Longitudinal associations between intersectional stigmas, antiretroviral therapy adherence, and viral load among women living with HIV using multidimensional latent transition item response analysis

Abstract

Background:

In the US, Women, especially Black and Latina women living in disadvantaged environments, are disproportionally affected by HIV. Women living with HIV (WLHIV) have higher rates of suboptimal antiretroviral therapy (ART) adherence, and detectable viral load (VL). Experiences of intersectional poverty, HIV, gender, and racial stigmas may increase the rates of detectable VL through suboptimal ART adherence.

Aims:

To explore longitudinal associations between intersectional stigmas, ART adherence, and detectable VL using multidimensional latent transition item response analysis.

Participants:

WLHIV (N = 459) in the [masked] sub-study of the [masked], from sites in Birmingham, AL, Jackson, MS, Atlanta, GA, and San Francisco, CA.

Assessment:

Experienced poverty, HIV, gender, and racial stigma, self-report ART adherence, and VL were assessed at four yearly follow-ups between 2016 and 2020.

Results:

We identified five classes of WLHIV with different combinations of experienced intersectional stigmas. Longitudinally, WLHIV with higher levels of poverty, gender, and racial stigma had higher odds of suboptimal ART adherence (<90%) (OR = 3.59, p < 0.001) and detectable VL (OR = 2.08, p = 0.028) compared to WLHIV with lower/moderately low stigmas levels. WLHIV in the highest stigma classes had higher odds of detectable VL, independently of ART adherence (Class 3: OR = 1.38, p = 0.016; Class 5: OR = 1.31, p = 0.046). These findings underscore the compounded effects of intersectional stigmas on HIV treatment outcomes.

Conclusion:

Intersecting experiences of HIV, racial, gender, and poverty stigmas can increase detectable VL risk through suboptimal ART adherence, although other mechanisms may also be involved. Recognizing the complexity of intersectional stigmas is essential for developing approaches to improve WLHIV’s HIV treatment outcomes.

1. Introduction

Women living with HIV (WLHIV), especially Black and Latina women living in socioeconomically disadvantaged environments, experience worse HIV treatment outcomes (i.e., poorer antiretroviral therapy [ART] adherence levels and higher rates of detectable viral load [VL]) (Geter et al., 2019; Crim et al., 2020; Bradley et al., 2019; Whittle et al., 2020), compared to men living with HIV. Psychosocial drivers of such disparities include HIV stigma and other co-occurring stigmas such as racism, sexism, and classism (Rice et al., 2018; Katz et al., 2013) that manifest as discrimination, unfair treatment, exclusion, and devaluation. Studies on stigma and HIV treatment outcomes have primarily focused on single forms of stigma. For example, studies have found significant associations between poverty stigma, HIV-related stigma, and HIV treatment outcomes (Sayles et al., 2009; Leddy et al., 2019; Turan et al.) This approach provides valuable but limited insights, as it overlooks the simultaneous effect of multiple stigmas on the HIV treatment outcomes of WLHIV (Rice et al., 2018; Leddy et al., 2019; Logie et al., 2011; Turan et al., 2019). Understanding the synergistic effect of co-occurring stigmas is critical for refining theories and developing interventions to address the unique challenges faced by WLHIV and ultimately improve HIV treatment outcomes.

Stigma has the effect of keeping people away and down (Link and Phelan, 2001). That is, people with stigmatized identities are excluded and devalued. Inclusion and being valued are two major fundamental needs, which when not met have important effects on well being (Baumeister et al., 2007; Leary, 1983). Different types of stigma may all contribute to thwarting these needs, but more research is needed before a theory can be developed to understand the unique and cumulative effects of different types of stigma in terms of twarthing these fundamental needs. Similarly, more research is needed before we can understand the intersectional effects of multiple types of stigma.

The intersectional stigma framework offers an optimal perspective to comprehend the simultaneous impact of co-occurring stigmas on HIV treatment outcomes. Intersectionality, a term and framework coined by Crenshaw, highlights the interconnected nature of social categories (e. g., racial identities) and systems of oppression (Turan et al., 2019; Crenshaw, 1991). The intersectional stigma framework is related but distinct, and emphasizing that multiple, intertwined social categories produce experiences of privilege and oppression. The intersectional stigma framework acknowledges that various stigmas, such as sexism, racism, and classism, can overlap to create unique lived experiences of unfair treatment, exclusion, and devaluation based on intersecting marginalized social identities (Crenshaw, 1991). Co-occurring stigmas interact in complex ways, shaping WLHIV’s lived experiences across various social contexts. For instance, WLHIV may face exclusion from social support networks due to poverty stigma, while experiencing devaluation or unfair treatment within healthcare settings due to racial and gender stigmas. Theoretically several factors shape the nature and extent of how stigma related to a certain identity or condition may affect health outcomes. These factors include features of stigmas: aesthetics, concealability, course, disruptiveness, origin, and peril (Pachankis et al., 2018). In addition, whether a person was born with the identity or not and to what degree identifies with the stigmatized group may be important predictors of the effect of a particular stigma (Pachankis et al., 2018; Yang et al., 2007). Unique combinations of experienced intersectional stigmas could potentially influence distinct HIV treatment outcomes via diverse mechanisms. However, more empirical research is needed to develop theories on the effects of these factors or dimensions of stigma, especially when they intersect.

WLHIV face intersectional stigmas arising from the confluence of multiple forms of discrimination and social marginalization related to gender, race/ethnicity, and socioeconomic status (Rice et al., 2018; Logie et al., 2011; Sangaramoorthy et al., 2017). These experiences have a profound, detrimental impact on WLHIV’s lives, leading to adverse downstream effects on mental health (e.g., depression and anxiety), and HIV treatment outcomes (Earnshaw et al., 2013a; Hook et al., 2023; Logie et al., 2013, 2019; Norcini Pala et al., 2022) through multiple mechanisms. For example, the Minority Stress Theory (Meyer, 2003) posits that chronic exposure to stressors such as discrimination and social exclusion adversely affects health outcomes through behavioral and physiological pathways. In the context of intersectional stigmas, WLHIV face a multitude of stigmas due to, for example, gender, race, and socioeconomic status, all of which adding to the list of stressors that they experience. Over time, these stressors can together affect health behaviors such as ART adherence. Experiences of discrimination may also undermine the patient-provider relationship, leading to medical mistrust (Turan et al., 2017, 2022). When WLHIV anticipate or experience discrimination in healthcare settings, they may avoid seeking care, miss appointments, or become non-adherent to ART, further exacerbating health disparities (Bogart et al., 2010; Earnshaw et al., 2013b). More research is needed before we can understand the unique and cumulative effects of different types of stigma as stressors.

Theoretically, intersectional stigma is still in its infancy. In a similar vein, intersectional theorists and methodologists are also still wrestling with different approaches for measuring and analyzing the inherent complexity of intersectional stigmas. Promising analytical approaches include Structural Equation Modeling (SEM) and Latent Class/Profile Analysis (LCA/LPA) (Turan et al., 2019; Logie et al., 2023). When used individually, however, these methods fail to capture the complex nature of these experiences. SEM models accurately estimate experienced intersectional stigmas accounting for measurement components (e.g., measurement error). But, unlike LCA, SEM does not account for non-linear effects of intersectional stigmas. Whereas LCA/LPA examine the non-linear relationship between experienced intersectional stigmas, they do not account for intersectional stigmas’ measurement components.

Integrating SEM and LCA is essential to capture the complex nature of intersectional stigmas and their effect on HIV treatment outcomes. SEM represents a variable-centered approach, while LCA employs a person-centered perspective. The variable-centered approach explores relationships among individual variables, such as intersectional stigmas and HIV treatment outcomes. In contrast, the person-centered approach aims to identify distinct groups of individuals characterized by unique combinations of intersectional stigmas. By integrating these two approaches, we establish an ideal framework for examining intersectional stigmas and HIV treatment outcomes. This framework accounts for measurement components of intersectional stigmas and non-linear relationships between intersectional stigmas and HIV treatment outcomes.

Finally, most studies on intersectional stigmas have relied on cross-sectional data, limiting our understanding of the dynamic nature of the experienced intersectional stigmas. Cross-sectional studies provide valuable insights into the complexity of intersecting stigmatized identities, but they represent a static snapshot of a dynamic phenomenon. The use of cross-sectional data does not shed light into the potential changes of intersectional stigma experiences and their downstream consequences on health outcomes. For example, an individual may report elevated levels of experienced HIV and racial stigma at a given time-point. A longitudinal approach might reveal a more complex narrative reflecting whether changes occur at multiple levels (e.g., intrapersonal, interpersonal, societal, structural). Consideration of time is critical in advancing our understanding of the dynamics of experienced intersectional stigmas, the impact on health trajectories, and identifying opportunities for interventions to alleviate health disparities among marginalized populations (Earnshaw et al., 2022). A longitudinal study employing Latent Class Growth Analysis investigated predictors of HIV stigma and everyday discrimination over time. This study revealed trajectories characterized by sustained low and consistently high stigma, which were linked with social factors and inequities (Logie et al., 2023). This emphasizes the importance of temporal perspectives in discerning nuanced patterns of stigmatization and its implications for the health outcomes of marginalized communities. This study examines the longitudinal impact of intersectional stigmas on HIV treatment out-WAVEcomes among a sample of women living with HIV in the US. Specifically, we aimed to 1) identify distinct groups of WLHIV based on their experiences with intersectional stigmas; 2) examine the dynamic transitions among these groups over time; 3) evaluate the group differences in HIV treatment outcomes; and 4) explore the mediating role of ART adherence in the association between intersectional stigma groups and viral load. To accomplish our aims, we integrated SEM, more precisely multidimensional item response theory (IRT), and the longitudinal version of LCA, namely latent transition analysis (LTA). We refer to this analytic approach as Multidimensional Latent Transition Item Response Theory (MLTIRT), which is the longitudinal extension of the existing multidimensional item response latent class model (Norcini Pala et al., 2022; Bartolucci et al., 2014).

2. Methods

2.1. Participants and procedures

In this study, a total of 459 WLHIV took part as participants. Women were enrolled in the Women’s Adherence and Visit Engagement (WAVE) sub-study of the Women’s Interagency HIV Study (WIHS) research project. Enrollment occurred at four distinct WAVE sites situated in Birmingham, Alabama; Jackson, Mississippi; Atlanta, Georgia; and San Francisco, California. Each participant provided written informed consent, and the study procedures received approval from the Institutional Review Boards at all participating sites. Data collection for this study took place between April 2016 and October 2020.

2.2. Measures

Experienced intersectional stigmas were measured through validated questionnaires assessing instances of discrimination, unfair treatment, exclusion, and devaluation in various social contexts over the past year.

To assess experienced poverty stigma, we used a 4-item subscale from the perceived poverty stigma survey (Mickelson and Williams, 2008). The items included statements such as People say negative or unkind things about me behind my back because of my financial situation, and I have been excluded from work, school, and/or family functions because of my financial situation. The items were rated on a 5-point scale ranging from Definitely Disagree to Definitely Agree. The subscale had good internal reliability in this sample (α = 0.87), with higher scores indicating a higher level of experienced poverty stigma.

Experienced HIV stigma was evaluated using a modified, 12 item, version of the experienced/enacted stigma subscale of the HIV Stigma Scale (Berger et al., 2001; Bunn et al., 2007). The items included experiences such as “People have avoided touching me if they know I have HIV”

The items were rated on a 4-point scale ranging from Strongly Disagree to Strongly Agree. In this study, the internal reliability of the scale was excellent (α = 0.94). Higher scores indicate greater levels of experienced HIV stigma.

We assessed experienced racial stigma using the 10-item Experiences of Discrimination scale (Krieger et al., 2005). We removed the item assessing racial stigma experiences in stores or restaurants because 1) it was highly correlated with the item assessing stigma in medical settings (r = 0.98); 2) it showed low endorsement (n = 11, 2.40%); and 3) its inclusion caused the model not to converge. Eight of the nine items assessed lifetime racial discrimination experiences on a 4-point scale (from never to four or more times), and one assessed the global experience of racism (i.e., How often do you feel that you, personally, have been discriminated against because of your race, ethnicity, or color?) on a 4-point scale (from Never to Often). The 9-item scale showed acceptable internal reliability in this study (α = 0.76). Higher scores correspond to higher racial stigma experiences.

Experienced sex-related stigma was measured using the 13-item Schedule of Sexist Events scale (Klonoff and Landrine, 1995). Participants were asked how often they experienced events such as being denied a raise, promotion, or job they deserved and being called sexist names. Each item was rated on a 4-point scale ranging from Never to Often. In this study, the scale internal reliability was good (α = 0.86). Higher scores indicate higher levels of experienced sex-related stigma.

Undetectable viral load was defined as fewer than 20 copies/mL, while a viral load of 20 or more copies/mL was considered detectable. Viral load was measured using a highly sensitive, automated real-time PCR assay that quantifies plasma HIV-1 RNA levels from blood samples drawn during the study.

ART adherence was assessed through a 3-item self-report measure (Wilson et al., 2016). The items assess, 1) the number of days in the last 30 days on which participants missed at least one dose of their medication; 2) how often participants took their medication as prescribed with five ordered response options including never, rarely, sometimes, usually, almost always, and always; and 3) participants’ ability to adhere to ART regimen over the past 30 days, rated on a 5-point scale including very poor, poor, fair, good, very good, or excellent. The item scores were transformed to a 0–100 scale, with zero as the poorest adherence level and 100 as perfect adherence level. We dichotomized the mean of the three items: 0 (≥90%) and 1 (<90%) (Byrd et al., 2019).

Sociodemographic characteristics included age, ethnicity, race, income, education, and substance use as covariates.

2.3. Data analysis plan

We used Latent Gold 6.1 (Vermunt and Magidson, 2013) and R Studio (Version 1.4.1717) (RStudio Team, 2020). Descriptive statistics included mean and standard deviation (SD) for normally distributed continuous variables, median and Interquartile Range (IQR) for non-normally distributed continuous variables, and count and percentage for nominal variables. To identify intersectional stigma latent classes, we used Multidimensional Latent Transition Item Response Theory (MLTIRT). Through MLTIRT, we 1) analyzed intersectional stigmas as latent traits, accounting for measurement error; 2) identified latent classes based on time-varying levels of intersectional stigmas; and 3) estimated participants’ probability to change classes over time. The measurement model consisted of four latent traits, corresponding to experienced poverty, HIV, gender, and racial stigma.

MLTIRT estimates the initial states or class membership at the first time-point and the probability of remaining in the same class or moving (transitioning) to a different class at the next follow-up. To clarify, in a hypothetical 2-class model, some participants who initially belong to Class 1 might stay in the same class or move to Class 2 at the next follow-up. Our study included four time points: T0 (used to determine the initial states), T1, T2, and T3. We found time-homogenous transition probabilities that did not change over time. Thus, we will refer to any given observation point as T and its preceding point as T-1.

In latent transition analysis, a class membership at T-1 predicts the class membership at T. Since class membership is a nominal variable, the transition probabilities are logit coefficients. Each class is used as a reference group to estimate the probability of moving from the reference group to a different class. For example, in a hypothetical 3-class model, we will have six coefficients, two for each class. The odds ratio (OR) calculated from the logit coefficients expresses the odds of staying in the same class (OR < 1) or moving to another class (OR > 1) over time.

To determine the optimal number of latent classes, we used the Bayesian Information Criterion (BIC), sample-adjusted BIC (aBIC), Akaike Information Criterion (AIC). Lower BIC, aBIC, and AIC indicate a better fit. The likelihood ratio test (LRT) was used to compare the fit of N-class with (N-1)-class model, wherein a significant p-value indicates a better fit of the N-class model. To estimate the precision with which each participant is assigned to a class, we used entropy, which ranges from 0 to 1 and values greater than 0.8 suggest accurate membership estimation. Model selection is also based on qualitative comparisons. Specifically, each additional class should differ qualitatively from the others, showing a unique pattern of intersectional stigmas (Jung and Wickrama, 2008; Nylund et al., 2007).

To examine the association between intersectional stigmas, VL and ART adherence, we used lagged Latent Markov Model (LMM) with logistic and multinomial regression and logit link. To estimate the LMM, we used the 3-step approach, which encompasses 1) the estimation of class membership probabilities using the observed data, 2) the assignment of participants to the most likely class (i.e., highest membership probabilities), and 3) refining the model using assignment feedback, an iterative process comparing the model predictions (group assignment; Step 2) with the actual data.

We initially performed the analysis using the low intersectional stigma class as our a-priori reference group to examine its association with ART adherence and VL. Subsequently, we employed the Wald test for pairwise comparisons to explore additional differences among all intersectional stigma classes. Every pair of latent classes were compared on both outcomes VL and ART adherence. We adjusted the p-values of the Wald tests for multiple comparisons using Benjamini-Hochberg correction (Chen, 2020).

We estimated the indirect effect of intersectional stigmas on VL through ART adherence using simulation-based, parametric bootstrapping with 10,000 replicates. Analyses were adjusted for age, income, race, ethnicity, and substance use and we implemented Full Information Maximum Likelihood to deal with missing data.

3. Results

3.1. Sample characteristics

Participants (N = 459; Table 1) had a mean age of 49.06 years. The majority identified as Black/African American, non-Latina, reported an income of $12,000 or less, and a High school/GED degree or lower as their highest level of education. On average, participants had been on ART for 103.74months (or 8.65years). At T0, nearly 30% of the participants had detectable VL. At T1 detectable VL was 34.86%, 38.99% at T2, and 22.66% at T3. The proportion of ART adherence <90% at T0 was 38.56% and remained overall stable over time with only small fluctuations (Table 1). Participants completely missing at follow-up were approximately 7% at T1, 13% at T2, and 24% at T3.

Table 1.

Sociodemographic characteristics of study participants.

	Overall
	N = 459
Age (mean [SD])	49.06 (9.45)
	383 (83.44)
Black/African-American, n (%)	25 (5.45)
Latina, n (%)
Highest level of education, n (%)
1. <High school/GED	129 (28.55)
2. High school/GED	141 (30.72)
3. Some college/Associate degree	142 (30.93)
4. 4 year College and above	39 (8.50)
Average yearly household income, n (%)
1. $12,000 or less	250 (54.47)
2. $12,001–24,000	99 (21.57)
3. $24,001–36,000	42 (9.15)
4. $36,001 or more	47 (10.24)
	103.74 (71.58)
Time on ART, months (mean [SD])	147 (32.02)
Any substance use at baseline n (%)
Detectable VL, n (%)
T0	136 (29.63)
T1	160 (34.86)
T2	179 (38.99)
T3	104 (22.66)
<90% ART adherence, n (%)
T0	177 (38.56)
T1	182 (39.65)
T2	185 (40.30)
T3	173 (37.69)

T0: Baseline assessment; T1-T3 yearly follow-up; ART: Antiretroviral Therapy; VL: Viral Load.

3.2. Intersectional stigma latent classes

We compared the fit indices of five nested models with latent classes ranging from two up to six (sTable 1). Although the 6-class model showed a comparatively better fit, one of the six classes was relatively small (6.59%; n ≈ 30) and showed a pattern similar to another class. Both classes had higher poverty and HIV stigma and in average lower levels of gender and racial stigma. Therefore, we chose the 5-class model, which consisted of unique intersectional stigma patterns (Fig. 1) with n greater than 10% in each class.

Fig. 1.

Profiles of the intersectional stigma latent class (N = 459).

Note: The values on the y axis are standardized latent trait scores, with mean = 0 and SD = 1; For each class, we reported the initial state n and %.

3.3. Characteristics of the intersectional stigma classes

Stigma scores consisted of standardized continuous latent trait scores with sample mean = 0 (SD = 1). We used zero as a reference to interpret the stigma levels within each class. Scores above 0 were considered moderately high (range 0 and 1) or high (range 1 and 2), whereas scores below 0 indicated moderately low (range − 1 and 0) or low (between − 2 and − 1) stigma levels. While this approach compared class stigma levels with the sample mean, it does not reveal between class stigma level differences. To this end, we used the Wald test to perform between-class comparisons of stigma scores.

Fig. 1 shows the different combinations of stigma types and levels by class. Using the sample average as the reference, we can group the five classes into two major groups: one with all stigma levels below the average (i.e., Class 1 and 2), and one with at least two stigma scores above the average (i.e., Class 3 through 5). In the following section, we describe the unique characteristics of each class with factor scores in parenthesis.

Class 1 (C1) – Low Intersectional Stigmas.

Low poverty (− 2.42) and HIV (− 2.96) stigma and moderately low gender (− 0.82) and racial (− 0.86) stigma.

C2 – Moderately Low Intersectional Stigmas.

All stigma scores ranged from 0 to − 1, indicating moderately low poverty (− 0.51), HIV (− 0.31), gender (− 1.17), and racial (− 1.02) stigma. Comparatively, C2 had significantly higher poverty and HIV stigma (p < 0.001) than C1, but no difference in gender and racial stigma (p > 0.05).

C3 – Moderately High Poverty, Gender, and Racial Stigma.

This class was characterized by moderately high poverty (0.53), gender (0.92), and racial (0.73) stigma and moderately low HIV stigma (− 0.30). The levels of poverty, gender, and racial stigma were significantly higher than C1 and C2 (ps < 0.001), whereas HIV stigma levels were significantly higher than C1 (p < 0.001) but not C2 (p > 0.05).

C4 – High HIV Stigma and Moderately High Poverty Stigma.

C4 showed high HIV stigma (1.646) and moderately high poverty stigma (0.50), but moderately low gender (− 0.39) and race (− 0.25) stigma. C4’s HIV stigma levels were significantly higher than C1, C2, and C3 (ps < 0.001). Poverty stigma levels were higher compared to C1 and C2 (ps < 0.001), but not C3 (p > 0.05). Although gender and racial stigma levels were below the mean, they were comparatively higher compared to C1, C2, and C3 (ps < 0.001).

C5 – High Intersectional Stigmas.

Stigma scores ranged from 1.39 to 1.91. C5’s levels for poverty, gender, and racial stigma were significantly higher compared to C1, C2, C3, and C4 (ps < 0.001). HIV stigma levels were significantly higher than C1, C2, and C3 (ps < 0.001) but not C4 (p > 0.05).

3.4. Latent transitions of intersectional stigma classes

The transition matrix in Table 2 shows that more than 50% of participants in each class had a high probability of staying in the same class over time. In the next sections, we report transition probabilities. The terms moved and stayed refer to the transition probability from any T-1 to T (i.e., the next follow-up).

Table 2.

Latent transition matrix of the intersectional stigma classes (N = 459).

Class T-1	Class T
	C1	C2	C3	C4	C5
C1	0.63	0.23	0.10	0.04	0.00
C2	0.13	0.68	0.08	0.10	0.01
C3	0.05	0.22	0.56	0.11	0.05
C4	0.05	0.16	0.08	0.55	0.15
C5	0.01	0.07	0.12	0.26	0.54

The proportion of women who stayed in the same class over time is shown in bold. T-1 is the previous time; there were four assessments: T0, T1, T2, and T3.

C1: More than 60% of women were likely to remain in C1 over time. More than 20% moved to C2, suggesting that they experienced a greater level of poverty and HIV stigma from T-1 to T. Approximately 10% of women in C1 moved to C3 or C4, indicating that women who experienced low intersectional stigmas (C1) had a low probability of facing increased levels of poverty, racial, and gender stigma or HIV and poverty stigma. The transition probability from C1 to C5 was zero namely, women who reported low intersectional stigmas did not experience high intersectional stigmas over time.

C2: Nearly 70% of women stayed in C2 over time, while some (13.1%) moved to C1, indicating a decrease in experienced HIV and poverty stigma levels. Less than 10% of women in C2 moved to C3 or C4, suggesting that few women reported an increase in poverty, gender, and race stigma or HIV and poverty stigma over time. The probability of moving from C2 to C5 was 0.8%, suggesting that women with moderately low experience of intersectional stigmas were unlikely to face high intersectional stigma levels over time.

C3: More than half of women stayed in C3 over time, while approximately 20% moved to C2, suggesting a decrease in experienced poverty, gender, and racial stigma over time. A small percentage (11.00%) went from experiencing moderately high poverty, gender, and racial stigma (C3) to high HIV and moderately high poverty stigma (C4). Notably, women in C3 have an equal probability of reporting a significant decrease (C1, 5.30%) or increase (C5, 5.31%) in intersectional stigma levels over time.

C4: More than 55% of women stayed in C4 over time, whereas approximately 15% were likely to experience increased (C5) or decreased (C2) levels of intersectional stigmas. Fewer women moved from C4 to C3 (8.00%; moderately high poverty, gender, and racial stigma) or to C1 (5.4%; low intersectional stigmas).

C5: More than half of women in C5 were likely to stay in C5 over time, while more than a fourth (26.21%) moved to C4, suggesting a significant reduction in poverty stigma levels and a steeper decrease in gender and racial stigma. Approximately 12% of women who reported elevated intersectional stigmas (C5) experienced a moderate decrease in poverty, gender-related-, and racial stigma and a steeper reduction in HIV stigma levels (C3). Less than 7% of women in C5 moved reported a dramatic decrease in experienced intersectional stigmas over time (i.e., from C5 to C2 or C1).

Overall, intersectional stigma experiences are stable over time, though a small proportion of women experience increased or decreased levels of intersectional stigmas or different patterns of experienced intersectional stigmas.

3.5. Intersectional stigmas and HIV treatment outcomes

Through lagged Latent Markov Model (LMM), we examined the relationship between intersectional stigmas, VL, and ART adherence at the initial state and over time simultaneously. Over time, coefficients correspond to the average effects across all time points. C1 (low intersectional stigma) was the intersectional stigma reference group.

Initial state – Table 3. Compared to C1, women in C3 had higher odds of detectable VL (OR = 2.08, p = 0.03). In addition, the odds of ART adherence <90% were significantly higher in C3 (OR = 3.59, p <0.001), C4 (OR = 2.12, p = 0.03), and C5 (OR = 2.81, p = 0.003), but not in C2 (OR = 1.74, p = 0.10). ART adherence <90% was associated with higher odds of detectable VL (OR = 2.25, p < 0.001).

Table 3.

Initial state associations between intersectional stigmas, VL, and ART adherence (N = 459).

	Detectable VL		<90% ART adherence
	OR (95% CI)	p	OR (95% CI)	p
C1(Ref. group)
C2	1.31 (0.67, 2.56)	0.43	1.74 (0.90, 3.37)	0.10
C3	2.08 (1.08, 3.98)	0.03	3.59 (1.93, 6.68)	<0.001
C4	1.64 (0.83, 3.22)	0.15	2.12 (1.08, 4.14)	0.03
C5	1.33 (0.66, 2.71)	0.43	2.81 (1.43, 5.50)	0.003
<90% ART adherence	2.25 (1.46, 3.48)	<0.001	–	–

Class 1–5 (C1–5); Adjusted for age, income, race, ethnicity, and substance use; OR: Odds ratio; ART: Antiretroviral Therapy; VL: Viral Load.

Over time – Table 4. The odds of detectable VL at T were higher for women in C5 at T-1 (OR = 1.50, p = 0.06), though the p-value did not reach statistical significance. The odds of ART adherence <90% at T were significantly higher among women in C3 (OR = 1.91, p = 0.002) and C5 (OR = 1.71, p = 0.02) at T-1. Finally, ART adherence <90% at T-1 significantly predicted higher odds of detectable VL at T (OR = 1.65, p < 0.001).

Table 4.

Longitudinal associations between intersectional stigmas, VL, and ART adherence (N = 459).

	Detectable VL [T]		<90% ART adherence [T]
	OR (95% CI)	p	OR (95% CI)	p
C1[T-1](Ref. group)
C2[T-1]	0.79 (0.53, 1.17)	0.24	1.06 (0.72, 1.56)	0.76
C3[T-1]	1.11 (0.75, 1.64)	0.60	1.91 (1.27, 2.86)	0.002
C4[T-1]	0.92 (0.61, 1.39)	0.69	1.35 (0.89, 2.06)	0.16
C5[T-1]	1.50 (0.98, 2.31)	0.06	1.71 (1.10, 2.67)	0.02
ART adherence <90%[T-1]	1.65 (1.27, 2.15)	<0.001	–	–

Class 1–5 (C1–5); Adjusted for age, income, race, ethnicity, and substance use; OR: Odds ratio; ART: Antiretroviral Therapy; VL: Viral Load.

Indirect associations:

Due to the higher odds of ART adherence <90%, women in C3 and C5 at T-1 also had higher odds of detectable VL at T (OR = 1.38, p = 0.02 and OR = 1.31, p = 0.05, respectively). The estimated total effects were OR = 1.53 (p = 0.06) for women in C3 and OR = 1.97 (p = 0.007) for women in C5.

3.6. Pairwise comparisons

Initial state

None of the adjusted p-values of the pairwise comparisons reached statistical significance.

Over Time

The Wald tests, with adjusted p-values, indicated significant differences in VL between C2 and C5 (p.adj = 0.01) and in ART adherence between C2 and C3 (p.adj = 0.008), and C2 and C5 (p.adj = 0.05). With C2 as the reference group (Table 5), we found significantly higher odds of detectable VL among women in C5 (OR = 1.91, p = 0.003) and marginally in C3 (OR = 1.41, p = 0.05) at T-1. Likewise, women in C3 and C5 at T-1 had higher odds of <90% ART adherence at T (OR = 1.80, p < 0.001 and OR = 1.61, p = 0.02, respectively). Indirect associations: Due to the increased odds of <90% ART adherence, the odds of detectable VL at T were significantly higher among women in C3 (OR = 1.34, p = 0.01) and, marginally, in C5 (OR = 1.27, p = 0.06). The estimated total effects were OR = 1.88 (p = 0.003) for women in C3 and OR = 2.42 (p < 0.001) for women in C5.

Table 5.

Longitudinal associations between intersectional stigmas, VL, and ART adherence (N = 459).

	Detectable VL [T]		<90% ART adherence [T]
	OR (95% CI)	p	OR (95% CI)	p
C2[T-1](Ref. group)
C1[T-1]	1.27 (0.86, 1.87)	0.24	0.94 (0.65, 1.37)	0.75
C3[T-1]	1.41 (1.00, 1.99)	0.05	1.80 (1.29, 2.50)	<0.001
C4[T-1]	1.17 (0.80, 1.70)	0.42	1.27 (0.89, 1.82)	0.19
C5[T-1]	1.91 (1.25, 2.90)	0.003	1.61 (1.08, 2.41)	0.02
ART adherence < 90% [T-1]	1.65 (1.29, 2.11)	<0.001	–	–

Class 1–5 (C1–5); Adjusted for age, income, race, ethnicity, and substance use; OR: Odds ratio; ART: Antiretroviral Therapy; VL: Viral Load.

4. Discussion

Due to multiple stigmatized identities, women living with HIV face a multitude of types of oppression and discrimination, ultimately affecting their HIV treatment outcomes. A large proportion of WLHIV identify as Black/African-American and/or Latina and live in poor socioeconomic conditions, exposing them to stigma based on race and poverty. Thus, in addition to HIV stigma, WLHIV may experience various intersections of gender, poverty, and racial stigmatization. Understanding the unique experience of intersecting forms of stigmatization, and its impact on HIV treatment outcomes is critical to inform tailored interventions for WLHIV, recognizing them as whole people with multiple identities. Specifically, different combinations of intersectional stigmas may operate through various pathways to affect HIV treatment outcomes. Tailored interventions to address such pathways could ultimately improve HIV outcomes among WLHIV.

In this study, we examined intersectional stigmas using MLTIRT, an innovative analytic approach to identify different patterns of experienced stigma types and intensity while accounting for the dynamic nature of such experiences. Experienced intersectional stigmas may vary or remain stable over time. We identified five distinct classes representing groups of WLHIV with unique combinations of type and intensity of intersectional stigmas. Class 1 (C1) represents women with low overall intersectional stigmas, while C2 has moderately low stigmas, specifically higher levels of poverty and HIV stigma. C3 is characterized by high levels of poverty, gender-based, and racial stigma, with moderately low HIV stigma. Differences in HIV treatment outcomes between C1 and C3 may be linked to higher poverty, gender, racial, and HIV stigma in C3. C2 and C3 differences are mainly attributed to elevated poverty, gender, and racial stigma in C3. C4 women have above-average HIV and poverty stigma levels, reporting higher stigmas overall compared to C1 and C2. C3 and C4 distinctions are marked by gender-based, racial, and HIV stigmas. C5 women experience the most intense levels of all stigmas, surpassing the sample mean and the other four classes. Notably, HIV stigma appears distinct from other forms of stigma in this study, as it is the only stigma type with a class (C4) where HIV stigma is high while other stigmas, such as poverty, gender-based, and racial stigma, remain moderate or low. This distinction might suggest that HIV stigma operates differently within certain groups of WLHIV, reflecting unique patterns of stigma experiences.

4.1. Transition probabilities

Taking into account the dimension of time in stigma research provides valuable insights into the evolving nature of stigma experiences and their associated outcomes. Researchers in the field have emphasized the need for investigations that consider temporal dynamics (Earnshaw et al., 2022). The transition probabilities we examined between classes shed light on the dynamics and temporal changes in experienced intersectional stigmas. Our findings suggest that, overall, the majority of women exhibited stable patterns of intersectional stigmas over time, remaining in the same class at each time point. This observation aligns with prior research in Canada, which uncovered the chronicity of HIV-related stigma and everyday discrimination over time among WLHIV (Logie et al., 2023). This underscores the importance of acknowledging how stigma associated with a chronic condition, such as HIV infection, may persist over time. Notably, the probability of remaining in the same class was higher for WLHIV in the first two classes with low or moderately low intersectional stigma levels. Participants in these classes had a low likelihood of transitioning to classes characterized by higher intersectional stigma levels. Conversely, women with at least two stigma levels above the average demonstrated a more complex transition probability pattern. Integrating temporal considerations into stigma research is crucial, not only for understanding the evolving nature of stigma experiences but also for informing effective interventions tailored to specific times and circumstances. Over time, the way stigma is experienced, and its resulting outcomes, undergo changes.

4.2. Intersectional stigmas and HIV treatment outcomes

The initial state associations shed light on the starting points of the longitudinal underlying processes. Interpreting only the over time, lagged coefficients may limit our understanding of the impact that intersectional stigmas have on HIV treatment outcomes. Initial state associations are particularly relevant when the underlying processes are stable (i.e., limited transitions between classes over time). More concretely, our findings suggest that women who experienced at least two types of stigmas with levels above the average are likely to report suboptimal adherence. For example, the odds of suboptimal adherence at the initial state were 3.587-fold higher for WLHIV, who experienced moderately high levels of poverty, gender, and racial stigma compared to their counterpart with overall low levels of intersectional stigmas (C1). Based on the transition probabilities, more than 50% of WLHIV who faced elevated levels of poverty, gender, and racial stigma consistently reported similar experiences over time. Chronic exposure to such a pattern of intersectional stigmas likely results in a consistently higher likelihood of suboptimal adherence. This reasoning may apply to WLHIV, who reported higher HIV stigma and poverty stigma and elevated levels of all intersectional stigmas. At the initial state and over time, ART adherence <90% predicted detectable VL. Hence, at each time point, WLHIV who experienced higher levels of at least two intersectional stigmas are likely to show suboptimal ART adherence and, consequently, increased risk for detectable VL.

In addition to this “initial vulnerability level,” WLHIV who faced poverty, gender and racial stigma, or all intersectional stigmas at a given time have higher odds of suboptimal ART adherence after a 12-month period. Furthermore, when compared with women who experienced moderately low levels of intersectional stigmas (C2), WLHIV who reported higher levels of poverty, gender, and racial stigma (C3), or all intersectional stigmas (C5) had a higher likelihood of detectable VL at the next follow-up independently of ART adherence.

Compared to previous studies that focused on individual forms of stigma, such as HIV-related stigma or racism in isolation, our study provides findings and clinically meaningful insights into the complex interplay of intersectional stigmas. Unlike the “single-stigma” studies, we were able to capture the cumulative burden of multiple intersecting stigmas experienced by WLHIV in a holistic fashion. Our findings provide empirical evidence that stigmas related to HIV, race, gender, and socioeconomic status operate simultaneously to exacerbate health disparities. While the mechanisms explaining the associations found in our study require further evaluation, our findings can help identify WLHIV who are more likely to be non-adherent to ART and have detectable VL. Sub-optimal ART adherence is undoubtedly a critical mechanism through which intersectional stigmas increase the vulnerability of WLHIV to detectable VL. Our findings, however, suggest self-reported adherence does not fully explain the adverse effect of intersectional stigmas on HIV treatment outcomes. Theoretical models, such as Minority Stress Theory (Meyer, 2003), offer a potential framework for understanding our findings by suggesting that chronic exposure to stressors like discrimination and social exclusion could lead to adverse health outcomes. Our study provides a foundation for further exploration into additional pathways beyond ART adherence, including both behavioral and physiological mechanisms that may contribute to HIV treatment outcomes disparities. Although these hypotheses remain speculative, they are worth exploring as they can contribute to the development of comprehensive and tailored approaches to address intersectional stigmas and improve WLHIV health outcomes.

Our results emphasize the urgent need for interventions for WLHIV aimed at mitigating the impact of intersectional stigmas on poorer HIV treatment outcomes, as documented in this study. Tailored interventions should be specifically designed for WLHIV who face intersectional stigmas, considering the unique patterns identified within different classes. For class 1 (C1) and class 2 (C2), where women reported low or moderately low intersectional stigmas, interventions may focus on further reducing stigma levels, especially poverty and HIV stigma. Class 3 (C3), characterized by high poverty, gender, and racial stigma, suggests the need for interventions addressing elevated levels of these specific stigmas. Women in class 4 (C4) experiencing high levels of HIV stigma and moderately high poverty stigma may benefit from interventions addressing both HIV− and poverty-related stigmas. Lastly, class 5 (C5), representing the most intense levels of all stigmas, requires comprehensive and intensive interventions across various dimensions of stigma.

We strongly believe that our findings have important implications at both the research and practice levels. For example, mixture models such as MLTIRT, which capture the unique combinations of stigmas faced by WLHIV over time, can overcome the limitations of variable-centered approaches that fail to account for individual differences. In addition, MLTIRT captures the dynamic nature of experienced intersectional stigmas. Our findings emphasize the need for healthcare professionals to receive comprehensive training on acknowledging, recognizing and addressing multiple forms of stigmas. This training should go beyond HIV stigma and encompass racial, gender, and socioeconomic stigmas to ensure that healthcare providers are equipped to offer holistic care. Research has shown that recognizing and validating these intersecting stigmas in healthcare settings can reduce the impact of medical mistrust and improve ART adherence.

From a public health policies perspective, our study findings stress the need for strategies to ensure equitable access to healthcare and social services for WLHIV from marginalized racial and socioeconomic groups. They encourage policy-makers to use intersectional approaches when designing interventions, improving healthcare systems’ response to the unique challenges faced by those who experience multiple forms of discrimination.

Finally, we believe that our study findings suggest to move beyond ART adherence and developing multidisciplinary care models that address the psychological and social impacts of stigmas. For example, integrating mental health services, social support, and legal aid (for housing or employment discrimination) could help address the broader consequences of stigmas on WLHIV’s well-being.

4.3. Limitations and offsetting strengths

Our study is not without limitations. Specifically worth noting are limitations regarding social desirability bias in reporting and the inability to determine causality. Participants may have provided responses that they perceived as socially desirable. For example, they may overstate their adherence to ART or underreport stigmatizing experiences due to concerns about judgment or stigma itself. However, our inclusion of viral load as an objective measure offset the potential for self-report bias in ART adherence reporting, and the use of anonymous data collection methods may have encouraged participants to provide more honest responses regarding stigmatizing experiences and ART adherence. Secondly, while the study identifies associations between intersectional stigmas, ART adherence, and viral load, this study cannot establish causality. It is possible that unmeasured variables or external factors could be influencing these relationships. This limitation is offset by the fact that the study’s longitudinal design allows for the examination of temporal associations, which, while not establishing causality, can provide strong evidence of relationships between variables. Finally, it’s important to acknowledge that the long-running nature of the cohort might lead to a lack of representation for women who have faced significant stigma, resulting in disengagement from care or weak participation in research studies. This potential bias could affect the generalizability of the findings to the larger population of women living with HIV. Specifically, the omission of women with limited engagement in care may bias results toward underestimating the impact of stigma.

4.4. Implications and conclusion

These findings have implications for the development and implementation of interventions that are designed to improve HIV treatment outcomes among women living with HIV experiencing intersectional stigma. The identification of distinct classes, each representing unique combinations of intersectional stigma types and intensities, provides a foundation for targeted interventions tailored to the specific needs of each group. The transition probabilities highlight the necessity of longitudinal monitoring to identify shifts in stigma experiences, guiding timely interventions to prevent escalation and improve overall mental and physical health outcomes for WLHIV. The insights gained underscore the urgent need for interventions that account for the intersectional stigma experienced by WLHIV, offering tailored and comprehensive strategies to mitigate adverse effects on HIV treatment outcomes and contribute to the overall well-being of women living with HIV.

Data availability

Data will be made available on request.