Introduction
Despite Canada’s considerable progress on sexual minority rights over the past two decades (Rau, Reference Rau2021), discriminatory attitudes against gay, bisexual and other men who have sex with men (GBM) persist (Government of Canada, 2023c; Hyman et al., Reference Hyman, O’Campo, Ansara, Siddiqi, Forte, Smylie, Mahabir and McKenzie2019). Biopsychosocial models and minority stress theory conceptualize discrimination as a social stressor (Bogart et al., Reference Bogart, Wagner, Galvan, Landrine, Klein and Sticklor2011; Brondolo et al., Reference Brondolo, Blair, Kaur, Major, Dovidio and Link2018; Meyer, Reference Meyer2003). Compared with heterosexual men, heightened stress in GBM, secondary to societal homonegativity, triggers maladaptive psychological pathways that increase the risk of depression, anxiety, substance use and suicidality (Hatzenbuehler, Reference Hatzenbuehler2009). This may contribute to GBM using mental health services (MHS) more frequently relative to heterosexual men (Hart et al., Reference Hart, Sharvendiran, Chikermane, Kidwai and Grace2023; Marbaniang et al., Reference Marbaniang, Rose, Moodie, Hart and Cox2022; Platt et al., Reference Platt, Wolf and Scheitle2018; Tjepkema, Reference Tjepkema2008), as a means to cope with negative psychological states that result from sexual minority stress (Giwa and Han, Reference Giwa and Han2022).
Research on MHS use among GBM has typically relied upon data from predominantly white samples (Ferlatte et al., Reference Ferlatte, Salway, Oliffe and Trussler2017; Kulick, Reference Kulick2013). However, GBM in Canada are not a racially monolithic group (Government of Canada, 2023b). In addition to homonegativity, GBM racialized as non-white encounter racism both from mainstream society at large and other GBM (Choi et al., Reference Choi, Paul, Ayala, Boylan and Gregorich2013; Giwa and Han, Reference Giwa and Han2022; Jackson et al., Reference Jackson, Mohr, Sarno, Kindahl and Jones2020; McConnell et al., Reference McConnell, Janulis, Phillips, Truong and Birkett2018; Sadika et al., Reference Sadika, Wiebe, Morrison and Morrison2020). Like homonegativity, racism negatively affects mental health (Brondolo et al., Reference Brondolo, Brady Ver Halen, Pencille, Beatty and Contrada2009; Paradies et al., Reference Paradies, Ben, Denson, Elias, Priest, Pieterse, Gupta, Kelaher and Gee2015). In the general population, racialized individuals tend to underutilize MHS and, instead, are more likely to engage fellow racialized community members for emotional support (Blumberg et al., Reference Blumberg, Clarke and Blackwell2015; Boukpessi et al., Reference Boukpessi, Kpanake and Gagnier2021; Chiu et al., Reference Chiu, Amartey, Wang and Kurdyak2018; Richman et al., Reference Richman, Kohn-Wood and Williams2007). Unfortunately, many racialized GBM experience a disavowal of their sexual identity from members of their racialized communities (Sadika et al., Reference Sadika, Wiebe, Morrison and Morrison2020), which limits access to stress-moderating emotional support (Bowleg, Reference Bowleg2013; Kim and Allen, Reference Kim and Allen2023; Nakamura et al., Reference Nakamura, Chan and Fischer2013). Thus, racialized GBM may have increased mental health care needs (Choi et al., Reference Choi, Paul, Ayala, Boylan and Gregorich2013; English et al., Reference English, Rendina and Parsons2018; Jackson et al., Reference Jackson, Mohr, Sarno, Kindahl and Jones2020; Layland et al., Reference Layland, Maggs, Kipke and Bray2022). However, it is unclear whether MHS use patterns in racialized GBM align more with white GBM or with trends observed in racialized communities.
In Canada, GBM represent one-half of people living with HIV (Government of Canada, 2022). HIV stigma represents an additional stressor for GBM living with HIV (GBMHIV) (Rzeszutek et al., Reference Rzeszutek, Gruszczyńska, Pięta and Malinowska2021; Smit et al., Reference Smit, Brady, Carter, Fernandes, Lamore, Meulbroek, Ohayon, Platteau, Rehberg, Rockstroh and Thompson2012) which can also contribute to poor mental health. For example, a study of 671 Canadian GBMHIV found HIV stigma to be associated with higher suicidality (Ferlatte et al., Reference Ferlatte, Salway, Oliffe and Trussler2017). For racialized GBMHIV, homonegativity, including culturally rooted heteronormative norms; racism; and HIV stigma may all coexist and intersect in different configurations in the social milieus they navigate (Arnold et al., Reference Arnold, Rebchook and Kegeles2014; Arscott et al., Reference Arscott, Humphreys, Merwin and Relf2020; Bogart et al., Reference Bogart, Wagner, Galvan, Landrine, Klein and Sticklor2011; MacCarthy et al., Reference MacCarthy, Bogart, Galvan and Pantalone2021).
Intersectionality is a critical theoretical framework developed by Black feminists (Crenshaw, Reference Crenshaw1989; Hill Collins, Reference Hill Collins2019). It focuses on neglected groups whose experiences are often essentialized to those of a larger dominant group (Bowleg, Reference Bowleg2012). Intersectionality examines how the interrelatedness and mutual construction of multiple systems of inequity (e.g., homonegativity, racism, HIV stigma) shape the experiences of people living at the intersections of such systems (Bowleg, Reference Bowleg2013; Hill Collins, Reference Hill Collins2019). As an analytical strategy, it diverges from methods that treat social identities of individuals (e.g., sexual identity, racial identity, living with HIV) as additive (i.e., mutually exclusive), ahistorical and socially invariant (Bowleg, Reference Bowleg2013; Cho et al., Reference Cho, Crenshaw and McCall2013; Hancock, Reference Hancock2007; Nayak, Reference Nayak, Trachsel, Gaab, Biller-Andorno, Tekin and Sadler2021). It also refrains from using identities as determinants of outcomes. Instead, intersectionality emphasizes the role of power or its surrogates (e.g., discrimination) to dynamically accord identities certain social positions, consequently affecting outcomes (Hill Collins, Reference Hill Collins2019; Sievwright et al., Reference Sievwright, Stangl, Nyblade, Lippman, Logie, Veras, Zamudio-Haas, Poteat, Rao, Pachankis, Kumi Smith, Weiser, Brooks and Sevelius2022).
Many studies on Black GBM in the US have leveraged intersectionality to better understand their disengagement from HIV care (Arnold et al., Reference Arnold, Rebchook and Kegeles2014; Arscott et al., Reference Arscott, Humphreys, Merwin and Relf2020; Lutete et al., Reference Lutete, Matthews, Sabounchi, Paige, Lounsbury, Rodriguez, Echevarria, Usher, Walker, Dickerson, Hillesheim and Frye2022). However, few studies have used an intersectionality-based framework to evaluate MHS use in GBM. MHS represent avenues through which discrimination-related stress and its negative effects on mental health (Moody et al., Reference Moody, Browning, Hossain and Clay2023) and on HIV care of GBM could be mitigated.
In this manuscript, our objectives are (1) to compare MHS use in different GBM groups (based on racial identity and HIV status) and (2) to estimate the mediating effect of self-reported interpersonal perceived discrimination (PD) between GBM groups and MHS use, using a quantitative intersectionality-based approach. Different quantitative methods have been used in intersectionality research (Bauer et al., Reference Bauer, Churchill, Mahendran, Walwyn, Lizotte and Villa-Rueda2021). We utilize a modification of causal mediation analysis, the three-way decomposition of causal effects (VanderWeele, Reference VanderWeele2013), as proposed by Bauer and Scheim (Reference Bauer and Scheim2019), to meet our objectives.
Methods
Study population
We used baseline data (February 2017 to August 2019) from Engage, a prospective cohort study of GBM in Montreal, Toronto and Vancouver. GBM (cisgender and transgender) ≥16 years who reported sexual activity with another man in the previous 6 months and provided informed consent in French or English were enrolled. Participants were recruited by respondent-driven sampling (RDS), an adapted form of the chain referral method, in which peer referrals and self-reported network sizes of participants are recorded (Gile et al., Reference Gile, Beaudry, Handcock and Ott2018). Engage study details have been described in detail elsewhere (Cox et al., Reference Cox, Apelian, Moodie, Messier-Peet, Hart, Grace, Moore, Lachowsky, Armstrong, Jollimore, Skakoon-Sparling, Rodrigues, Tan, Maheu-Giroux, Noor, Lebouché, Tremblay, Olarewaju and Lambert2021; Hart et al., Reference Hart, Moore, Noor, Lachowsky, Grace, Cox, Skakoon-Sparling, Jollimore, Parlette, Lal, Apelian, Sang, Tan and Lambert2021; Moore et al., Reference Moore, Cui, Skakoon-Sparling, Sang, Barath, Wang, Lachowsky, Cox, Lambert, Noor, Grace, Jollimore, Apelian, Lal, Parlette and Hart2021).
At study visit, all participants completed computer-assisted self-administered questionnaires and underwent tests for HIV and other sexually transmitted and blood borne infections.
Study definitions
Exposure was categorized into four groups. Group 1: white HIV-negative GBM, Group 2: white GBMHIV, Group 3: racialized HIV-negative GBM and Group 4: racialized GBMHIV. We operationalize the definition of racialization as used by the Government of Canada (2023a). Racialized GBM included Indigenous men, those listed as visible minorities under Canada’s Employment Equity Act (South Asian, Chinese, Korean, Japanese, Black, Filipino, Arab, Latin American, Southeast Asian and West Asian; Government of Canada, 1995) and those self-identifying as persons of colour.
Our categorization recognizes that homonegativity, racialization and HIV stigma are social phenomena which differentially affect the social positions of members of the exposure groups. Further, it emphasizes that the experiences of racialized GBMHIV cannot be understood as the sum of experiences of racialized HIV-negative GBM and white GBMHIV.
The mediator was PD scores measured as continuous values on a short version of the Everyday Discrimination Scale (EDS; Williams et al., Reference Williams, Yan, Jackson and Anderson1997), previously used in a national Canadian survey (Hyman et al., Reference Hyman, O’Campo, Ansara, Siddiqi, Forte, Smylie, Mahabir and McKenzie2019). In the first half of the EDS, PD is measured using five questions not bound to a timeframe and without attribution to sexual identity, racial identity or/and living with HIV. Questions range from an assessment of minor daily hassles (i.e., being treated with less respect or courtesy) to more consequential experiences (i.e., being threatened or harassed). Responses are scored on a 5-point Likert scale (1 – never, 5 – at least once a week). Total scores (range: 5–25) are calculated by summing responses over the five questions. Scores are directly proportional to the chronicity of PD (Lewis et al., Reference Lewis, Cogburn and Williams2015). Since the score is attribution-free, we posited that the total score encapsulates the interconnectedness of discriminatory experiences. We used the total score for quantitative analyses. In the second half of the EDS, participants indicate all identities to which they attribute discrimination. We counted attributions to sexual identity, racial identity and HIV for each exposure group and examined these descriptively but not in quantitative analyses.
The outcome was MHS use in the prior 6 months, determined as a binary variable (yes/no). MHS use was defined as having seen or talked to an MHS provider (physician, psychologist, psychiatrist, social worker and community-based counsellor) about one’s mental or emotional health.
Statistical analyses
We excluded GBM with missing data for the outcome (n = 63) or those diagnosed with HIV in the enrolment year or for the first time through the study (n = 15). The latter were excluded assuming their experiences of HIV stigma would be dissimilar to GBM who had lived longer with HIV.
Study population characteristics across the three cities were described with and without RDS adjustment. This adjustment accounts for participants’ self-reported peer network sizes and is city specific. It uses inverse probability weights to account for the likelihood of being sampled (i.e., to account for the higher probability of those with large network sizes being sampled, therefore, preventing selection bias from unduly affecting estimates) (McCreesh et al., Reference McCreesh, Frost, Seeley, Katongole, Tarsh, Ndunguse, Jichi, Lunel, Maher, Johnston, Sonnenberg, Copas, Hayes and White2012). RDS-unadjusted estimates are presented whenever data combined across the three cities (i.e., three-city data) are used.
For the causal mediation analysis, three-city data were used. Conventionally, the experiences of white HIV-negative GBM (Group 1) are generalized to most GBM, particularly in research unrelated to HIV (Bowleg, Reference Bowleg2013; Kulick, Reference Kulick2013; Sadika et al., Reference Sadika, Wiebe, Morrison and Morrison2020). Here, Group 1 serves as the reference to facilitate the evaluation of results from the vantage point of underrepresented and differentially disprivileged GBM groups.
To fit a causal mediation model on cross-sectional baseline data, we made two assumptions regarding the temporality of measured variables. First, we assumed that sexual identity, racial identity and HIV status preceded PD (mediator). Due to being in an inequitable society, homonegativity, racism and HIV stigma are consequences of GBM identity, racialization and living with HIV. Thus, we believe our first assumption is valid. Second, we assumed that for most participants, PD (not time-bound) preceded the outcome and remained relatively unchanged during the assessment period of the outcome (prior 6 months). To our knowledge, there were no sociopolitical events in Montreal, Toronto and Vancouver 6 months prior to enrolment (i.e., between 2016 and 2019) that may have caused a significant change in average PD. Therefore, we believe that our second assumption is reasonable.
Traditional causal mediation analysis decomposes the total effect (TE) of the exposure on the outcome into a pure direct effect (PDE, unmediated effect) and a total indirect effect (TIE, mediated effect) (VanderWeele, Reference VanderWeele2013). This assumes that the mediator has the same effect on the outcome for all exposure groups. Discrimination is not experienced independent of historical and social contexts. Hence, PD may have different meanings and effects for exposure groups. The three-way decomposition further separates the TIE into a pure indirect effect (PIE) and a mediated interaction effect (MIE). Statistically, this introduces an interaction between the exposure and the mediator. This allows PD to have heterogeneous effects on the outcome for different exposure groups (Bauer and Scheim, Reference Bauer and Scheim2019). Interpretations for PDE, PIE, MIE and TE are presented in Table 1.
Table 1. Interpretations of three-way decomposition of mediated effect estimates with a binary outcome
Source: Adapted from Bauer and Scheim (Reference Bauer and Scheim2019).
a Application examples are provided using white HIV-negative GBM (Group 1) as the reference exposure group and racialized HIV-negative GBM (Group 3) as the comparator exposure group. Interpretations for white GBM living with HIV (white GBMHIV, Group 2) and racialized GBMHIV (Group 4) can be made similarly.
We used the imputation-based approach of natural effect models to operationalize the three-way decomposition (Vansteelandt et al., Reference Vansteelandt, Bekaert and Lange2012). This method estimates PDE and PIE directly without resorting to complex formulae, used in traditional three-way decomposition. EDS scores were standardized by mean-centring and dividing by the standard error of the scores’ distribution. We obtained odds ratios (ORs) for PDE, PIE, MIE and TE, comparing different exposure groups to Group 1, after controlling for confounders. Confounders were age, having a lifetime chronic mental health condition diagnosed >6 months ago (yes/no), Canadian citizenship (yes/no), being cisgender (yes/no) and city of enrolment. Conventionally, confounders are chosen to meet the assumptions required for causal mediation analysis, namely, control for exposure–outcome, exposure–mediator and mediator–outcome confounding, and no exposure-induced mediator–outcome confounding (no confounder for the mediator–outcome relationship is itself affected by the exposure) (VanderWeele, Reference VanderWeele2016). However, as noted in Table 1, potential outcomes are defined based on intervening on the mediator and not on the exposure (i.e., we consider racial identity and HIV seropositivity as non-modifiable). Thus, our interpretations in Table 1 hold under the assumptions of sufficient confounder control for the mediator–outcome relationship (VanderWeele and Robinson, Reference VanderWeele and Robinson2014) and no exposure-induced mediator–outcome confounding. Primary findings are presented without including RDS weights, as no consensus exists on their incorporation in natural effect models (Avery et al., Reference Avery, Macpherson, Flicker and Rotondi2021; Gile et al., Reference Gile, Beaudry, Handcock and Ott2018), although there exists ongoing discussion (Yauck et al., Reference Yauck, Moodie, Apelian, Fourmigue, Grace, Hart, Lambert and Cox2021).
We performed three sensitivity analyses to verify the robustness of our natural effect model findings. (1) We compared our primary results with those after accounting for RDS weights. (2) We considered socioeconomic factors (income, education and employment), social support and resilience as confounders of the mediator–outcome relationship. These could potentially be affected by the exposure and violate a key assumption of causal mediation analysis (i.e., no exposure-induced confounding). Therefore, they were not included in the primary model. Findings from models in which they were included as separate confounders (in addition to primary model confounders) were nonetheless compared to primary results. Mean scores on the Medical Outcomes Study Social Support Survey Instrument (Sherbourne and Stewart, Reference Sherbourne and Stewart1991) and the Connor Davidson Resilience Scale-2 (Vaishnavi et al., Reference Vaishnavi, Connor and Davidson2007) were used as measures of social support and resilience. (3) Finally, we compared primary results with those in which we lagged the outcome by 1 year. This was done to assess if a potential overlap between the assessment timeframes of the mediator and outcome (i.e., simultaneous or reversed temporality) biased the primary results. We did not treat this as the primary model for several reasons. There was a 29% loss to follow-up (LFU) between enrolment and first follow-up visit at 1 year, leading to an important loss of sample size and hence power. Moreover, the long lag between variables may dilute the estimated mediated effect. For this sensitivity analysis, we additionally accounted for LFU by using inverse probability weights (Willems et al., Reference Willems, Schat, van Noorden and Fiocco2018). This weighting may reduce bias due to selective LFU but also increases standard errors relative to an unweighted analysis, further decreasing power.
All analyses were performed in Stata 17.0 and R statistical software using the RDS and medflex packages. More details of the statistical procedures are provided in the supplementary section.
Results
Study population
We included 1127, 504 and 740 GBM participants from Montreal, Toronto and Vancouver, respectively. RDS-adjusted mean age of participants ranged between 34.8 years (95% CI: 32.6, 36.9; Toronto) and 36.9 years (95% CI: 35.5, 38.2; Montreal). Racialized GBM constituted 27.8% (95% CI: 22.6%, 33.1%; Montreal); 37.6% (95% CI: 30.5%, 44.6%; Toronto) and 45.6% (95% CI: 39.2%, 52.0%; Vancouver) of the study population. The proportions of GBMHIV were 13.8% (95% CI: 10.2%, 17.4%; Montreal), 19.1% (95% CI: 14.8%, 23.5%; Toronto) and 19.4% (95% CI: 13.8%, 24.9%; Vancouver). Mean duration of living with HIV was 17.5 years (95% CI: 14.7, 20.4; Montreal), 17.1 years (95% CI: 13.8, 20.3; Toronto) and 15.1 (95% CI: 12.4, 17.8; Vancouver). Approximately, one quarter of participants used MHS in the prior 6 months across all three cities (Montreal, 26.2%, 95% CI: 21.8%, 30.6%; Toronto, 30.1%, 95% CI: 24.1%, 37.2%; Vancouver, 24.7%, 95% CI: 19.6%, 29.8%).
Montreal had the highest proportion of white HIV-negative GBM (61.2%, 95% CI: 55.7%, 66.6%) and Toronto the highest proportion of white GBMHIV (14.5%, 95% CI: 10.8%, 18.2%). Vancouver had the highest proportion of both racialized HIV-negative GBM (38.1%, 95% CI: 31.9%, 44.2%) and racialized GBMHIV (7.5%, 95% CI: 3.7%, 11.4%) (Table 2).
Table 2. Characteristics of participants recruited across Montreal, Toronto and Vancouver in the Engage cohort study via respondent-driven sampling
RDS: respondent-driven sampling; 95% CI: 95% confidence intervals; SD: standard deviation.
a Other (sexual identity) includes straight (n = 7), questioning (n = 7), asexual (n = 2), two-spirit (n = 15) and responses that do not include any of the categories listed (n = 15).
b Racialized includes Indigenous peoples, those listed as visible minorities under the Employment Equity Act of Canada or those that self-identify as persons of colour.
c Chronic mental health conditions include any of the following mental health provider diagnosed conditions: substance use disorder, anxiety disorder, post-traumatic stress disorder (PTSD), attention deficit disorder or attention deficit hyperactivity disorder (ADD or ADHD), bipolar disorder, borderline personality disorder or major depressive disorder.
PD scores and MHS use in the prior 6 months by exposure groups
The EDS demonstrated good internal consistency (study Cronbach’s α > 0.90) for all three cities. Racialized GBMHIV in Montreal had the lowest mean EDS score (7.2, 95% CI: 5.8, 8.6), and racialized GBMHIV in Toronto had the highest (13.0, 95% CI: 9.2, 16.7).
Between cities, MHS use was variable across GBM groups. In Montreal, MHS use was highest in racialized GBMHIV (33.6%, 95% CI: 11.2%, 56.1%); in Toronto, it was highest in white GBMHIV (35.0%, 95% CI: 16.8%, 53.2%) and racialized HIV-negative GBM (35.3, 95% CI: 23.0%, 47.5%); and in Vancouver, it was highest in white GBMHIV (37.7%, 95% CI: 20.9%, 54.5%) (Table 3).
Table 3. Mental health services use in the past 6 months by GBM enrolled in the Engage cohort study by intersectional categories and mean perceived discrimination scores measured on the Everyday Discrimination Scale (EDS)
Group 1: white HIV-negative, Group 2: white living with HIV, Group 3: racialized HIV-negative, Group 4: racialized living with HIV.
MHS use P6M: mental health services use in the past 6 months; RDS: respondent-driven sampling; SD: standard deviation; 95% CI: 95% confidence interval.
Overall represents the total study population combining across the three cities, these are not RDS weighted as RDS weights are city specific.
a EDS range: 5–25.
b Total participants: Group 1: n = 1376, Group 2: n = 327, Group 3: n = 577, Group 4: n = 91.
There were 64 missing values for EDS scores (27 for Group 1; 8 for Group 2; 24 for Group 3; 5 for Group 4). If a participant indicated that they preferred not to respond to any of the five items on the EDS, the total score was not calculated, and that participant was assigned a missing value.
Attributions of discrimination
We used three-city data. As mentioned, these data do not serve an analytical purpose but are presented to showcase the distribution of discriminatory experiences attributed to sexual identity, racial identity and HIV status.
In Group 1 (white HIV-negative GBM), 46.6% reported a single form of discrimination mostly attributed to sexual identity. In Group 2 (white GBMHIV), 20.5% reported two forms of discrimination mostly attributed to sexual identity and HIV status. In Group 3 (racialized HIV-negative GBM), 38.3% reported two forms of discrimination attributed mostly to sexual and racial identities, whereas in Group 4 (racialized GBMHIV), 17.6% reported discrimination attributed to sexual identity, racial identity and HIV status. However, across the four groups, attributions to discrimination based on sexual identity, racial identity and HIV status were varied. Individuals with multiply marginalized identities did not necessarily attribute discrimination to all aspects of their identities. Moreover, 25.3% (Group 4) to 47.1% (Group 1) of participants reported no experiences of discrimination based on their sexual identity, racial identity and HIV status (Table 4).
Table 4. Attribution(s) of discrimination to sexual identity, racial identity and HIV status across exposure groups using data combined across the three cities
Patterns for each city were consistent with those presented for the total study population here.
Note: HIV status indicates HIV seropositivity.
Group 1: white HIV-negative, Group 2: white living with HIV, Group 3: racialized HIV-negative, Group 4: racialized living with HIV.
a No discrimination includes participants who marked never to all five questions on the first part of the EDS (i.e., total EDS score of 5) or those that indicated not having faced discrimination due to their sexual identity, racial identity and HIV status in the second part of the EDS.
b Attribution to racial identity among groups that included white GBM (n = 86): 32 were not Canadian citizens; 29 identify their ethnicity as English Canadian, 8 French Canadian, 39 of Eastern or Western European ancestry and 10 as mixed ethnicity; 66 were primarily English speakers, 13 primarily French speakers and 7 spoke either Spanish, Russian, Slovak, Chechen, Turkish or Portuguese primarily. No information on Jewish/Islamic heritage or political affinity was available. None identified as persons of colour.
c Attribution to HIV status among groups that included HIV-negative GBM (n = 22): 5 were born in the 1960s, 14 were born between 1978 and 1987 and 3 in the 1990s; 12 were born either in Syria, Rwanda, Lebanon, Sri Lanka, Cameron, Slovakia, Turkey, Mexico, Philippines or Brazil.
d There is a discrepancy in the number of missing values with Table 3 (n = 64). This is because participants that did not respond to any question on the first part of the EDS were marked as a missing value when calculating the total EDS score but could still indicate whether they experienced discrimination due to their sexual identity, racial identity or HIV status in the second part of the EDS.
Note that enumerating multiple forms of discrimination is not equivalent to an intersectionality-based approach. Attributions have been presented to highlight that GBM who experience discrimination because of multiply marginalized identities do not experience an ambiguous form of ‘intersectional’ discrimination. GBM may be able to recall ‘identifiable’ forms of discrimination (i.e., homonegativity, racism and HIV stigma) due to existing systems of inequity to different extents (Lewis et al., Reference Lewis, Cogburn and Williams2015). Thus, enumeration may help to identify actionable items for intervention. However, exclusive enumeration of multiple (but individual) forms of discrimination may lend itself to an understanding that frames experiences of discrimination as compartmentalized and therefore additive. An intersectionality-based approach acknowledges that experiences of discrimination due to different systems of inequity overlap with each other (i.e., cannot be examined independent of each other) to different degrees and are dynamically constructed within historical and social contexts.
Natural effect model estimates
Group 2 (white GBMHIV) had higher overall MHS use odds (TE [OR] 1.71; 95% CI: 1.27, 2.29) compared with Group 1 (white HIV-negative GBM). PD did not significantly mediate MHS use (PIE [OR], 1.02; 95% CI: 0.98, 1.05). As evinced by the PDE, if PD levels were reduced to that of Group 1, higher MHS use odds in Group 2 (OR, 1.68; 95% CI: 1.27, 2.24) would still be observed.
Relative to Group 1, Group 3 (racialized HIV-negative GBM) did not have significantly different overall MHS use odds (TE [OR], 0.95; 95% CI: 0.74, 1.22). Higher PD levels (compared with Group 1) were associated with higher odds of MHS use (PIE [OR], 1.09; 95% CI: 1.02, 1.17). If PD levels were reduced to that of Group 1, we estimate the odds of MHS use in Group 3 would concomitantly reduce, albeit not statistically significantly (PDE [OR], 0.84; 95% CI: 0.65, 1.09).
Relative to Group 1, TE (OR), 1.23, 95% CI: 0.73, 2.01; PIE (OR), 1.08, 95% CI: 0.99, 1.17 and PDE (OR), 1.23, 95% CI: 0.73, 2.15 for Group 4 (racialized GBMHIV) were not significantly different.
PD did not appear to have heterogeneous effects on MHS use for Groups 2, 3 and 4, relative to Group 1. This is indicated by non-significant MIE estimates (Table 5).
Table 5. Odds ratios for MHS use in the past 6 months across different intersectional groups, with perceived discrimination measured on the Everyday Discrimination Scale as the mediator
* Indicates statistically significant results.
Group 1: white HIV-negative, Group 2: white living with HIV, Group 3: racialized HIV-negative, Group 4: racialized living with HIV.
Confounders adjusted for in the model: age, city, having a chronic mental health condition, Canadian citizenship and cisgender status.
Estimates when missing data (for EDS and having a chronic mental health condition) were multiply imputed using chained equations are comparable to those presented here.
Sensitivity analysis findings
When we included RDS weights, there were no changes in the point estimates for TE, PDE, PIE or MIE. This may indicate that point estimates are not influenced by participants’ network sizes or may reflect the conditional nature of mediation models (in contrast to simple means and proportions, where RDS weighting is recommended). However, it may also indicate that the distribution of factors affecting PD and MHS use is homogenous across exposure groups. When additional confounders for socioeconomic indicators, social support or resilience were included separately, the difference in point estimates from Table 5 estimates varied between 0% and 5%. These factors, as measured in the Engage study, do not appear to be strong confounders. When the outcome was lagged by 1 year, PDE estimates (and consequently TE estimates) for Group 2 were reduced by 26%. We are unsure of the reasons for these reductions. However, these reductions must be interpreted cautiously; estimates could be biased if the underlying LFU mechanism depends on variables not available in the dataset such as incarceration (i.e., missingness is not at random) (Li et al., Reference Li, Shen, Li and Robins2013). The difference for other estimates varied between 0% and 9.7% across exposure groups. PIE remained the same for all groups, highlighting the consistency of PD’s effects on MHS use.
Results from sensitivity analyses are presented in Supplementary Tables S2–S6.
Discussion
Using a quantitative intersectionality approach (Bauer and Scheim, Reference Bauer and Scheim2019; Public Health Agency of Canada, 2022), we found MHS use to be higher among white GBMHIV but not significantly different for racialized GBM (irrespective of HIV status), compared with white HIV-negative GBM. Additionally, relative to white HIV-negative GBM, PD was a significant mediator of MHS use only for racialized HIV-negative GBM.
Higher MHS use among white GBMHIV compared with white HIV-negative GBM was not mediated by PD. This is explained by comparable PD levels between the two groups. In Canada, unemployment and lower income, both associated with increased stress (American Psychological Association, 2017; Baum et al., Reference Baum, Fleming and Reddy1986), are more prevalent among GBM than heterosexual men (Kinitz et al., Reference Kinitz, Gould, Shahidi, MacEachen, Mitchell, Craig-Venturi and Ross2023; Waite et al., Reference Waite, Ecker and Ross2019). Further, in the current sample, we observed lower employment and income among white GBMHIV relative to white HIV-negative GBM (Supplementary Table 7). Hence, higher MHS use in white GBMHIV could represent one means of coping with structural inequity-related stressors. Recent reviews call for equity-promoting labour interventions for GBM and other people living with HIV at structural levels (policy, community and institutional levels) (Kinitz et al., Reference Kinitz, Gould, Shahidi, MacEachen, Mitchell, Craig-Venturi and Ross2023; Maulsby et al., Reference Maulsby, Ratnayake, Hesson, Mugavero and Latkin2020). It may be worthwhile to consider how MHS may be integrated into such interventions. Additionally, the EDS only assesses interpersonal discrimination, and future studies should include structural discrimination measures, to better elucidate its effect on MHS use among different groups of GBM.
Our finding that increased PD is associated with higher MHS use for racialized HIV-negative GBM aligns with previous studies (Burgess et al., Reference Burgess, Tran, Lee and van Ryn2007; Evans and Sheu, Reference Evans and Sheu2019; Richman et al., Reference Richman, Kohn-Wood and Williams2007). However, these studies take a unidimensional approach considering only racial or sexual identity. We add to existing literature without assuming a priori that racialized HIV-negative GBM rank one identity over the other. Although not statistically significant, lower TE for racialized HIV-negative GBM (Group 3) compared with white HIV-negative GBM (Group 1) indicates that MHS use among racialized HIV-negative GBM may partly mirror underutilization observed in racialized communities (Chiu et al., Reference Chiu, Amartey, Wang and Kurdyak2018; Richman et al., Reference Richman, Kohn-Wood and Williams2007). Cultural norms influence MHS use patterns in racialized individuals in the general population (Boukpessi et al., Reference Boukpessi, Kpanake and Gagnier2021; Chiu et al., Reference Chiu, Amartey, Wang and Kurdyak2018). Thus, while higher MHS use among racialized HIV-negative GBM may represent one coping strategy for increased PD-related stress, the impact that cultural norms may have on this association also needs to be considered.
Racialized individuals (in the general population) have also been noted to use MHS primarily when self-management of mental/emotional health becomes untenable (Boukpessi et al., Reference Boukpessi, Kpanake and Gagnier2021; Chiu et al., Reference Chiu, Amartey, Wang and Kurdyak2018; Richman et al., Reference Richman, Kohn-Wood and Williams2007). Although it is unclear if racialized GBM (irrespective of HIV status) use MHS as a last resort, addressing this research gap may be crucial to develop timelier MHS for multiply marginalized individuals. Another consideration is that racialized individuals experiencing chronic discrimination are more likely to have heightened vigilance for discrimination (Brondolo et al., Reference Brondolo, Blair, Kaur, Major, Dovidio and Link2018). In such individuals, ambiguous experiences may then be perceived as discriminatory (Lewis et al., Reference Lewis, Cogburn and Williams2015), cyclically exacerbating stress. Assuming the interrelatedness of discriminatory experiences (i.e., experiencing discrimination as a continuum rather than as discrete unrelated events because of the non-divisibility of identities in multiply marginalized GBM) increases their chronicity, it is possible that racialized GBM have increased vigilance for discrimination. Hence, for mental health practitioners working with racialized GBM, the extent to which both culture and increased vigilance affect MHS use may be important considerations.
The paucity and limited funding of GBM-affirmative MHS and predominance of medical models in mental healthcare (that disregards social context) are systemic barriers previously identified to deoptimize MHS for GBM (McIntyre et al., Reference McIntyre, Daley, Rutherford and Ross2011). Recently, some improvements on these barriers have been made. In 2022, the Canadian federal government announced a $100 million plan for Two-Spirit, Lesbian, Gay, Bisexual, Transgender, Queer/Questioning, Intersex, Asexual+ communities (The Globe and Mail, 2022), and intersectionality-guided psychotherapeutic approaches are emerging (Adames et al., Reference Adames, Chavez-Duenas, Sharma and La Roche2018; Jackson et al., Reference Jackson, Mohr, Sarno, Kindahl and Jones2020; Nayak, Reference Nayak, Trachsel, Gaab, Biller-Andorno, Tekin and Sadler2021). However, the impact these measures will have on discrimination-related stress among GBM, especially those with multiple marginalized identities, remains to be seen. Individual-level intersectionality-based interventions without macrolevel reforms are unlikely to sustainably mitigate discrimination-related stress in GBM. Concurrently, interventions such as those that address homonegativity in racialized communities, and racism and HIV stigmatization in GBM communities are required (Sievwright et al., Reference Sievwright, Stangl, Nyblade, Lippman, Logie, Veras, Zamudio-Haas, Poteat, Rao, Pachankis, Kumi Smith, Weiser, Brooks and Sevelius2022). An often-overlooked aspect of intersectionality is its emphasis on building coalitions across marginalized groups to collectively oppose overlapping systems of inequity (Carastathis, Reference Carastathis2013; Sievwright et al., Reference Sievwright, Stangl, Nyblade, Lippman, Logie, Veras, Zamudio-Haas, Poteat, Rao, Pachankis, Kumi Smith, Weiser, Brooks and Sevelius2022). Going forward, it may be useful for researchers to develop and evaluate coalitional anti-discriminatory multilevel interventions.
Our findings have several limitations that merit discussion. Our analyses were contingent on data available in the Engage Cohort Study (CIHR Canadian HIV Trials Network), whose primary aims were different from the objectives addressed in this manuscript. To improve statistical power, we grouped multiple ethnicities of non-European descent as racialized. Despite this, we were inadequately powered to make definitive inferences about Group 4 and about the MIEs for any group. While racialized GBM across ethnicities may share some cultural commonalities and similar discriminatory experiences (Giwa and Han, Reference Giwa and Han2022; Sadika et al., Reference Sadika, Wiebe, Morrison and Morrison2020), we are cognizant that grouping them risks homogenizing them. We advocate for larger studies with oversampling of GBM with non-European ethnicities. As the EDS was administered only at baseline, we cannot be definitive regarding the temporality in our primary analysis. However, we found consistent results when the outcome was lagged by 1 year, making it less likely that the primary findings were affected by reverse causation. Longitudinal PD measurements may further enrich our understanding of its time-varying effects. The EDS was not developed for intersectional analysis, and certain items on the scale correspond more to specific racialized experiences than those related to homonegativity or HIV stigmatization. Thus, the EDS may not be suited for comparison across multiple social categories (Harnois et al., Reference Harnois, Bastos, Campbell and Keith2019). Additionally, the overall EDS score may encapsulate experiences to other forms of discrimination (e.g., ableism). We recommend the development of better quantitative tools to measure intersecting forms of discrimination, while cautioning against positivist stances that assume intersectional experiences represent a single fixed reality (Bowleg, Reference Bowleg2008). Although the MIE allows for PD to vary heterogeneously between exposure groups, it estimates a constant within-group average effect of PD. However, even within the same group, PD may be shaped by the degree of identification with one’s social identities and their importance to self-image (Richman et al., Reference Richman, Kohn-Wood and Williams2007), and as observed in Table 4, within-group experiences of PD are varied. Mixed methods studies that can generate both between-group and within-group average effects while also detailing the complexity and diversity of within-group PD experiences may be more suited for intersectionality-based research (Hankivsky and Grace, Reference Hankivsky, Grace, Hesse-Biber and Johnson2015). We note that our primary analysis relied on estimates that were not RDS-adjusted, which in the strictest view, limits our findings to the study sample. However, as our sensitivity analysis shows, adjusted estimates were comparable to unadjusted estimates indicating that our findings may be generalizable to the larger population of GBM living in Montreal, Toronto or Vancouver. Given that best statistical practices for inclusion of RDS weights in the specific context of natural effect models have not been established, the concordance between RDS unadjusted and adjusted estimates is reassuring. Our results show a positive association between PD and MHS use in racialized HIV-negative GBM. However, previous research has also indicated that discrimination may impair the accomplishment of cognitively demanding tasks (Richman and Lattanner, Reference Richman and Lattanner2014) like MHS use. Therefore, it is possible that discrimination in different contexts may have different associations with MHS use.
The present study found that discrimination may be associated with MHS use at both baseline and 1-year follow-up. Longer longitudinal studies investigating the effects of intersecting forms of discrimination on MHS use trends among GBM are warranted. Simultaneously, we present natural effect models as a viable method for three-way decomposition, to reveal underlying mechanisms not apparent through the sole examination of TEs. However, more research to improve quantitative intersectionality measures and methodology is required.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S2045796024000143.
Availability of data and materials
The datasets generated and/or analysed during the current study are not publicly available due to privacy concerns for this ongoing cohort study. Deidentified participant data used in this analysis are stored at the British Columbia Centre for Excellence in HIV/AIDS (BCCFE). For information regarding these databases, please contact Joseph Cox, a principal investigator on the Engage Study.
Acknowledgements
We (the authors) work in academia in different capacities (the first author is a doctoral student, and the co-authors are faculty at varying stages of their careers). We come from diverse disciplinary backgrounds (Biostatistics, Epidemiology, Infectious Diseases Medicine, Psychology, Public Health and Sociology). Most of the authors identify as cisgender sexual minority men, but co-authors also include non-binary individuals, transgender sexual minority men and cisgender heterosexual women and men. We come from diverse economic, racial and ethnic backgrounds (including Francophone and Anglophone Canadians). We also have different citizenships. The first author identifies as a non-American Indigenous person of the Khasi tribe of India. However, none of the authors identify as Indigenous Peoples of Canada or as persons living with HIV. We respectfully acknowledge that each of us contributed to this manuscript from traditional, unceded First Nation territories. Specifically, this manuscript was developed in Montreal, on the traditional and unceded territory of the Mohawk (Kanien’kehá:ka) Nation.
The manuscript is motivated by an understanding that inequities in MHS for GBM cannot be addressed without interventions on social, economic and political factors that generate and reinforce societal hierarchies, influencing the social and material conditions in which individuals are born, grow, live, work and age. Intersectionality theory which forms the base of this work originates from coalitional anti-oppressive traditions and critical realist thought of Black and Chicana feminists. However, we acknowledge that the methodology of this manuscript was informed by biostatistical principles originating in empirical positivist traditions. We interpret our findings empirically but situate them within the framework of intersectionality.
The authors thank Jesse Gervais at UQAM (Université du Québec à Montréal) for helping with the manual coding of natural effect models. They also extend their gratitude to all office staff, community engagement committee members and community partner agencies. Most of all, the authors thank the study participants in Engage, who contributed their time and information, and without whom, this work would not have been possible.
Financial support
The Engage Cohort Study was led by Principal Investigators in Montreal by Joseph Cox and Gilles Lambert; in Toronto by Trevor A. Hart and Daniel Grace and in Vancouver by Jody Jollimore, Nathan J. Lachowsky and David M. Moore. Engage/Momentum II was funded by the Canadian Institutes for Health Research (CIHR, #TE2-138299; #FDN-143342; #PJT-153139), the Canadian Foundation for AIDS Research (CANFAR), the Ontario HIV Treatment Network (OHTN, #1051), the Public Health Agency of Canada (#4500345082) and Toronto Metropolitan University. TAH was supported by an Endgame Leader Chair Award in Gay and Bisexual Men’s Health from the Ontario HIV Treatment Network. NJL was supported by a Scholar Award from the Michael Smith Foundation for Health Research (#16863). DG was supported by a Canada Research Chair in Sexual and Gender Minority Health. IM was supported by a training grant from the Fonds de recherche du Québec-Santé, Healthy Brains Healthy Lives Doctoral Fellowship; EEMM is a Canada Research Chair (Tier 1) in Statistical Methods for Precision Medicine and acknowledges the support of a chercheur de mérite career award from the Fonds de Recherche du Québec, Santé. SSS was supported by postdoctoral fellowship awards from CIHR and the CTN.
Competing interests
No competing interests to declare.
Ethical standards
The Engage cohort study was approved by the research ethics boards of the following institutions: the Research Institute of the McGill University Health Centre, Toronto Metropolitan University, St. Michael’s Hospital, University of Toronto (Health Sciences), University of Windsor, University of British Columbia, University of Victoria and Simon Fraser University.
Open access
