tract is the largest source of CD4+ T cells (the most abundant HIV reservoirs) and inflammation of the gastrointestinal tract is an important contributor to ongoing viral replication and HIV persistence [76–79]. In addition to HIV-associated inflammation, behavioral predictors of inflammation—including psychosocial stress, smoking, consuming high-fat diets and irregular sleep patterns—are ubiquitous among PWUD experiencing homelessness [80–82]. Chronic inflammation has been strongly associated with significant morbidity and mortality, including causes of death that are common among homeless populations, such as cancer and heart disease [33, 83]. As a result, the inflammatory and immune consequences associated with these exposures may have contributed to the association between homelessness and the VACS Index that was not mediated by ART adherence.

This study has limitations. Since many study variables were measured by self-report, socially desirable reporting and recall bias may have influenced the measurement of stigmatized behaviors including substance use. However, the validity and reliability of self-report methods among PWUD has been demonstrated previously [84]. Our findings may not be generalizable to other populations of PWUD since the ACCESS study is not a random sample. The VACS Index was recently modified to include body mass Index, albumin and white blood cell count to improve the discrimination of mortality [85]; however these measures were not available in our study. The measure of mental illness did not assess the severity of psychiatric symptoms, which may have mediated the association between homelessness and the VACS Index. Lastly, it is possible that residual confounding influenced the association between homelessness and the VACS Index since this was an observational study, although the no-cost universal access to ART in the study setting minimizes confounding due to financial barriers to care. Nevertheless, measures of food insecurity and behavioral predictors inflammation (e.g., psychosocial stress, consuming high-fat diets and irregular sleep patterns) were not available in the present study. These factors may have contributed to residual confounding given that these factors may moderate ART effectiveness and the association between homelessness and HIV disease progression [70–73, 80–82].

Conclusions

In summary, homelessness was significantly associated with HIV disease progression, as measured by the VACS Index, among a prospective cohort of PWUD living with HIV over a 12-year follow-up period. While this association was partially mediated by ART adherence, the majority of the association appeared to be mediated by factors other than engagement with HIV care. These findings have important implications for health interventions designed to address the intersecting challenges associated with homelessness and substance use among people living with HIV. Future efforts are needed to quantify and address important risk factors for HIV disease progression and mortality among homeless PWUD living with HIV.

Data Availability

The data cannot be made publicly available due to the ethical agreements of these studies.