Relationship between Housing Characteristics and Care Outcomes among Women Living with HIV: Latent Class Analysis

Abstract

Housing plays a critical role in the care outcomes of individuals living with a HIV, yet few studies have examined the unique housing profiles of women living with HIV (WLH), especially among those belonging to low-income racial/ethnic minority groups. In this study, authors conducted a latent class analysis to generate latent profiles of women (N = 1,501) according to their housing status and household characteristics and assessed associations between classes and sociodemographic and behavioral characteristics and between classes and three HIV care outcomes: retention in care, viral suppression, and sustained viral suppression. A three-class model was selected using five dichotomized housing indicators: housing status, head of household status, living with minors, living with another adult, and disclosure of HIV status to adults with whom they live. While class 1 and class 2 had comparable HIV care outcomes, women in class 3—who were predominately unstably housed, living alone, and did not disclose their HIV status with those with whom they lived—had significantly lower crude odds of successful care outcomes. When accounting for sociodemographic and behavioral factors, significant differences in retention in care persisted. Findings underscore the need for tailored interventions for subgroups of low-income WLH and provide contextual details around the role of housing experiences.

People living with HIV are vulnerable to housing instability and are often at risk of experiencing homelessness, the most extreme form of housing deprivation (U.S. Department of Housing and Urban Development, 2014). Among people living with HIV (PLH), lack of stable housing is a significant barrier to accessing medical care, adherence to antiretroviral therapy (ART), viral suppression, and reduction of transmissions (Aidala et al., 2016). Although sustained use of ART is the strongest predictor of survival among PLH, not all groups benefit equally from this medical advancement. Women living with HIV (WLH) who experience homelessness and housing instability are less likely to be linked to treatment, and among those linked to treatment, a range of sociodemographic and psychosocial factors such as stigma, childcare burden, and poverty may impact their ability to benefit from care and treatment (Anastos et al., 2005; Aziz & Smith, 2011; Losina et al., 2009; Riley et al., 2007; Wolitski et al., 2008).

The Ryan White HIV/AIDS Program, funded by the Health Resources and Services Administration (HRSA), provides primary HIV medical care, medications, and medical case management services to PLH in the United States who are low-income and have no other source of HIV medical services. Eligibility for Ryan White Part A Program–funded services is limited to individuals who have a gross household income of up to 400% of the federal poverty level (FPL). Women make up 26.2% of national Ryan White Program recipients (HRSA, 2021). Among women who are Ryan White Program recipients, 69.8% are living at or below a household income of 100% of the FPL, indicating extreme poverty, compared with 60.7% of the national average of people in the Ryan White Program. Additionally, the majority of women in the Ryan White Program (83.6%) belong to racial and ethnic minority groups, primarily African American/Black and Hispanic/Latina women (HRSA, 2021).

For racial and ethnic minority WLH facing housing instability and homelessness, intersections of gender, ethnicity, HIV status, and poverty compound vulnerability to poor HIV care outcomes and contribute to challenges in addressing persistent health disparities. In general, WLH are more likely to live in environments with elevated rates of poverty, crime, and illicit drug use; have low health literacy; and experience stigma and discrimination (Aziz & Smith, 2011; Riley et al., 2007). Further social and structural factors such as lack of insurance coverage and transportation; mistrust of health systems; and lack of access to high-quality, culturally competent, and consistent care present as barriers that put unstably housed WLH at increased risk of delayed entry into care and poor outcomes in retention and adherence to care (Thompson et al., 2012). Furthermore, many of the social and structural factors that place racial and ethnic minority WLH at risk of acquiring HIV continue to contribute to poor outcomes after linkage to care and treatment.

Housing, which is often conceptualized as one factor influencing an individual’s ability to maintain health, can be more fully understood as a structural mechanism connected to and influencing a myriad of other risk factors. People without stable and adequate housing are vulnerable to compounded negative health consequences, including but not limited to poor HIV outcomes, mental health issues, isolation, substance use issues, and they face challenges to maintaining stable relationships and cohesive networks of social support, which are known to positively impact an individual’s health status (Aidala et al., 2016). In addition to detrimental effects of housing instability and homelessness on an individual’s health and well-being, there are also negative consequences at the social and community level. For all PLW, adherence to ART is essential to achieving sustained viral suppression, which is the ultimate goal of the HIV care continuum. Achieving viral suppression not only minimizes an individual’s disease progression, but also reduces risk of transmission and minimizes the emergence of drug resistance (CDC, 2019). While studies have shown that adherence predicts viral suppression, assessments of clinical data demonstrate that experience of homelessness continually predicts less viral suppression and that WLH in the face of housing instability are at heightened risk of care interruption and treatment discontinuation (Riley et al., 2019; Thakarar et al., 2016).

There is evidence in the literature to suggest the need for prioritizing the importance of housing to achieve sustained viral suppression for low-income WLH. Yet, attempts to improve access, retention, and adherence to ART among racial and ethnic minority WLH facing housing instability are limited by the ability to identify and intervene during such experiences. While the overrepresentation of single adult men in homeless and HIV populations has been well documented, less attention has been given to understanding the lived experiences and profiles of WLH, especially among those who belong to racial and ethnic minority groups (Parashar, 2016). There is a need for a greater understanding of the unique housing experiences of women and the impact on housing on HIV care outcomes as a deeper understanding of housing experiences and characteristics can provide insight for necessary clinical screening tools and a better understanding of how to more comprehensively meet the needs of some of the most vulnerable groups of  WLH.

The purpose of our study was to conduct a latent class analysis to identify patterns of housing situations and characteristics among a group of low-income WLH. Furthermore, the goal was to identify sociodemographic and behavioral characteristics associated with group membership and assess the impact of latent housing groups on three HIV care outcomes: retention in care, viral suppression, and sustained viral suppression. This study is one of the few that closely examine characteristics of housing experiences and their role on retention in care, viral suppression, and sustained viral suppression among a racially and ethnically diverse sample of WLH.

METHOD

Data for this study included merged data from client intake, service, laboratory, and biannual health assessments of WLH who received services under the Ryan White Part A Program (RWP) in Miami-Dade County during 2017. The data were collected through questionnaires administered to clients by case mangers across 10 agencies. Miami-Dade County is a metropolitan area with one of the highest HIV new-case incidence rates in the United States and is designated as a geographic hot spot for HIV, so it’s a target for HIV prevention efforts (CDC, 2020). Data used in this study were from women who were 18 years or older and were considered enrolled in care in 2017, defined as receiving at least one medical case management or peer education service from the RWP during 2017. Clients referred to the program only for ancillary services from an out-of-network provider; those who were missing health assessment forms; cases closed due to mortality, relocation, or financial ineligibility; and auto-closure cases due to lack of follow-up were excluded from the analysis.

The data were collected at medical case management sites. Intake, health assessment, and laboratory data were then merged for each participant for analysis. The original sampling frame included data from 1,609 women. The final sample used in the analysis included 1,501 women. We excluded cases due to missing data: sustained viral suppression (n = 102 missing), living arrangement (n = 1 missing), and race/ethnicity unknown (n = 5 missing). A further description of decisions for exclusion and the handling of missing data is included in study design and analysis.

Study Design and Analysis

The analysis for this study was conducted in four primary phases. First, latent class analysis was performed to generate latent classes based on housing status and household characteristics. Second, bivariate descriptive analyses were conducted to compare sociodemographic and behavioral characteristics with class membership. Next, multinomial logistic regressions assessed the association between class membership and sociodemographic and behavioral characteristics. Finally, logistic regression were used to examine the association between class memberships and HIV care outcomes.

Latent class analysis was chosen as an appropriate statistical methodology because it permits detection of latent groups (classes) that cannot be directly observed using categorical indicator variables alone (Okeke et al., 2018). Thus, this statistical method permits a more nuanced understanding of housing profiles among the sample by identifying otherwise invisible groups of women who share similar patterns of characteristics. Latent classes were modeled according to patterns of housing status and household structure using five dichotomized variables: housing status (homeless/institutional housing vs. stably housed), head of household status (female headed vs. non–female headed [including joint head, male head, and no head of household, such as when institutionalized]), living with minors (yes or no), living with other adults (yes or no), and disclosure of HIV status to adults with whom they live (yes or no/no adults).  Table 1 provides a comprehensive list of variables used in this study. The best fit model from latent classes generated by PROC LCA in SAS 9.4 was selected by assessing the lowest Bayesian Information Criterion (BIC) and a higher entropy, suggesting a better distinction between classes (Celeux & Soromenho, 1996; Schwarz, 1978).

Table 1:

List of Variables Included

Variable	Response Options
Housing variables
What is your housing status?a	Homeless/institutional housing (e.g., transitional, residential, healthcare, or correctional facilities) or stably housed
Head of household situationa	Female-headed or non–female headed (e.g., joint headed, male headed, no head)
Are you living with minors (<18 years old) in the household?	Yes, no
Are you living with other adults in the household?	Yes, no
Have you disclosed your HIV status to the adults with whom you live?	Yes or no/no adults
Sociodemographic variables
Age (years)	18–34, 35–49, 50+
Race/ethnicity	Non-Hispanic Black, Hispanic, Haitian, non-Hispanic White
Preferred language	English, Spanish, other
U.S. born	Yes, no
Household income	<100% FPL, ≥100% FPL
Are you currently working?	Yes, no
Do you have transportation to appointments for healthcare/dental care/social service appointments?	Yes, no
Behavioral variables
Have you been feeling depressed/anxious?	Yes, no
Do you currently receive or need mental health services?	Yes, no
Have you ever experienced domestic violence/abuse?	Yes, no
Has drug/alcohol use resulted in any problem in daily activity, legal issue, or hazardous situation?	Yes, no
Does your drug use affect your adherence?	Yes, no
Would you like substance use treatment now?	Yes, no
Do you have a support system?	Yes, no
HIV care outcome variables
Retained in care (engaged in HIV care at least twice within one year ≥3 months apart)	Yes, no
Viral suppression (having a viral load test of <200 copies/ml in the last laboratory test of the year)	Yes, no
Sustained viral suppression (all viral load tests <200 copies/ml in the year)	Yes, no

Note: FPL = federal poverty level.

^a When collecting these data, medical case managers gave participants the additional response options included in the parentheses. The responses were then dichotomized for the analysis in this study.

Bivariate descriptive analyses were conducted using chi-square tests to identify the sociodemographic and behavioral factors associated with latent class membership. Sociodemographic variables included age, race/ethnic group, U.S. nativity, preferred language, household income, employment, and access to transportation. Additional behavioral variables included having social support, feeling depressed or anxious, receiving or needing mental health services, problematic drug use, drug use affecting adherence, needing substance use treatment services, and ever experiencing domestic violence. Next, multinomial logistic regressions were conducted to assess the association between class membership and sociodemographic and behavioral characteristics in multivariate models. Adjusted odds ratios (aORs) and 95% confidence intervals (CIs) were computed.

Last, logistic regression models were conducted to examine the association between latent classes and retention in care, viral suppression, and sustained viral suppression. To model class membership and its impacts on the outcomes of interest, three different models were run. In the first model (crude model), only class membership was included as a predictor. In the second model (sociodemographic model), we included class membership and controlled for sociodemographic variables. In the third model (full model), we controlled for sociodemographic variables as well as behavioral variables. Retention in care was defined as having two or more medical care visits three or more months apart within a one-year period. Viral suppression was defined as having a viral load test of <200 copies/ml in the last laboratory test of the year. Sustained viral suppression was defined as having a viral load count of <200 copies/ml in all viral load tests within 2017 and for at least two tests at least three months apart (Crepaz et al., 2018). If any test in 2017 was ≥200 copies/ml (not suppressed), women were considered not to have achieved sustained viral suppression in 2017. For those women with only one suppressed viral load test in 2017 (or those with only two tests less than three months apart), we included the last viral load test result of the prior year. Those women with only one suppressed viral load test in 2017 and no viral load test in the prior year were excluded from the analysis, as were women who had more than one suppressed viral load test but the tests were less than three months apart. To assess the impact of excluding women based on missing viral load data, we ran a sensitivity analysis to compare the results of the logistic regressions on sustained viral suppression with the full sampling frame of 1,609 women to the outcomes of the final sample used in this study. There were no significant changes in our findings during this analysis. ORs and 95% CIs were computed for all analyses. All analyses were performed using SAS Version 9.4. The Florida International University institutional review board approved the procedures for this study.

RESULTS

The final sample included data from 1,501 women, 18 years of age or older, who were living with HIV and were enrolled in the Miami-Dade County RWP during 2017. A three-class model for the latent classes was selected as a best fit based on BIC (135.28) and entropy (0.86). Table 2 shows the distribution of participants according to housing indicators that determined class membership. Class 1 (56%) consisted of primarily stably housed women, majority non–female headed households, 30% having at least one minor in the household, all having another adult in the household, and all having disclosed their HIV status to the adults with whom they live. Class 2 (39%) consisted of entirely stably housed women, majority female headed, 26.6% with at least one minor in the household, majority with no other adult in the household, and none with reported HIV disclosure to adults with whom they live. Class 3 (5%) consisted entirely of women who were homeless or unstably housed, majority non–female headed, majority with no minors, no other adults, and no reported HIV disclosure.

Table 2:

Distribution of Housing Status and Household Structure Variables by Class Membership among Women Living with HIV

	Total Sample	Class 1	Class 2	Class 3
	(N = 1,501)	(n = 805)	(n = 655)	(n = 41)
Characteristic	n (%)	n (%)	n (%)	n (%)	p
Housing status					<.0001
Stably housed	1,425 (94.9)	770 (95.7)	655 (100.0)	0 (0.0)
Homeless/institutional housing	76 (5.1)	35 (4.4)	0 (0.0)	41 (100.0)
Female-headed household					<.0001
No	695 (46.3)	441 (54.8)	227 (34.7)	27 (65.9)
Yes	806 (53.7)	364 (45.2)	428 (65.3)	14 (34.2)
Minor(s) in the household					.0552
No	1,080 (72.0)	564 (70.1)	481 (73.4)	35 (85.4)
Yes	421 (28.1)	241 (30.0)	174 (26.6)	6 (14.6)
Other adult(s) in the household					<.0001
No	502 (33.4)	0 (0.0)	467 (71.3)	35 (85.4)
Yes	999 (66.6)	805 (100.0)	188 (28.7)	6 (14.6)
Disclosure to adult with whom you live					<.0001
No/no adult	696 (46.4)	0 (0.0)	655 (100.0)	41 (100.0)
Yes	805 (53.6)	805 (100.0)	0 (0.0)	0 (0.0)

Notes: Class 1 = stably housed, female headed, living with another adult, some minors, and disclosure; Class 2 = stably housed, non–female headed, no other adults, some minors, and no disclosure; Class 3 = unstably housed, non–female headed, no minors, no adults, and no disclosure.

Table 3 shows the results of the bivariate analyses exploring correlates of class membership. The majority of women in the full sample identified as belonging to a racial/ethnic minority group, with 41.7% of the sample identifying as non-Hispanic Black, 30.3% Hispanic, 24.9% Haitian, and 3.1% non-Hispanic White. A slight majority of the women were foreign born (57.9%), and half (50.6%) were older than 50 years old. Of the entire sample of women, 90.2% were retained in care, 87.8% were virally suppressed, and 76.6% had sustained viral suppression in 2017.

Table 3:

Distribution of Sociodemographic and Behavioral Characteristics by Class Membership among Women Living with HIV

	Total Sample	Class 1	Class 2	Class 3
	(N = 1,501)	(n = 805)	(n = 655)	(n = 41)
Characteristic	n (%)	n (%)	n (%)	n (%)	p
Race/ethnicity					<.0001
Non-Hispanic Black	626 (41.7)	344 (42.7)	258 (39.4)	24 (58.5)
Hispanic	455 (30.3)	260 (32.3)	189 (28.9)	6 (14.6)
Haitian	373 (24.9)	178 (22.1)	189 (28.9)	6 (14.6)
Non-Hispanic White	47 (3.13)	23 (2.9)	19 (2.9)	5 (12.2)
Age (years)					<.0001
18–34	222 (14.8)	141 (17.5)	71 (10.8)	10 (24.4)
35–49	520 (34.6)	302 (37.5)	205 (31.3)	13 (31.7)
50+	759 (50.6)	362 (45.0)	379 (57.9)	18 (44.0)
Preferred language					.0002
English	753 (50.2)	415 (51.6)	307 (46.9)	31 (75.6)
Spanish	397 (26.5)	226 (28.1)	167 (25.5)	4 (9.8)
Other	351 (23.4)	164 (20.4)	181 (27.6)	6 (14.6)
Nativity					.0084
Foreign born	869 (57.9)	453 (56.3)	400 (61.1)	16 (39.0)
U.S. born	632 (42.1)	352 (43.7)	255 (38.9)	25 (61.0)
Household income					<.0001
≥100% FPL	736 (49.0)	381 (47.3)	350 (53.4)	5 (12.2)
<100% FPL	765 (51.0)	424 (52.7)	305 (46.6)	36 (87.8)
Work status					<.0001
Working	747 (49.8)	368 (45.7)	370 (56.5)	9 (22.0)
Not working	754 (50.2)	437 (54.3)	285 (43.5)	32 (78.1)
Has transportation to appointments					.0086
Yes	1,392 (92.7)	751 (93.3)	608 (92.8)	33 (80.5)
No	109 (7.3)	54 (6.7)	47 (7.2)	8 (19.5)
Feeling anxious or depressed					.1357
No	1,239 (82.5)	658 (81.7)	551 (84.1)	30 (73.2)
Yes	262 (17.5)	147 (18.3)	104 (15.9)	11 (26.8)
Receives or needs mental health services					<.0001
No	1,225 (81.6)	642 (79.8)	562 (85.8)	21 (51.2)
Yes	276 (18.4)	163 (20.3)	93 (14.2)	20 (48.8)
Ever experienced domestic violence					.0020
No	1,381 (92.0)	738 (91.7)	611 (93.3)	32 (78.1)
Yes	120 (8.0)	67 (8.3)	44 (6.7)	9 (22.0)
Problematic drug use					<.0001
No	1,466 (97.7)	786 (97.6)	650 (99.2.)	30 (73.2)
Yes	35 (2.3)	19 (2.4)	5 (0.8)	11 (26.8)
Drug use affects adherence					<.0001
No	1,450 (96.6)	777 (96.5)	640 (97.7)	33 (80.5)
Yes	51 (3.4)	28 (3.5)	15 (2.3)	8 (19.5)
Would like substance use treatment now					<.0001
No	1,478 (98.5)	796 (98.9)	649 (99.1)	33 (80.5)
Yes	23 (1.5)	9 (1.1)	6 (1.0)	8 (19.5)
Has support system					<.0001
No	209 (13.9)	81 (10.1)	116 (17.7)	12 (29.3)
Yes	1,292 (86.1)	724 (89.9)	539 (82.3)	29 (70.7)
Retained in HIV care					.0009
No	147 (9.8)	73 (9.1)	63 (9.6)	11 (26.8)
Yes	1,354 (90.2)	732 (90.9)	592 (90.4)	30 (73.2)
Viral suppression					.0532
No	183 (12.2)	96 (11.9)	77 (11.8)	10 (24.4)
Yes	1,318 (87.8)	709 (88.1)	578 (88.2)	31 (75.6)
Sustained viral suppression					.0021
No	352 (23.5)	185 (23.0)	148 (22.6)	19 (46.3)
Yes	1,149 (76.6)	620 (77.0)	507 (77.4)	22 (53.7)

Note: FPL = federal poverty level.

The distributions of all sociodemographic variables were significantly different by class membership, as seen in Table 3. The majority of individuals in class 3 (58.5%) identified as non-Hispanic Black, compared with 42.7% and 39.4% in class 1 and 2, respectively. The majority of individuals in class 3 were U.S. born (61.0%), compared with 43.7% and 38.9% in class 1 and 2, respectively. Similarly, the distribution of all behavioral variables (except for feeling depressed or anxious) and HIV care outcome variables were significantly different by class membership. The percentages of women having achieved sustained viral suppression in 2017 were 77.0%, 77.4%, and 53.7% for classes 1, 2 and 3, respectively.

Table 4 shows results from multinomial logistic regressions, which modeled the odds of being in class 2 and 3 versus class 1 for demographic and behavioral characteristics. When compared with class 1, individuals in class 2 had significantly higher odds of being 50 years of age or older versus 35–49 years of age (aOR = 1.65; 95% CI [1.30, 2.10]), were less likely to be not working versus working (aOR = 0.61; 95% CI [0.48, 0.78]), and were more likely to report no support system versus having a support system (aOR: 1.99; 95% CI [1.45, 2.73]). Moreover, when compared with class 1, individuals in class 3 were more likely to be living with a household income below 100% of the FPL (aOR = 3.90; 95% CI [1.38, 11.04]), report receiving or needing mental health services (aOR = 3.16; 95% CI [1.33, 7.50]), report needing substance use treatment now (aOR = 4.20; 95% CI [1.18, 14.91]), and report having no support system (aOR = 4.13; 95% CI [1.82, 9.38]).

Table 4:

Adjusted Odds Ratios and 95% Confidence Intervals of Class Membership by Characteristics of Women Living with HIV Obtained from Multinomial Logistic Regression Output

	Class 2 vs. Class 1	Class 3 vs. Class 1
Characteristic	aOR [95% CI]	aOR [95% CI]
Age group (years)
18–34 vs. 35–49	0.74 [0.52, 1.05]	1.37 [0.53, 3.56]
50+ vs. 35–49	1.65 [1.30, 2.10]	1.04 [0.46, 2.32]
Race/ethnicity
Non-Hispanic Black vs. White	0.98 [0.51, 1.88]	0.63 [0.18, 2.30]
Hispanic vs. White	0.90 [0.40, 2.06]	0.26 [0.04, 1.70]
Haitian vs. White	1.00 [0.43, 2.31]	0.27 [0.03, 2.06]
Preferred language
Other vs. English	1.21 [0.65, 2.25]	1.28 [0.18, 9.02]
Spanish vs. English	0.86 [0.48, 1.56]	0.37 [0.06, 2.39]
U.S. Born
No vs. yes	1.07 [0.73, 1.56]	2.12 [0.74, 6.09]
Household income
<100% FPL vs. ≥100% FPL	0.98 [0.77, 1.24]	3.90 [1.38, 11.04]
Not currently working
Yes vs. no	0.61 [0.48, 0.78]	1.04 [0.42, 2.52]
Has transportation to appointments
No vs. yes	1.03 [0.68, 1.58]	2.10 [0.81, 5.46]
Feeling anxious or depressed
Yes vs. no	1.21 [0.85, 1.71]	0.51 [0.20, 1.31]
Receives or needs mental health services
Yes vs. no	0.75 [0.53, 1.07]	3.16 [1.33, 7.50]
Ever experienced domestic violence
Yes vs. no	1.03 [0.67, 1.60]	1.50 [0.56, 4.00]
Problematic drug use
Yes vs. no	0.42 [0.14, 1.27]	2.71 [0.76, 9.68]
Drug use affects adherence
Yes vs. no	0.96 [0.47, 1.94]	1.69 [0.51, 5.69]
Would like substance use treatment now
Yes vs. no	1.46 [0.47, 4.49]	4.20 [1.18, 14.91]
Support system
No vs. yes	1.99 [1.45, 2.73]	4.13 [1.82, 9.38]

Notes: Class 1 = stably housed, female headed, living with another adult, some minors, and disclosure; Class 2 = stably housed, non–female headed, no other adults, some minors, and no disclosure; Class 3 = unstably housed, non–female headed, with no minors, no adults, and no disclosure; FPL: federal poverty level. Bolded text indicates significant differences.

Table 5 shows results from logistic regression models comparing the odds of care outcomes between housing classes. Women belonging to class 3 had significantly lower crude odds of being retained in care (OR = 0.27; 95% CI [0.13, 0.57]), being virally suppressed (OR = 0.42; 95% CI [0.20, 0.88]), and having sustained viral suppression (OR = 0.35; 95% CI [0.18, 0.65]) compared with women in class 1. When adjusting for demographic characteristics, individuals in class 3 persisted having significantly lower odds of being retained in care (aOR = 0.30; 95% CI [0.14, 0.63]) and having sustained viral suppression (aOR = 0.48; 95% CI [0.25, 0.94]) compared with class 1. When running the full model that included class membership, sociodemographic, and behavioral risk variables, individuals in class 3 continued to have significantly lower odds of being retained in care (aOR = 0.33; 95% CI [0.14, 0.76]). The association with viral suppression and sustained viral suppression was not retained in the full model.

Table 5:

Crude and Adjusted Odds Ratios and 95% Confidence Intervals for HIV Care Outcomes by Household Structure Class Membership for Women Living with HIV

	Crude Modela		Sociodemographic Modelb		Full Modelc
Outcome	Class 2 vs. Class 1	Class 3 vs. Class 1	Class 2 vs. Class 1	Class 3 vs. Class 1	Class 2 vs. Class 1	Class 3 vs. Class 1
	OR [95% CI]	OR [95% CI]	aOR [95% CI]	aOR [95% CI]	aOR [95% CI]	aOR [95% CI]
Retained in care	0.94 [0.66, 1.34]	0.27 [0.13, 0.57]	0.79 [0.55, 1.14]	0.30 [0.14, 0.63]	0.79 [0.55, 1.15]	0.33 [0.14, 0.76]
Viral suppression	1.02 [0.74, 1.40]	0.42 [0.20, 0.88]	0.87 [0.62, 1.21]	0.51 [0.23, 1.11]	0.84 [0.60, 1.18]	0.70 [0.30, 1.66]
Sustained viral suppression	1.02 [0.80, 1.31]	0.35 [0.18, 0.65]	0.87 [0.67, 1.21]	0.48 [0.25, 0.94]	0.85 [0.66, 1.11]	0.61 [0.30, 1.26]

Notes: Class 1 = stably housed, female headed, living with another adult, some minors, and disclosure; Class 2 = stably housed, non–female headed, no other adults, some minors, and no disclosure; Class 3 = unstably housed, non–female headed, with no minors, adults, and no disclosure. Bolded text indicates significant differences.

^a Crude model only includes class membership as a predictor.

^b Sociodemographic model includes class membership as a predictor while controlling for demographic variables, which include age, race/ethnicity, U.S. nativity, preferred language, federal poverty level, employment, and access to transportation.

^c Full model includes class membership as a predictor while controlling for sociodemographic and behavioral variables, which include age, race/ethnicity, U.S. nativity, preferred language, federal poverty level, employment, access to transportation, receives or needs mental health services, feeling depression or anxious, ever experience domestic violence, support system, problematic drug use, drug use affect adherence, and would like substance use treatment now.

When further examining the results from the full model on the three primary HIV care outcomes of interest, results from Table 6 indicate that, in addition to women in class 3 having significantly lower odds of being retained in care, women who were not currently working also had lower odds of being retained in care (aOR = 0.62; 95% CI [0.41, 0.93]). Women who were older (50+ versus 35–49) had higher odds of being retained in care (aOR = 1.54; 95% CI [1.02, 2.31]) and also had higher odds of being virally suppressed (aOR = 1.93; 95% CI [1.33, 2.80]). Women who were non-Hispanic Black (aOR = 0.24; 95% CI [0.06, 0.92]) and Haitian (aOR = 0.13; 95% CI [0.03, 0.62]), women who had a household income less than 100% of the FPL (aOR = 0.63; 95% CI [0.44, 0.91]), and women who reported problematic drug use (aOR = 0.18; 95% CI [0.07, 0.48]) had lower odds of being virally suppressed. Women with household income below 100% of the FPL also had lower odds of achieving sustained viral suppression (aOR = 0.59; 95% CI [0.44, 0.78]), as were those who reported feeling anxious or depressed (aOR = 0.65; 95% CI [0.45, 0.95]), and those who reported problematic drug use (aOR = 0.28; 95% CI [0.11, 0.67]).

Table 6:

Adjusted Odds Ratios and 95% Confidence Intervals for the Full Model Assessing HIV Care Outcomes While Controlling for Class Membership, Sociodemographic, and Behavioral Variables for Women Living with HIV

Characteristic	Retention in Care aOR [95% CI]	Viral Suppression aOR [95% CI]	Sustained Viral Suppression aOR [95% CI]
Class
Class 2 vs. Class 1	0.79 [0.55, 1.15]	0.84 [0.60, 1.18]	0.85 [0.66, 1.11]
Class 3 vs. Class 1	0.33 [0.14, 0.76]	0.70 [0.30, 1.66]	0.61 [0.30, 1.26]
Age group (years)
18–34 vs. 35–49	0.64 [0.40, 1.03]	0.85 [0.55, 1.32]	0.76 [0.53, 1.09]
50+ vs. 35–49	1.54 [1.02, 2.31]	1.93 [1.33, 2.80]	2.12 [1.60, 2.83]
Race/ethnicity
Non-Hispanic Black vs. White	0.79 [0.29, 2.22]	0.24 [0.06, 0.92]	0.65 [0.30, 1.42]
Hispanic vs. White	0.45 [0.14, 1.52]	0.31 [0.07, 1.39]	0.93 [0.35, 2.49]
Haitian vs. White	0.75 [0.19, 2.93]	0.13 [0.03, 0.62]	0.52 [0.19, 1.39]
Preferred language
Other vs. English	1.12 [0.39, 3.24]	1.12 [0.45, 2.80]	0.88 [0.43, 1.81]
Spanish vs. English	1.62 [0.69, 3.81]	0.69 [0.26, 1.78]	1.02 [0.49, 2.11]
U.S. born
No vs. yes	1.73 [0.92, 3.24]	2.31 [1.25, 4.26]	1.38 [0.88, 2.17]
Household income
<100% FPL vs. ≥100% FPL	1.08 [0.73, 1.60]	0.63 [0.44, 0.91]	0.59 [0.44, 0.78]
Not currently working
Yes vs. no	0.62 [0.41, 0.93]	0.95 [0.65, 1.38]	0.83 [0.62, 1.11]
Has transportation to appointments
No vs. yes	1.53 [0.72, 3.25]	0.84 [0.47, 1.49]	0.73 [0.46, 1.15]
Feeling anxious or depressed
Yes vs. no	1.12 [0.63, 1.94]	0.78 [0.48, 1.24]	0.65 [0.45, 0.95]
Receives or needs mental health services
Yes vs. no	1.45 [0.83, 2.56]	1.07 [0.66, 1.74]	0.90 [0.62, 1.32]
Ever experienced domestic violence
Yes vs. no	1.51 [0.72, 3.16]	1.00 [0.55, 1.79]	1.03 [0.64, 1.65]
Problematic drug use
Yes vs. no	0.54 [0.19, 1.52]	0.18 [0.07, 0.48]	0.28 [0.11, 0.67]
Drug use affects adherence
Yes vs. no	1.15 [0.45, 2.95]	1.76 [0.70, 4.45]	1.65 [0.78, 3.45]
Would like substance use treatment now
Yes vs. no	0.58 [0.19, 1.77]	0.83 [0.28, 2.48]	0.87 [0.33, 2.34]
Support system
No vs. yes	1.04 [0.62, 1.73]	1.11 [0.68, 1.82]	0.99 [0.68, 1.45]

Notes: Class 1 = stably housed, female headed, living with another adult, some minors, and disclosure; Class 2 = stably housed, non–female headed, no other adults, some minors, and no disclosure. Class 3 = unstably housed, non–female headed, with no minors, adults, and no disclosure; FPL = federal poverty level. Bolded text indicates significant differences.

DISCUSSION

This study sought to identify patterns of housing and household characteristics among a cohort of low-income WLH using latent class analysis, assess the predictors of membership in the latent groups, and identify the relationships between latent groups and HIV care outcomes. Results from the analyses indicate three unique classes of WLH as they relate to patterns in their reported housing status and household characteristics. Women in class 1 tended to be stably housed, in non–female headed homes, some living with minors, and living with at least one adult who was aware of their HIV status. Women in class 2 were all stably housed, tended to be in female-headed homes, some living with minors, majority with no other adults, and did not disclose their HIV status. Women in class 3 tended to be in homeless or unstably housed situations, with no minors, non–female headed household, no other adults, and did not disclose their HIV status to those with whom they live. When comparing these three unique classes, women in class 1 and class 2 had comparable HIV care outcomes. Significant and persistent differences in HIV care outcomes among women belonging to class 3 suggest the need for targeted and tailored interventions for specific subgroups of low-income WLH.

Unstably housed women who live alone, with no minors, no other adults, and reported not disclosing their HIV status (class 3) are a unique group who are at increased risk of poor HIV care outcomes. Moreover, women belonging to this high-risk class were significantly more likely to be living with a household income below 100% FPL, indicating more extreme poverty, receiving or needing mental health services, needing substance use treatment, and having no support system. Our findings suggest that women experiencing high-risk living situations are also more likely to experience risky behavioral situations and have unmet service needs. These compounded experiences can further exacerbate poor HIV outcomes.

When assessing the association between classes and retention in care, viral suppression, and sustained viral suppression, there were significant and persistently worse outcomes among women belonging to class 3. Even when controlling for sociodemographic and behavioral risk variables, belonging to class 3 was associated with worse outcomes in retention in care. Retention in care is a critical component in the HIV care continuum (HRSA, 2017; Kay et al., 2016). If not addressed in a timely manner, lack of retention in care can quickly lead to longer-term medical treatment discontinuation (Lee et al., 2018). Findings suggest that women whose profiles meet those of women in class 3 are in critical need of targeted services to increase retention in care, as they are significantly more likely to experience disengagement.

Women who were older had higher odds of being retained in care and of being virally suppressed. This is consistent with other literature that indicates that PLW in general who are older (45 years of age or older) have higher viral suppression rates, compared with all other age groups (CDC, n.d.). Younger WLH may also experience challenges in maintenance of HIV care, suggesting that efforts should focus on assisting younger WLH with programs to increase timely linkage and retention in care (Muthulingam et al., 2013). Efforts should also focus on the needs of Black and Haitian WLH, as women who were non-Hispanic Black and Haitian had lower odds of viral suppression. Current strategies implemented may not take into account cultural and linguistic differences present in minority groups and may not be fully addressing the barriers to care faced by these populations. Findings also indicate that poverty and drug use were associated with lower odds of viral suppression and sustained viral suppression, and being anxious or depressed was also significantly associated with lower odds of sustained viral suppression, suggesting that interventions that do not address poverty, substance use, and mental health issues will fall short of current needs among low-income WLH.

Taken as a whole, our findings highlight the factors that drive poor HIV care outcomes among a diverse group of low-income WLH. In line with syndemic theories of understanding HIV risk and outcomes, understanding the multiple risk factors that contribute to adverse health outcomes can provide a deeper understanding of the experiences of WLH facing housing instability and homelessness. Syndemic theory refers to presence of two or more risk factors that impact an individual and whose interaction contributes to excess burden and disease among certain marginalized populations (Singer & Clair, 2003). Because of the complex nature of housing instability among WLH and its interrelatedness with multiple risk factors that place WLH at risk of both acquiring HIV and having adverse care outcomes, it is difficult to assess the causal nature of housing instability and homelessness. The identified characteristics and needs of WLH who were less likely to be retained in care and achieve sustained viral suppression, even after controlling for class membership (i.e., housing conditions), suggest the need to address compounded barriers including poverty, unemployment, mental health needs, substance use, and racial and ethnic disparities in access to and use of available treatment. Furthermore, researchers and practitioners should acknowledge and address how individual-level outcomes of retention in care and viral suppression are associated with sociodemographic, behavioral, and structural factors such as younger age, being Black and Haitian, poverty, and housing status. Findings suggest that intersections of race, gender, HIV status, and poverty may jointly affect health status and access to quality healthcare (Caiola et al., 2014). Thus, attention to the structural barriers that interact with individual-level attributes that prevent WLH from accessing or receiving quality care should be addressed and further examined.

Limitations

Limitations of this study should be noted. First, by nature of the data used for this study, the sample does not include WLH who were not engaged in care. Women who are not engaged in care may be more likely to be homeless or unstably housed and face compounded risk factors. Still, our sample is made up entirely of women who are under- or uninsured and low-income, which is a strength of the study’s contributions. Studies that aim to recruit women who are out of care are needed to better understand their experiences. A second limitation of the current study is that the data that were accessible for this study were self-reported for the purposes of care delivery and did not utilize validated measurements of substance use or mental health and could have resulted in underreporting of stigmatized topics. Next, this study used a crude measure of housing instability (e.g., living in transitional, residential, healthcare, or correctional facilities). Although there is no standard definition for housing instability, the term generally encompasses a range of experiences that also include moving frequently, staying in overcrowded homes, and spending the majority of income on rent (U.S. Department of Health and Human Services, n.d.). Thus, this study did not include all factors in understanding housing instability that have been used in other housing-related studies such as that by Brisson and Covert (2015). Future studies that include more details on multiple types of housing instability and how each of these types of housing instability affects care would be beneficial. Moreover, more details of the housing situations are critical to further understand and identify presentations of high-risk housing situations. Several limitations in the data available include the lack of detail on relationships with those with whom the women live and their specific roles and responsibilities in the household. Last, this is a cross-sectional study. Future longitudinal studies will be valuable to understand changes across time among housing conditions and whether these variables predict changes in risk for poor care outcomes over time.

Conclusion

This study extends the literature by providing a contextual understanding of the housing and household factors that play a role in treatment discontinuation and HIV care outcomes among a cohort of low-income, minority WLH. Although only a small proportion of individuals in the sample (5%) belonged to the most at-risk group (class 3), significant and persistent differences in HIV care outcomes suggest some women with unique housing patterns are in need of targeted interventions. These findings demonstrate the impact and importance of assessing housing status and characteristics as an important condition among WLH that adversely affects retention in care and viral suppression.

Women who are unstably housed, live alone, or with at least one other adult who does not know their HIV status are in need of strategies to increase retention in care and may benefit from strategies that address behavioral characteristics to improve ART adherence to achieve viral suppression. Furthermore, women who are living in extreme poverty, are in need of mental health and substance use services, or without a support system to rely on may be more likely to experience these housing situations. Interventions geared toward alleviating substance use, mental health needs, lack of social support, and housing needs are required to improve HIV care outcomes among some of our most vulnerable groups of minority WLH. Further suggestions include providing integrated care for WLH that includes social, behavioral, and mental health care. Additionally, more efforts to increase culturally responsive care strategies for Black and Haitian women are needed. Interventions that incorporate the input from these minority groups in program design and delivery are critical to adequately understand and address their unique needs and the existing barriers to current healthcare systems. Finally, next steps must include focusing policy efforts on present structural barriers including affordable and subsidized housing to minimize homelessness and adverse consequences associated with poverty.

 

This work was supported by Award Number R01MD013563-02S1 and in part by U54MD012398, K01MD013770, and F31MD015234-01 from the National Institute on Minority Health and Health Disparities at the National Institutes of Health.