Combating HIV stigma in low‐ and middle‐income healthcare settings: a scoping review

Abstract Introduction Nearly 40 years into the HIV epidemic, the persistence of HIV stigma is a matter of grave urgency. Discrimination (i.e. enacted stigma) in healthcare settings is particularly problematic as it deprives people of critical healthcare services while also discouraging preventive care seeking by confirming fears of anticipated stigma. We review existing research on the effectiveness of stigma interventions in healthcare settings of low‐ and middle‐income countries (LMIC), where stigma control efforts are often further complicated by heavy HIV burdens, less developed healthcare systems, and the layering of HIV stigma with discrimination towards other marginalized identities. This review describes progress in this field to date and identifies research gaps to guide future directions for research. Methods We conducted a scoping review of HIV reduction interventions in LMIC healthcare settings using Embase, Ovid MEDLINE, PsycINFO and Scopus (through March 5, 2020). Information regarding study design, stigma measurement techniques, intervention features and study findings were extracted. We also assessed methodological rigor using the Joanna Briggs Institute checklist for systematic reviews. Results and discussion Our search identified 8766 studies, of which 19 were included in the final analysis. All but one study reported reductions in stigma following the intervention. The studies demonstrated broad regional distribution across LMIC and many employed designs that made use of a control condition. However, these strengths masked key shortcomings including a dearth of research from the lowest income category of LMIC and a lack of interventions to address institutional or structural determinants of stigma. Lastly, despite the fact that most stigma measures were based on existing instruments, only three studies described steps taken to validate or adapt the stigma measures to local settings. Conclusions Combating healthcare stigma in LMIC demands interventions that can simultaneously address resource constraints, high HIV burden and more severe stigma. Our findings suggest that this will require more objective, reliable and culturally adaptable stigma measures to facilitate meaningful programme evaluation and comparison across studies. All but one study concluded that their interventions were effective in reducing healthcare stigma. Though encouraging, the fact that most studies measured impact using self‐reported measures suggests that social desirability may bias results upwards. Homogeneity of study results also hindered our ability to draw substantive conclusions about potential best practices to guide the design of future stigma reduction programmes.

disease progression [15,16]. PLWH experiences of enacted healthcare stigma are widely documented across the globe. [17][18][19] Large-scale stigma reduction efforts will need to be a cornerstone of planned global efforts to end the AIDS epidemic by 2030 [20].
Researchers have traditionally divided stigma into internalized, enacted, anticipated and perceived stigma. Internalized stigma or "self-stigma" occurs when targets of stigma internalize the negative attitudes and perceptions projected onto them [21]. Enacted stigma refers to overt acts of discrimination and hostility directed at those perceived to have the stigmatized status, and anticipated stigma result from fear of enacted stigma [22]. Lastly perceived stigma pertains to how PLWH perceive their partners, friends, family and community to treat and view HIV and PLWH in general [6]. The various forms of stigma are often cyclical [23]: for example PLWH facing enacted stigma may be denied critical health services which may in turn validate their fears of anticipated stigma or expose them to negative experiences that deepen internalized stigma [21]. Interrupting stigma will therefore require interventions at multiple points in the cycle. Its inherently social nature will also require multi-level interventions that target not only individual-level behaviours but also change at the interpersonal, social, organizational and societal levels [24].
A rich history of HIV stigma interventions has been documented in four known reviews to date [25][26][27][28]. Most employ a narrative approach to highlight the positive intervention effects reported by the majority of included studies, while also pointing out that these gains mask shortcomings in terms of short duration of effects [26], lack of methodological rigor [25] and inconsistent measures used across studies [28]. The fourth review, the only meta-analysis, quantifies a significant but small effect across included studies to arrive at a similar conclusion regarding issues with study quality [27]. These four reviews are broadly inclusive of all studies on HIV stigma interventions regardless of the target population, intervention setting and global region. But their analyses do not focus on healthcare settings, the individuals who work in these settings, or on high priority regions.
This scoping review aims to expand on past reviews by assessing the design and impact of stigma reduction interventions conducted in healthcare settings of low-and middle-income countries (LMIC). It also examines the methodological quality of these studies with particular attention to how stigma was measured. We opted to conduct a scoping approach (as opposed to other approaches such as a systematic review) in order to characterize the nature of evidence for these stigma interventions [29]. The substantial heterogeneity in approaches to measuring stigma and the interventions for reducing it presented challenges in specifying a clearly defined question (e.g. "does this strategy effectively reduce this specific outcome?") also made a scoping review the more practical choice [30]. We focus on healthcare stigma in LMIC because of the unique combination of factors that shape stigma in these settings such as heavier HIV burden [31,32], less developed primary healthcare systems [33,34], and the layering of HIV stigma with discrimination towards other marginalized identities such as sexual minorities [35], commercial sex workers [36] or people who inject drugs [37]. In such settings, pursuit of global "best practices" for combating stigma can miss opportunities inadvertently results in the direct importation of western-style interventions. By investigating studies in LMIC where the inherent challenges of stigma reduction are further compounded by resource constraints and other limitations, this review describes progress to date in the control of stigma in settings where these programmes may have the greatest impact.

| METHODS
The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Guidelines were followed to conduct the review and analysis [30].

| Search strategy
A public health librarian (author SLH) created the literature search strategy after meeting with two members of the research team to clarify goals and further define selection criteria. The search strategy was built and tested for sensitivity in Embase using Emtree subject headings and keywords, and the search strategy was translated to three other databases: Ovid MEDLINE, PsycINFO and Scopus. There were no language or time restrictions placed on the search, which was conducted on 29 to 30 November 2018. Several additional texts were added in the course of revisions up through 5 March 2019.
Four conceptual domains were used to build the list of relevant search terms: health outcome (i.e. HIV), region (i.e. LMIC), key topic area (i.e. stigma in healthcare settings) and study design (i.e. intervention). Terms selected for each domain to include widely used language, acronyms and phrasings common in the medical and health sciences literature. Search terms defining the regional scope of the paper included a list of the 164 countries defined as low, low-middle and upper-middle income by the World Bank [38], as well as inclusion of a set of search terms commonly used to describe LMIC published by the Health Sciences Library at the University of North Carolina [39]. Terminology related to stigma and interventions were tested and reduced down to the terms in the Appendix S1 to reduce the sensitivity and increase the specificity of the search. Lastly, we screened the reference lists of selected articles to locate other potentially relevant studies and scanned the Internet for relevant grey literature including non-peer-reviewed reports or documents. Specific search terms used in the Embase search strategy are provided in the Appendix S1.

| Study selection
Authors MKS and RHX reviewed the resulting titles and abstracts to identify articles that: (1) specified reduction in HIV-related stigma in healthcare settings as at least one of the study aims, (2) included healthcare workers as members of the population targeted by the intervention and (3) took place in an LMIC as defined in the previous section. Note that while our topical interest centred on enacted stigma, we applied this criteria liberally in order to include studies that described issues of stigma in the context of patient-provider encounters, regardless of whether the authors explicitly used the term "enacted stigma. " We defined healthcare workers as professionally trained medical personnel such as doctors, nurses, physician's assistants, pharmacists, but not situationally trained lay workers such as community health workers or volunteer support staff. We drew this distinction because professional healthcare workers hold disproportionate power over their patients by way of their authority to prescribe (or withhold) life-saving drugs, to perform (or abstain from) necessary medical procedures, or to refer patients to other medical specialists. Though practice around the inclusion of conference abstracts in scoping reviews vary, our team opted to include them in order to address the publication bias that tends to favour more frequent and more rapid publication of reports with positive results [40].
Next, the same two authors conducted a full-text review to further determine its appropriateness for inclusion in the final analysis. In cases where multiple articles described the same intervention, only the reference with the most complete information was retained. All aspects of the review and text screening were carried out by authors MKS and RX independently in the software program Rayyan [41]. Discrepancies in decisions regarding article inclusion were resolved through discussion. Figure 1 shows the flow diagram for study selection.

| Data extraction and bias assessment
Author MKS created an initial draft of a data extraction chart containing study features of interest a priori. Authors MKS and RHX then assessed chart utility by conducting data extraction for the first five articles, after which they compared results and amended the chart to better reflect study features and characteristics relevant to the analysis. A full data extraction was then carried out independently on each article by MKS and RHX, who met for a final time to resolve any differences in classifications.
The following data were extracted: study region, specific target population, sample size, study design, type of stigma measure, intervention characteristics and primary results. Approaches for stigma measurement were categorized into the three domains as identified by Nyblade et al. [42] and included providers' awareness about the negative impacts of stigma, their knowledge about HIV transmission as it relates to fear of occupational exposure, moral or value-based associations they have of HIV with socially taboo behaviours, and enacted stigma (i.e. actual measures of discriminatory behaviours on the part of providers).
Interventions were then categorized in three ways. First we determined the levels at which changes were sought: individuallevel strategies such as improving provider awareness of stigma versus institutional-level changes such as implementation of guidelines for universal precautions. Second, we classified interventions using the four intervention types defined by Brown et al. [26]: information-based training, skills building, contact with affected groups such as PLWH, and institutional changes such as altering clinic policy around PLWH care (e.g. ending the practice of segregating inpatients by HIV status). Last, we conducted a risk of bias assessment for all studies in the final sample set using the Joanna Briggs Institute checklist for systematic reviews [43]. Checklists specific to randomized control trials (RCT) and quasiexperimental studies were applied separately to studies of corresponding design. Responses to each criteria were assessed using "yes" if criteria were met, "no" if they were not met, "N/A" if the question did not apply to the particular study, and "unclear" if the information could not be determined from the available text. We then assigned a final "overall rating" to each study based on the following algorithm: studies failing to meet more than two criteria were classified as "poor, " those that failed to meet exactly two criteria as "fair, " and less than two as "good. " Criteria classified as "N/A" or "unclear" were treated as equivalent to half of a "no;" for example a study whose criteria rating included a single "no" and two "unclears" would receive an overall rating of "fair. "

| RESULTS AND DISCUSSION
The initial search produced 8766 results (Embase = 2605; Ovid Medline = 2215; PsycINFO = 1089; Scopus = 2857), and after removing duplicates, 4369 studies remained. Review  of the titles and abstracts identified 80 studies for full-text examination, after which 19 studies were found to meet criteria regarding the outcome of interest (enacted healthcare stigma), the study population (healthcare workers) and study design (intervention; Figure 1). Most of the studies were published as full articles in peer reviewed journals [44][45][46][47][48][49][50][51][52][53][54][55][56][57], with the remainder published as conference abstracts [58-60] or public reports [61][62][63] (Table 1). According to the World Bank geoscheme [89], a third of studies were located in sub-Saharan Africa [45,48,57,59,61,63,83] followed by South Asia [44,54,55,58,60,62]. Among the 10 studies located in East Asia and the Pacific, six took place in China and India which account for over half of people living with HIV in the region [50][51][52]75]. Only one study was conducted in the Middle East and Northern Africa [53] (Figure 2). Additionally, two studies had sites in multiple countries, most of which were collectively located in sub-Saharan Africa [47], but one of which also had a single site in Latin American and the Caribbean [56]. The most common study design was a single group pre/post-test comparison [44,45,47

| Intervention strategies and impact
Interventions conducted by every study in this review focused on the individual-level changes which included strategies such as improvement of HIV knowledge or reduction in discriminatory attitudes. Among these, six studies [44,50,59,60,62,63] additionally included strategies focused on institutional-level change, such as implementation of universal precautions, provision of preventive medical equipment like gloves or sharps containers and reversal of discriminatory clinical practices such as marking patient medical records to signal their HIV status to colleagues.
Using the classification strategy introduced by Brown et al., every study in this review was found to have employed at least two strategies as part of the intervention, with some using three [53,54,75] or at most four [59,62] approaches in the same study. Information-based training was included in every study, an approach most commonly deployed through didactic lectures or group discussion. Skills-building was used in all but three of the studies [47,55,60] and involved techniques such as role playing, group brainstorming or sharing of personal experiences. Contact with affected groups was used in less than half [47,[53][54][55]58,59,62,75] of studies, and institutional changes such as desegregation of patients living with HIV from other patients or implementation of universal precautions was used in five studies [44,50,59,60,62]. Physicians and nurses made up the intervention target population of every studyincluding students still in training for medicine and nursingamong which half [44,46,48,49,51,57,59,[61][62][63]75] also included nonclinical staff such as laboratory technicians or administrative staff in their intervention activities.
Regarding impact, all but one study [47] reported that their intervention was effective in reducing HIV stigma in healthcare settings. However, in 10 of the studies [44,45,48,[55][56][57][58][59]63,76], the overall conclusion was drawn from selectively focusing on the subset of stigma scale items where a

| Stigma measurement
As a function of our study inclusion criteria, all interventions in this review addressed stigma enacted by healthcare providers in healthcare settings. Yet measures varied widely in terms of measurement development, the groups surveyed and the domains assessed (i.e. stigma awareness, HIV knowledge, value-based judgements and performance). A summary of measure characteristics is provided in Table 3.
In terms of stigma scale development, about half used scales or indices developed by the authors themselves in the course of the study [46,48,50,51,[57][58][59]61,63]. Of these, three described steps taken to validate the measure, whether by conducting psychometric analyses of the scales [46,57] or by pilot-testing questions with local community members [48]. The remainder of studies relied on existing stigma measurement tools which were either used "as is" [44,45,47,52,55,56,60,75] or with modifications informed by consultation with local experts [53,62,63] or by pre-testing the instrument in target communities [49,54].
The remaining three studies that surveyed providers approached them as potential witnesses to discriminatory behaviours taking place in their own workplace using the HIV/ AIDS Stigma Instrument (HASI) [45,47,56]. The next most common group surveyed were patients who were asked either as clients evaluating the quality of care received at study clinics [44,60], as witnesses to discriminatory treatment in clinics [47], or as future patients in term of their willingness to seek HIV-related care in a study clinic following the intervention [51]. Each of these four studies utilized patient-related measures in conjunction with those for providers. In two final studies, study investigators and staff directly observed provider behaviours to assess evidence of stigma reduction. Allam et al. [60] observed the persistence of recommended practices in study clinics such as removing HIV status from patient case records or integrating hospital wards for inpatients with and without HIV. Li et al. [50] used an unannounced standardized patient approach to observe and compare provider behaviours before and after the intervention.

| Quality assessment
Appraisal of study quality was restricted to full reports (i.e. all but the three conference abstracts [58-60]) and conducted  separately depending on whether it was an RCT or quasi-experimental study. The most common weakness of the three RCTs included in this review [46,48,75] (Table 2A) was the lack of blinding of participants or study staff to treatment allocation, an expected result given the nature of behavioural interventions. Among the 15 quasi-experimental studies (Table 2B) the most common criteria that studies failed to meet was inclusion of a control group [44,45,51,52,54,57,62] and completeness or the reporting of participant follow-up over time [45,[51][52][53]57,60,63]. Many of the quasi-experimental studies also lacked adequate detail to inform appraisal by our reviewers (MKS and RHX), particularly in regards to reliability of the outcome measures. It is worth noting that the Li et al. [50] study using standardized patients, though classified here as a quasi-experimental design, was conducted in the context of a larger RCT that evaluated the intervention using survey-based stigma outcome measures [90] but was not included in this review to avoid redundancy.

| Summary of findings
This scoping review describes a body of rich body of research by adding 10 new studies to the collective body of research identified by earlier reviews on this topic [25][26][27][28]. This body of research exhibits some key strengths. First, the 21 studies included in this review represent LMIC across five of the seven World Bank regions [38], nine of which were located in or had at least one site in sub-Saharan Africa [45][46][47][48]56,57,59,61,63] which bears 70% of the global burden of HIV [91]. Regarding study design over half of the sample used a control condition [46,49,53,55,56,58,61,63], including three RCTs [46,48,75]. Every intervention in this review employed at least two intervention components, and seven studies utilized stigma measures that had been developed or modified to account for the local cultural context. Lastly, all but one [47] of the 21 studies included in this review reported a significant effect of the intervention in reducing stigma in healthcare settings. Though the consistency of positive results across these studies is encouraging, several features of these studies as a group merit discussion. First, we observed a dearth of studies taking place within the lowest income category of LMIC, as well as in Eastern Europe & Central Asia and Latin America & the Caribbean, regions where HIV incidence is rising [92,93] or persistent at high levels [94,95]. Second and in terms of interventions, relatively few (N = 6) sought to address institutional or structural determinants of stigma [44,50,59,60,62,63]. And although the fact that many interventions combine multiple components can be seen as a strength, this can also hinder investigators' ability to identify the most impactful activities, an insight which could inform stigma control efforts under budget constraints. Third, methodological quality across studies varied widely, with only five studies ranked as "good" [49,50,55,56] and over half ranked as "poor" [44][45][46]51,52,54,[58][59][60]62,96]. Lastly, the homogeneity of the findings that nearly all tested intervention were effective, while encouraging, may be a byproduct of the upward bias that can be induced with the use of self-reported outcomes that trigger social desirability bias. The uniformity of results also hindered our ability to draw substantive conclusions about specific best practices to inform future stigma reduction programmes. Qualitative assessment of the relative magnitudes of effect may potentially inform a few such insights, though the one meta-analysis on this topic conducted by Mak et al. cited a number of challenges to synthesizing findings including the "diversity of outcome measures used" [27].
The inherent challenges of stigma measurement are a topic of longstanding debate [22,42,77,97] and merit special consideration. As a multifaceted phenomenon, stigma measurement demands complex tools to operationalize its myriad domains. Yet the tremendous diversity in the measures across the studies in this reviewscale components range from three [75] to 96 [54] itemsundermines our ability draw greater insight into the state of stigma reduction as a whole. In addition, measurement of some stigma domainsnamely HIV knowledge and stigma awarenesscommonly rely on use of hypothetical questions. This technique tests the underlying assumption that stigma is driven by ignorance of its effects on its targets or by fear of accidental transmission. But hypothetical questions are likely a poor proxy for how respondents behave in real life [97] and can lead to ambiguous interpretations, particularly in certain language groups and cultures [98]. Lastly, though enacted stigma is challenging to measure given low provider willingness to self-report discriminatory behaviours and their tendency to alter behaviours under observation [42,99], it can also provide the most substantive metric to assess whether stigma reduction interventions achieved their goals. Several measures of enacted stigma identified in this review sidestepped the problem of social desirability bias by using techniques other than self-reported measures, whether by asking providers [45,47,56] and/or patients [44,45,47,60] to bear witness to enacted stigma, or by deploying standardized patients to covertly observe provider behaviours [50]. Though these measures are not without their own limitationsfor example the HASI-N may inadvertently solicit multiple reports of the same stigma incidenttheir novelty and potential for reducing social desirability bias may motivate more research to further refine and improve techniques like these. A summary of these techniques and their potential limitations are provided in Table 3.

| CONCLUSIONS
Our findings highlight several important knowledge gaps to guide future research directions. First, more research is needed to push the development of objective, reliable and transportable stigma measures to facilitate more meaningful programme evaluation and comparison across studies. Second, a common design weakness of studies in this review is the inability to blind participants or study staff to the intervention during the trial (see Table 2A). The modified Zelen design [100,101] may offer a partial solution to this problem by not informing control arm participants of their trial participation until the study end. Finally, though many incorporated studies address attitudes towards key populations or HIV-related risk behaviours (e.g. injection drug use, commercial sex, same sex behaviours), they do not do so in a systematic way that could distinguish it from stigma towards HIV infection alone. A way to disentangle the components that make up intersectional stigma will be necessary in order to address its myriad manifestations.
Nearly 40 years into the epidemic and with wider availability of treatment and prevention tools, the persistence of HIV stigma in healthcare settings remains a matter of great urgency. Moreover, combating healthcare stigma in LMICs will require particular attention to the unique combination of resource constraints, high HIV burden and diverse cultural contexts that shape stigma [102][103][104][105][106][107][108]. More recent proposals by Hatzenbuehler et al. [109] to regard stigma as a driver of health inequity may inform a social determinants framework to mobilize more novel and interdisciplinary approaches to stigma reduction. By focusing on research conducted in LMIC, this review highlights research crucial to informing impactful and sustainable programmes to tackle the underlying drivers of stigma.