Leaving no one behind? An equity analysis of the HIV care cascade among a cohort of people living with HIV in Manitoba, Canada

Background Manitoba is a central Canadian province with annual rates of new HIV infections consistently higher than the Canadian average. National surveillance statistics and data from the provincial HIV care program suggest that epidemiological heterogeneity exists across Manitoba. New HIV cases are disproportionately reported among females, Indigenous-identifying individuals, and those with a history of injection drug use. Given the heterogeneity in acquisition, it is of interest to understand whether this translates into inequalities in HIV care across Manitoba. Methods A sample of 703 participants from a clinical cohort of people living with HIV in Manitoba, with data current to the end of 2017, was used to conduct cross-sectional, disaggregated analyses of the HIV care cascade to identify heterogeneity in service coverage and clinical outcomes among different groups receiving HIV care in Manitoba. Equiplots are used to identify and visualize inequalities across the cascade. Exploratory multivariable logistic regression models quantify associations between equity variables (age, sex, geography, ethnicity, immigration status, exposure category) and progression along the cascade. Adjusted odds ratios (AOR) and 95% confidence intervals (95%CI) are reported. Results Equity analyses highlight inequalities in engagement in and coverage of HIV-related health services among cohort participants. Equiplots illustrate that the proportion of participants in each cascade step is greater for those who are older, white, non-immigrants, and report no history of injection drug use. Compared to those living in Winnipeg, participants in eastern Manitoba have greater odds of achieving virologic suppression (AOR[95%CI] = 3.8[1.3–11.2]). The odds of Indigenous participants being virologically suppressed is half that of white participants (AOR[95%CI] = 0.5[0.3–0.7]), whereas African/Caribbean/Black participants are significantly less likely than white participants to be in care and retained in care (AOR[95%CI] = 0.3[0.2–0.7] and 0.4[0.2–0.9], respectively). Conclusions Inequalities exist across the cascade for different groups of Manitobans living with HIV; equiplots are an innovative method for visualizing these inequalities. Alongside future research aiming to understand why inequalities exist across the cascade in Manitoba, our equity analyses can generate hypotheses and provide evidence to inform patient-centred care plans that meet the needs of diverse client subgroups and advocate for policy changes that facilitate more equitable HIV care across the province.


Background
Manitoba is a central Canadian province where annual rates of new HIV infections are consistently higher than the national average (7.9 vs. 6.9 per 100,000 population, respectively, in 2018) [1]. Injection drug use (33.9%), condomless anal sex between men (24.4%), and condomless vaginal (heterosexual) sex (20.9%) are the most commonly identified HIV risk exposures in Manitoba [2], and new infections in 2018 were disproportionately high, compared to the rest of Canada, among individuals identifying as Indigenous (50% vs. 19.3%) and female (40% vs. 29.3%) [1,3]. Additionally, notable heterogeneity in rates of new HIV infection exists across the province by geography, age, and sex [3]-in 2018, 77.6% of new diagnoses occurred in Winnipeg, the provincial capital and main urban centre, and among newly diagnosed females, 11.6% were ≤ 19 years (compared to 1.6% of males) and 14.0% were ≥ 60 years (compared to 3.1% of males) [3]. At the end of 2018, Manitoba Health, Seniors and Active Living (MSHAL), estimated that 1572 people were living with HIV in the province (personal communication, J. Paul, April 16, 2020), and the Manitoba HIV Programthe primary provider of HIV care in the province-estimated that approximately 1400 people were in care in the same year [2].
Over the past decade of HIV research, heavy emphasis has been placed on the HIV care cascade ("the cascade")-a framework and analytic tool providing insights into the continuum of care services for people living with HIV [4][5][6]-and its simplified counterparts the 90-90-90 Initiative [7]. Conventionally, cascades use aggregate data to illustrate the proportion of individuals in a population of people living with HIV who have been diagnosed, linked to HIV care services, retained in care, then initiated and sustained on HIV treatment to, ultimately, reach virologic suppression. Using aggregate data to illustrate the continuum of HIV care for an entire population is useful insofar as it can provide a general picture of points of "leakage" or "bottlenecks" within a health system or care program. However, relying on aggregate data to paint a picture of an entire population risks obscuring the underlying heterogeneity among and between individuals and groups who make up the population. To generate evidence that can help to inform the development and optimization of interventions and programs addressing inequities in HIV care, it is crucial to conduct equity analyses that generate disaggregated cascades to showcase nuances and highlight inequalities across the cascade steps within a population.
In 2015, all 193 Member States of the United Nations agreed upon the 2030 Agenda for Sustainable Development, comprising seventeen Sustainable Development Goals (SDGs), which built upon the Millennium Development Goals (MDGs) introduced fifteen years earlier [8]. At the core of the SDGs is the notion of leaving no one behind, which, "represents the unequivocal commitment … to reduce the inequalities and vulnerabilities that leave people behind and undermine the potential of individuals and of humanity as a whole" [9 , p. 6]. This idea underscores the interconnectedness of the SDGs and principles of health equity [10]-a noted limitation of the MDGs [11]. As such, under the auspices of the SDGs [12], there is a need for research that focuses on identifying (health) inequalities that exist, examining the factors that perpetuate and exacerbate these inequalities, understanding how specific inequalities are related to broader health inequity [13], and developing strategies to minimize or, ideally, eliminate them. As noted in SDG 17, to adequately assess (in)equities, it is necessary that data are disaggregated by socioeconomic, demographic and other relevant, context-specific characteristics [11,12].
Publicly available HIV epidemiological data in Manitoba are limited to reports published by the Public Health Agency of Canada (PHAC) [1] and MHSAL [3], which focus solely on surveillance data. As such, local understandings of inequalities in HIV care and clinical outcomes among different groups in the province are rudimentary. In 2013-through the support of a multi-site program of research, Advancing Primary Healthcare for Persons Living with HIV in Canada (the LHIV Study), funded by the Canadian Institutes of Health Research [14]-a prospective clinical cohort of people living with HIV and/or receiving HIV care in Manitoba was established as an embedded research project within the Manitoba HIV Program [15]. The establishment of the LHIV-Manitoba cohort opens up numerous analytic opportunities to better understand HIV epidemiology in Manitoba and, for the first time, provides access to de-identified, individual-level clinical data, allowing for disaggregated analyses to take place.
Here, building upon previous work [15][16][17], we use equiplots to present disaggregated cascade analyses (by age, sex, geography, ethnicity, immigration status, and HIV exposure category) that visualize inequalities in service uptake and clinical outcomes among LHIV-Manitoba cohort participants who were alive as of 31 December 2017 and had received an HIV diagnosis on or before that date. Exploratory multivariable logistic regression analyses are used to quantify these inequalities, generate hypotheses, and provide guidance for future cascade research in Manitoba. In conjunction with future research to understand why identified inequalities exist across the cascade [6] and how these inequalities contribute to health inequities [13], our examination of the cascade through an equity lens [18,19] will provide Manitoba's provincial care program with evidence needed to develop patient-centred care plans that meet the needs of heterogeneous client subgroups, and to advocate for policy changes addressing inequities in HIV care across the province.

Study setting
Manitoba has a population of 1.36-million people, spanning over 550,000 km 2 . Approximately 57% of Manitoba's population lives in the capital city of Winnipeg, 37% in the western, eastern, and southern regions (rural), and 6% in the north (rural-remote) [20]. HIV care in Manitoba is primarily provided through the Manitoba HIV Program, comprising three clinics-two in Winnipeg and one in a smaller, rural city in southwestern Manitoba. As such, the majority of Program clients living in rural and rural-remote regions of the province are required to travel substantial distances to attend clinic appointments. The Manitoba HIV Program employs a multidisciplinary care model encompassing a full complement of health and social service providers, including HIV specialists and family physicians, nurses, pharmacists, social workers, and other allied health professionals. Despite Canada's publicly funded healthcare system, no Canadian provinces or territories have a single drug plan that provides universal coverage for all, and out-of-pocket expenses associated with prescription medications vary substantially depending on an individual's insurance coverage [21].

Data sources
Data used to generate disaggregated, cross-sectional analyses of the Manitoban cascade are derived from the LHIV-Manitoba clinical cohort dataset, comprising individual-level, de-identified clinical data linked to provincial administrative health databases. Clinical cohort data are linked to the most recently available administrative health datasets and are reported up to 31 December 2017. Manitoba's administrative datasets include individual-level records for nearly all contacts within the province's healthcare system. Linked administrative data within the cohort include variables pertaining to physician visits, hospital admissions, prescription drug dispensation, and laboratory testing, including quantitative HIV plasma viral load data [15]. The process of establishing the cohort, a complete profile of cohort participants, and a comprehensive description of the datasets is detailed in earlier work [15]. Briefly, the cohort includes 890 unique individuals, comprising 63.6% of the estimated 1400 clients receiving care from the Manitoba HIV Program and 56.6% of the estimated 1572 people living with diagnosed HIV in the province. All adults (≥18 years) living with HIV and/or receiving HIV care in Manitoba were eligible for participation, unless under the jurisdiction of the Public Guardian and Trustee of Manitoba or otherwise unable to make their own healthcare decisions.
Potential participants have the opportunity to agree to any combination of three separate components of the LHIV-Manitoba cohort study: (i) have their clinical data collected; (ii) have their clinical data linked to provincial administrative health datasets; and/or (iii) indicate interest in being approached about participating in future HIV research studies [15]. In total, 725 (81.5%) cohort participants agreed to have their data extracted from clinical records, and 703 (78.9%) of those participants also agreed to have their clinical data anonymously linked to administrative health databases. A limited clinical dataset exists for an additional 165 individuals whose clinical records were reviewed posthumously and linked to administrative health databases. In total, the linked dataset comprises 868 participants. Missing data were minimal throughout the cohort dataset-9.1% (n = 81) of all participants who were alive at enrolment are missing observations from at least one key variable of interest [16]. Missing data points are excluded from relevant analyses. All cohort participants who were alive as of 31 December 2017, had received a positive HIV diagnosis at any point on or before the same date, and had provided written, informed consent to have their clinical data reviewed and linked to administrative health datasets were included in a baseline HIV care cascade model and subsequent equity analyses.

Equity analyses
An HIV care cascade has previously been developed and specifically tailored to accommodate available data sources within Manitoba [16]. The existence of a single insurer responsible for payment of the majority of health services in the province (MHSAL) and linkable, population-based administrative health databases afforded opportunities for developing sensitive and data-intensive indicator definitions [16,17] similar to definitions used in other provinces [22,23]. Table 1 outlines the established definitions for five indicators representing each step of the Manitoban cascade, which were derived data from both clinical cohort data (CD4 cell count data) and linked administrative health datasets (viral load, physician visits, and drug dispensation data).
Throughout the paper, the term "inequality" describes measured differences in outcomes between groups, whereas "inequities" refer to the implications of population-level inequalities, which, in the context of public health, are understood to be fundamentally unjust or unfair [13]. To examine inequalities across the Manitoban care cascade, we identified relevant equity variables available within the cohort by which each cascade step indicator was then disaggregated. These variables-which have been recommended for use in equity analyses previously [10,11,19]-included age; sex; geographic location of residence; self-identified ethnicity; immigration status; and primary HIV exposure category, identified using a "risk hierarchy" framework [24]. A participant's geography is categorized by provincial Regional Health Authority (Supplementary Figure, see Additional File 1), which is inferred from the postal code of residence recorded in the cohort database [25]. Participants are categorized as "immigrants" if they were foreign-born and had immigrated to Canada in 2001 or later [26] (Maritim C, McClarty L, Leung S, Bruce S, Restall G, Migliardi P, Becker M. HIV treatment outcomes among newcomers living with HIV in Manitoba, Canada JAMMI In press), or "non-immigrants" if they were either Canadian-born or foreign-born, having immigrated to Canada before 2001.

Equiplots
To aid in the visualization of equity analyses, equiplots [27] were used to present the proportion of participants within each group-defined by key equity variables-who reach each cascade step, thus illustrating inequalities in progression along the cascade. Data were excluded from equiplot analyses for participants living outside of Manitoba or without a known permanent address. Equiplots were generated in Stata 15.1 (College Station, TX) using code available through the International Centre for Equity in Health, Universidad Federal de Pelotas, Brazil [27].

Multivariable logistic regression models
The in care, retained in care, on treatment, and virologically suppressed cascade steps (as defined in Table 1) were used as dichotomous outcome variables and the aforementioned equity variables were included as categorical exposure variables in multivariable logistic regression models. For each of the four outcomes of interest, six individual models were constructed to quantify associations between individual equity variables and reaching each cascade step. All equity variables were converted into dummy variables in the models and were tested for collinearity prior to inclusion. Cohort participants living outside of Manitoba and those who had no known address recorded in their medical files were excluded from multivariable models quantifying inequalities across the cascade, disaggregated by geographic region (Model 3 in Table 4).
Adjusted odds ratios (AOR) are presented using 95% confidence intervals (95%CI) to assess statistical significance. Crude analyses from unadjusted models are presented in the Supplementary

Results
In total, 703 cohort participants were alive and had received an HIV diagnosis as of 31 December 2017. Distributions of key equity variables among these participants, disaggregated by cascade step, are presented in Table 2.

Examining the cascade through an equity lens
Analyzing the Manitoba cascade data through an equity lens highlights a number of inequalities in the proportion of cohort participants who reached the in care, retained in care, on treatment, and virologically suppressed steps. Data corresponding to each of the presented equiplots are presented in Table 3.

Age
Disaggregated by age group, cascade data ( Fig. 2 [26]. "Non-immigrants" are either Canadian-born participants or foreignborn participants who immigrated to Canada before 2001 d "Other risk" includes recipients of blood/blood products, perinatal acquisition, occupational exposure

Sex
The equiplot illustrating disaggregated cascade estimates for male and female participants (Fig. 3) highlights similarities between groups at each step with distinguishable, though small, differences in the in care (91.3 and 89.3%, respectively) and virologically suppressed (75.0 and 73.0%, respectively) cascade steps (Table 3). Data presented in Table 4 (Model 2) indicates that female participants tend to have greater odds of being in each successive cascade step compared to male participants, but these differences are not statistically significant.

Geography
Inequalities are observed between the proportions of cohort participants in each cascade step when data are disaggregated by geographic region (Fig. 4). Given that the majority (81.9%) of cohort participants reside in the Winnipeg, the proportion of Winnipeg-based cohort participants in each cascade step are similar to the cascade estimates for the entire cohort, depicted in Fig. 1. Visual analysis of equiplot data suggests very little leakage occurs between cascade steps among participants living in eastern Manitoba. However, Fig. 4 and Table 3 highlight substantial drop-off between the alive and diagnosed and in care steps among western Manitobans (100 to 76.0%), and between the on treatment and virologically suppressed steps among those living in northern (84.0 to 72.0%) and southern regions (77.8 to 55.6%). Logistic regression Model 3 (see Table 4) indicates that compared to those living in Winnipeg, the odds of being in care are significantly lower for participants living in western Manitoba (AOR = 0.33, 95%CI:0.12-0.90). Furthermore, compared to participants living in Winnipeg, the odds of being virologically suppressed for those living in eastern Manitoba is nearly four-times greater (AOR = 3.82, 95%CI:1.31-11.17).

Ethnicity
Compared to all other ethnicity categories white cohort participants comprise the greatest proportion in each step across the cascade (Fig. 5). The proportions of participants identifying as sub-Saharan African/Caribbean/black (ACB) and those in the Other ethnicity category are relatively low in the in care step (83.5 and 83.3%, respectively), whereas the proportions of Indigenous participants in the on treatment and virologically suppressed steps are relatively low (77.0 and 65.1%, respectively). Accordingly, Model 4 in Table 4 highlights that compared to white participants, the adjusted odds of being in care is significantly lower for ACB participants (AOR = 0. 34

Immigration status
After controlling for age group, sex, and ethnicity, a participant's immigration status was not found to significantly influence their odds of being in a given cascade step (see Table 4, Model 5). However, the equiplot in Fig. 6 and data in Table 3 highlight a number of important inequalities between groups. In general, the proportion of participants who had immigrated to Canada in 2001 or later is notably lower than the proportion of non-immigrant participants in the in care (81.1% vs 92.2%, respectively), retained in care (77.8% vs 87.4%, respectively), and on treatment (75.6% vs 82.4%) cascade steps.

HIV exposure category
Disaggregating cascade data by HIV exposure category (Fig. 7) revealed few inequalities across cascade steps. Most notably, the proportion of participants whose primary HIV risk is injection drug use (IDU) categorized as virologically suppressed (61.8%) is at least 15% lower than the same for all other groups (Table 3). Compared to male participants reporting condomless anal sex with  [26]. "Non-immigrants" are either Canadian-born participants or foreignborn participants who immigrated to Canada before 2001 d "Other risk" includes recipients of blood/blood products, perinatal acquisition, occupational exposure other men (MSM), participants with a history of IDU are half as likely to be included in the virologically suppressed step (Model 6 in Table 4, AOR = 0.48, 95%CI: 0.27-0.85). The proportion of participants reporting condomless sex-either anal or vaginal-as their primary HIV exposure categories are distributed similarly across cascade steps (Fig. 7), although a slight inequality emerges between these groups at the virologically suppressed step (78.6% vs 74.8%, respectively).

Discussion
In general, our data indicate that Manitobans living with HIV are progressing well along the cascade and are nearly meeting 90-90-90 targets-81.5% of those alive and diagnosed are on treatment and 91.3% of those on treatment are virologically suppressed (Fig. 1). However, equity analyses highlight important inequalities in progression along the HIV care cascade among cohort participants within different sociodemographic groups. Disaggregating our cascade data calls attention to clear inequalities in HIV care and outcomes by age, geography, and ethnicity among cohort participants. In general, individuals who are younger and non-white are less likely than their counterparts to reach subsequent cascade steps. Notable heterogeneity also exists along the cascade based on participants' reported HIV exposure categories, with MSM progressing relatively well participants reporting IDU having relatively poor odds of reaching virologic suppression. These trends are not unprecedented; similar inequalities across the cascade have been noted in a variety of contexts [28][29][30][31][32]. While our multivariable logistic regression analyses highlight specific inequalities of statistical significance across the cascade, the use of equiplots to analyze disaggregated cascade data is an innovative and important method for identifying inequalities that, although not statistically significant, are highly relevant and should be considered during programmatic planning and design to address population-level inequities in HIV-related health outcomes. On the quest to generate evidence that can inform policy and program development aimed at minimizing health inequities, using cascade data, both aggregated and disaggregated, to identify inequalities in health outcomes and service access, delivery, and utilization is necessary, but insufficient. As Seckinelgin [6] and Zamora and colleagues [11] have argued, employing additional methodologies, such as qualitative inquiry and community-based participatory research, to inform policy and program design is crucial.
When it was first introduced in 2011 by Gardner and colleagues [4], the spectrum of engagement of HIV care, which ultimately became the HIV care cascade, was framed as an analytic tool for mapping individual-and population-level progression through the continuum of HIV care services. Specifically, Gardner's model [4] provides a framework through which to determine the proportion of individuals in various stages along the continuum, and to "explore the potential impact of interventions to improve engagement in care" (p.795). However, over time, the cascade, and its simplified counterparts, the 90-90-90 Initiative [7] and the 95-95-95 Fast-Track targets [33], have been adopted or endorsed by global technical and policy normative bodies (e.g. UNAIDS [7,33] and the World Health Organization [34]), and used to guide and influence international HIV policy development [6]. Expanding the utility of the cascade framework from an analytic tool to a large-scale decision-and policy-making framework is problematic because, as Seckinelgin [6] notes, "the model itself does not analyse the broader socio-political and economic conditions that interact with individuals' experiences of HIV and that inform their decisions to engage with health services" [6]. In the process of developing health policies that align with the principles of health equity and the SDG commitment to leaving no one behind, it is essential for decision-makers to thoroughly consider  [26]. "Non-immigrants" are either Canadian-born participants or foreignborn participants who immigrated to Canada before 2001 g Multivariable models 5 and 6: n = 703; adjusted for age group, sex, and ethnicity h "Other risk" includes recipients of blood/blood products, perinatal acquisition, occupational exposure how social determinants of health influence and manifest inequalities in health outcomes and access to health services [10,35]. As we demonstrated, performing equity analyses using HIV care cascade data, and illustrating inequalities along the cascade using equiplots, concisely draws attention to points along the cascade at which specific groups of individuals are unable to optimally engage in HIV care or reach target health outcomes. Still, these analyses provide insufficient context or explanation for leakages along the cascade. In order to appreciate nuances in observed inequalities across the cascade, and to understand, for example, why people are having a hard time engaging in their HIV care and how to best support equitable access for all, complementary research approaches, namely qualitative inquiry and community-based participatory research, are necessary [6,11,36].
Next, using our analyses in this paper as an example, we demonstrate one way to expand the utility of innovative data visualization techniques, such as the equiplot. In Fig. 4, obvious discrepancies exist in the proportions of cohort participants from different geographic regions in Manitoba falling within each cascade step. Of particular interest to the Manitoba HIV Program may be the  relatively low proportions of individuals living in western Manitoba categorized as in care and the substantial leakage at the virologically suppressed step among participants living in southern Manitoba. Indeed, previous research has also identified substantial geographic heterogeneity in engagement in HIV care [31,37], some of which may be attributable to limited access to services due to physical distance [38], or other individual-, community-, and structural-level barriers [32,37,39]. Although our disaggregated analyses have provided a useful starting point for understanding geographic heterogeneity in HIV care in Manitoba, further mixed methods explorations will be necessary to delve into understanding the complex circumstances that shape inequities along the cascade before meaningful recommendations can be made to inform local programming or policy. A next step to understand geographic inequalities will require further disaggregating data (e.g. by sex, age, socioeconomic status) to uncover whether specific groups within geographic regions are further vulnerable  to suboptimal engagement. Once a reasonable level of granularity is achieved in identifying "key groups" who may require additional support to engage in HIV care, program adjustments and policy development should then be based upon further-contextualized understandings of barriers to engagement in care through meaningful community involvement in program and policy decisions [11,36]. Policy and programmatic decisions aimed at reducing health inequities must incorporate nuanced conceptualizations of how the various factors that influence access to and engagement with necessary and appropriate health services interact and overlap [39] to create specific conditions that prevent individuals from progressing through the HIV care cascade and other care continua [6]. If these complexities are not considered, and instead the linear logic inherent to the cascade [6] is privileged, policies intended to reduce gaps in equity will continue to miss the mark.

Study limitations
This study has a few limitations that must be noted. First, a number of limitations inherent to the design of our clinical cohort have been described in detail elsewhere [14][15][16][17]. Opportunities to participate in the cohort are introduced to individuals in the context of their clinic appointments with the Manitoba HIV Program; participation is optional and does not impact the way that HIV care and other services are received. Still, we have to assume that selection bias may be influencing our analyses, and actual engagement in HIV care among the clinic population may be lower than we are able to assess from the cohort. For the same reasons, we cannot presume that our findings are generalizable to the broader population of people living with HIV in Manitoba, although previous work suggests that these data are reasonably representative of larger population in HIV care in Manitoba [15]. Second, using the LHIV-Manitoba cohort as a starting point for the first step of the cascade means that we cannot ascertain information about the proportion of people living with undiagnosed HIV in Manitoba and thus limits our ability to generate provincial estimates for all 90-90-90 targets. Finally, available data were limited such that we were unable to analyse the Manitoban HIV care cascade by income, level of education, or other socioeconomic status (SES) indicators. This will be an important addition to this work, which we will undertake as we move forward with more detailed analyses of our cohort data.

Conclusion
Our findings highlight a need for further investigation into the complex and dynamic circumstances that shape the lives of people living with HIV in Manitoba and, ultimately, influence the ability of certain groups to engage in their HIV care. While our cascade equity analyses provide a useful starting point to work toward achieving health equity and leaving no one behind for people living with HIV in Manitoba, eliciting meaningful policy and programming change will require deeper, more comprehensive work to understand barriers and facilitators to engagement in care.