Stated-preference research in HIV: A scoping review

Discrete choice experiments (DCE), conjoint analysis (CA), and best-worst scaling (BWS) are quantitative techniques for estimating consumer preferences for products or services. These methods are increasingly used in healthcare research, but their applications within the field of HIV research have not yet been described. The objective of this scoping review was to systematically map the extent and nature of published DCE, CA, and BWS studies in the field of HIV and identify priority areas where these methods can be used in the future. Online databases were searched to identify published HIV-related DCE, CA and BWS studies in any country and year as the primary outcome. After screening 1,496 citations, 57 studies were identified that were conducted in 26 countries from 2000–2017. The frequency of published studies increased over time and covered HIV themes relating to prevention (n = 25), counselling and testing (n = 10), service delivery (n = 10), and antiretroviral therapy (n = 12). Most studies were DCEs (63%) followed by CA (37%) and BWS (4%). The median [IQR] sample size was 288 [138–496] participants, and 74% of studies used primary qualitative data to develop attributes. Only 30% of studies were conducted in sub-Saharan Africa where the burden of HIV is highest. Moreover, few studies surveyed key populations including men who have sex with men, transgender people, pregnant and postpartum women, adolescents, and people who inject drugs. These populations represent priorities for future stated-preference research. This scoping review can help researchers, policy makers, program implementers, and health economists to better understand the various applications of stated-preference research methods in the field of HIV.


Introduction
Discrete choice experiments (DCE), conjoint analysis (CA) and best-worst scaling (BWS) are quantitative methods for estimating individuals' stated preferences for products and services [1][2][3][4]. These methods, which are based in economic/marketing theory, have been used widely in research pertaining to marketing, transportation, the environment, and other fields [5]. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 DCE, CA and BWS have also been applied to health economics and medical research to elicit preferences from patients, providers and policy makers, covering broad themes such as delivery of health services, resource allocation, development of outcome measures, prophylaxis and treatment products, and employment [6][7][8][9][10].
Stated-preference methods have also expanded into the field of HIV research, having been used to elicit preferences and attitudinal obstacles for HIV testing, service delivery, antiretroviral treatment (ART) and HIV prevention products [11][12][13][14][15][16][17]. Given the increasing emphasis on patient-centered approaches to HIV care delivery that are adaptive to health systems and their resource constraints, the potential for stated-preference research to address priority HIV research areas is significant [18][19][20][21]. However, it is unclear how DCE, CA and BWS methodologies have been applied in the HIV research field to date and what priority areas exist in which these methods can best be used to better understand the HIV epidemic. The objective of this scoping review was to systematically map the extent and nature of published DCE, CA, and BWS studies in the field of HIV research and identify priority areas where these methodologies can be used in the future. We chose to focus on HIV in light of the magnitude of the global HIV epidemic which has resulted in a substantial and diverse body of HIV-specific literature, our prior HIV research experience, and the current status of HIV as a chronic disease requiring a longitudinal, patient-centered care model that can be informed by stated-preference research.

Synopsis of stated-preference methods
There are several key steps and concepts underpinning stated-preference research methods, which are reviewed in depth elsewhere and summarized here as follows (Fig 1) [7,20]. First, DCE, CA and BWS are subtypes of stated-preference research that utilize conjoint measurements (i.e. measurements taken on all parameters simultaneously) to draw comparisons between sets of defined alternatives [22]. DCE is really a distinct conjoint approach different from other CA methods although it is sometimes referred to as choice-based CA [23]. In our review, we use CA to refer to non-discrete choice conjoint analytic approaches. (Contingent valuation is another subtype of stated-preference research that is not discussed in this review.) CA methods include ranking and ratings-based subtypes. In ratings-based CA, each scenario is rated independently (in theory) of the other scenarios, while in rankings-based CA, the evaluations are not independent of each other. In DCE, respondents consider multiple scenario configurations simultaneously, and in BWS (also called MaxDiff analysis), respondents must choose the 'best' (i.e. most preferred) and 'worst' (i.e. least preferred) options among a choice task containing at least three alternatives. There are three subtypes of BWS (i.e. object, profile and multi-profile case) that differ in the complexity of the items under consideration (Fig 2) [24].
For all types of stated-preference research, the investigator must first identify and present the characteristics (known as "attributes") that compose each hypothetical set (known as "choice sets") that will be presented to the study participants. Each choice set contains various attributes which in turn contain various options or increments (termed "levels"). The choice sets selected by participants thus represent their preferences over other choice sets (i.e. hypothetical alternatives). In cases where the number of attributes and levels make for an unreasonably large number of unique choice sets for a participant to select from, a fractional factorial design is often employed to reduce the number of choice sets while maintaining the statistical integrity of the model [20]. Responses are then analyzed to quantify the relative importance of each attribute, how/whether the participant's preferences are influenced by the attributes, and the trade-offs participants are willing to make between varying hypothetical alternatives [20]. As one would expect, the selection of attributes and levels is critical to study design in order to accurately reflect the preferences that would be made by a population in a real-world context [25].

Protocol and registration
The protocol was developed by our research team using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews (PRISMA-ScR) [26]. The final protocol was registered with the Open Science Framework (https://osf.io).

Eligibility criteria
We utilized the following eligibility criteria to broadly map the literature on stated-preference research in HIV: 1) the target study population is people living with HIV (PLWH) or persons at risk of HIV infection (as identified by the authors of the report), healthcare workers interfacing with PLWH, or policy makers addressing HIV-related issues; and 2) DCE, CA, or BWS methodology was used to elicit and analyze preferences of the study population. All published manuscripts and conference abstracts published in any year, language, or country were included. Reviews and opinion pieces were excluded.

Outcomes
Scoping reviews are a relatively novel form of knowledge synthesis that follow a systematic approach to map the evidence on a topic to identify key concepts, theories, sources and knowledge gaps [26]. Using this approach, the primary outcomes of our scoping review were to determine the extent (i.e. number of studies), range (i.e. variety of study types), and nature (i.e. characteristics) of eligible studies.

Data sources and search strategy
We conducted a systematic search following the Cochrane Collaboration guidelines and report our findings using the PRISMA-ScR (S1 Table) [26-28]. Our search criteria was informed by recent systematic reviews of the literature of health-related, stated-preference studies (S2 Table)

Study selection
Titles and abstracts were imported into Endnote X8 (Thompson Reuters, Philadelphia, United States), duplicates were removed, and they were screened by two authors (John Humphrey and Katherine McDonald) with potential eligibility determined by consensus with a third author (Gregory Zimet) when eligibility was unclear. Full texts of potentially relevant records were retrieved and assessed for eligibility. Reference lists of all potentially eligible articles and reviews were also searched for additional titles relevant to the search, as well as to search for methodological details missing from the included report, as necessary.

Critical appraisal of individual studies
Eligible studies were included in the review regardless of their methodological quality or risk of bias. However, a quality assessment was conducted to gain a fuller understanding of the nature of the available evidence. Given a lack of a standard quality or bias assessment tool for stated-preference studies, a customized quality assessment tool was created utilizing domains from the Lancsar and Louviere 2008 and the Quality Assessment Tool for Observational Cohort and Cross-Sectional Studies published by the National Heart, Lung, and Blood Institute (NHLBI), and pilot tested on five random studies [3,34]. The following domains were assessed: reported methodology concordance (e.g. that studies reporting using CA are using CA and not DCE) [23]; participation rate (< or � 50%); recruitment method; whether inclusion/exclusion criteria were explicit; whether a sample size justification, power description, or variance and effect estimates were provided; design type (full or fractional factorial); whether a forced choice was used and if so, whether a justification was provided; method of profile generation and allocation to choice sets; whether respondents were randomly allocated to versions; whether coverage of attributes and levels was checked via piloting; whether understanding and complexity was checked via piloting; and, how data were collected (e.g. face-to-face, phone-, self-or computer-administered questionnaires). Checklist items that were not present in the report (either because the item in question was not performed or not reported by the authors) were scored 'n/s'. The critical appraisal data extraction was conducted by two of the authors (John Humphrey, Katherine McDonald) for each study through a separate process following the initial data extraction. These data were then summarized in a table for all full-text articles. Although all reports were included in the critical appraisal, conference abstracts were excluded from this summary table given that the word limits of abstracts may have precluded descriptions of all of the items in our checklist.

Search results
The selection process based on PRISMA guidelines is illustrated in Fig 3.[28] The search yielded 1,496 citations, 57 of which were included in the study following the screening process. Table 1 shows a summary of included studies organized by category (HIV prevention, HIV counselling and testing, HIV care and service delivery, and ART) followed by year(s) in which each study was conducted (or published, if the study year(s) were not reported), from earliest to most recent.

Characteristics of studies
Studies were conducted in 27 countries from 2001-2017, with the majority conducted in the United States (30%) and South Africa (12%), while Germany, United Kingdom, and Thailand each contributed 5-6 (9-11%) studies ( Table 2). The frequency of published studies reported during three-year periods from 2000 to 2017 also increased over time (Fig 4). The largest proportion of studies covered themes relating to HIV prevention (44%), while other studies covered HIV counselling and testing (17.5%), HIV care and service delivery (17.5%), and antiretroviral therapy (21%). DCE, CA and BWS methods were used in 63%, 37% and 4% of studies, respectively, and most studies did not report using any form of probability sampling. PLWH comprised 35% of respondent groups. People who were HIV-negative but considered at risk of HIV were the most common population studied overall (46%) and included adolescents and young women, members of serodiscordant couples, adults with other sexually-transmitted diseases, and male truck drivers. Participant groups from key populations included men who have sex with men (MSM), female sex workers, transgender women, and people who use drugs. The median sample size was 288 participants, and 59% of studies sampled between 100 and 500 participants. The most common methods of attribute determination were literature review and key informant interviews; 56% of studies reported using � 2 methods to determine attributes. The number of attributes generated ranged from 3 to 13, with 74% of studies selecting between 5 and 7 attributes. The number of choice sets ranged from 3 to 49, with 79% administering 8-16 choice sets. Stated-preference research in HIV

HIV care and service delivery
The majority of studies in this category were conducted in low and middle-income countries including South Africa, Ghana, Thailand, Malawi, Ethiopia, Mozambique, and Vietnam. Four studies assessed reproductive health preferences, including the attributes that influence facility choice among HIV-infected women of childbearing age in Ethiopia and Mozambique [13], and fertility planning for women, youth, and others affected by HIV [71][72][73]. Other studies assessed how to configure health services for PLWH more generally. In one study from South Africa, cost, staff attitude, wait time, and clinic branding constituted major barriers to ART uptake and adherence in resource-poor settings [68]. In the United Kingdom, preferences for shorter appointment waiting times, longer opening hours, and the type of HIV care provider (general practitioner vs HIV clinic) were prioritized [14], while flexible medical and psychosocial support were priority features according to policy makers and PLWH in Thailand [70]. Three studies specifically assessed preferences of policy makers regarding HIV program design, identifying preference for prevention interventions [74] among high-risk groups [70] and compared to other non-HIV related public health interventions [69].

Antiretroviral therapy
Eleven studies addressed patient and physician preferences for various attributes of ART, all of which were conducted in high-resource settings in North America and western Europe. Many of these studies addressed ART attributes that would optimize adherence and quality of life for patients, finding that avoiding major side effects [15,16,75,77,[81][82][83], long-term safety [16,80,82], treatment effectiveness [15,16,81], limited drug-drug interactions [81], and regimen convenience (e.g. tablet count and size, co-formulated tablets, dosing frequency) [15,75,76] were among the most important drivers of treatment choice for patients. Overall, the choice of ART was highly affected by patient preferences, with the majority of attributes studied being important to patients to varying degrees.

Critical appraisal of individual studies
The author-reported methodology was not concordant in 20% of studies, which in all cases were presented as CA but instead described a DCE (Table 3 and S3 Table). The eligibility criteria were explicitly reported in 69% of studies, while 10% of studies provided any sample size justification, power description, or variance and effect estimate. The majority (88%) of studies used a fractional factorial design and most (65%) did not specify whether a forced choice was used. The method of profile generation and allocation to choice sets was specified in 41% of studies, with Sawtooth (Provo, UT), SPSS (Armonk, NY), Ngene (Sydney, Australia) and SAS (Cary, NC) software used in 73% of these studies. Face-to-face interviews were used to collect data in 73% of studies, followed by web surveys (16%) and self-administered questionnaires (12%). Understanding and complexity of the questionnaire was checked (e.g. through piloting) in 47% of studies, and 59% of studies did not report that participants were randomly allocated to versions of the questionnaire.

Discussion
In our scoping review of DCE, CA and BWS research pertaining to HIV, 57 studies were identified covering a variety of themes relating to HIV prevention, care and treatment across diverse settings and populations. The majority of studies were DCEs (63%) followed by CA (37%). BWS was conducted in only two studies possibly reflecting the more recent introduction of BWS in health research [31]. Our review supports the increasing applications of statedpreference methods in the field of HIV research as well as the diverse uses of stated-preference research to advance knowledge about the global HIV epidemic. The studies in our review offer key lessons for HIV policy and service delivery. First, most studies addressed HIV prevention products, including pre-exposure prophylaxis, microbicides, and vaccines. Accelerating HIV prevention is a major target in the global HIV response, and understanding client preferences and attitudes for prevention products, which may differ across population groups, is critical to maximizing the uptake and impact of these products [53,87]. Clinical trials of pre-exposure prophylaxis and female prevention products (e.g. microbicide gels, vaginal rings) have identified adherence as a major factor influencing these products' effectiveness, and adherence is influenced by consumer preferences and attitudes [88]. Adherence is also a key determinant of ART effectiveness, and a number of studies examined patient attitudes and preferences towards ART. Multiple studies addressed the HIV care cascade, including HIV testing and service delivery. Reaching the UNAIDS 90-90-90 targets globally, in which 90% of people with HIV know their status, 90% of those who know their status are on ART, and 90% of those on ART achieve viral suppression, requires a thorough understanding of the barriers to achieving each of these targets in general and key populations. The stated-preference studies in our review contribute to this goal, providing insights about HIV testing preferences, key attributes of ART from the perspectives of PLWH, and preferences for HIV programs among PLWH and policy makers.
Our review sheds light on several priorities for future stated-preference research. First, only 17 of 57 (30%) studies were conducted in sub-Saharan Africa where the burden of HIV is highest, and none of these studies addressed patient preferences for ART. There is significant opportunity to conduct stated-preference studies in sub-Saharan Africa given that they are generally low risk, low cost and simple to implement. Second, as HIV care programs expand to serve increasing numbers of patients, there is a need to identify patient and provider preferences for patient-centered models of care that will improve efficiency, retention on ART, and viral suppression. Examples includes models of differentiated care for stable patients and patients requiring additional support, decentralized models of ART delivery, models of integrated HIV and maternal and child health care for pregnant and postpartum women and their HIV-exposed infants, and adolescent-friendly care delivery. Third, as novel pharmacologic agents and formulations become available for HIV treatment, prevention, and possibly even sustained HIV remission, understanding patient attitudes and preferences for these products will be needed to enable their delivery, optimize their uptake, and promote patient adherence. Fourth, key populations are a priority in the HIV response and traditional public health interventions may not be suitable to address these populations' unique needs and experiences. This includes the need to understand HIV care preferences for older children and adolescents who are cognitively capable of participating in stated-preference research but have seldom been included in such research to date. Finally, the critical appraisal also highlighted several quality areas that should be addressed in future stated-preference research. First, 20% of studies were misclassified as CA rather than DCE. Distinguishing between these two paradigms is important given their different approaches [23]. Second, methodological details (e.g. eligibility criteria, use of a forced choice, methods of profile generation and allocation to choice sets) were not specified in a high proportion of studies, and little more than half of studies reported checking participants' understanding and complexity of the choice tasks through piloting. Guidelines for conducting and reporting health-related CA, DCE and BWS studies have been published to help guide investigators in these areas [3,10,89]. Third, 74% of studies in our review reported using primary qualitative data collection (i.e. key informant interviews and/or focus groups) to develop attributes. Attribute selection is a critical step in stated-preference research, as the integrity of the choice sets depends entirely on the attributes used. Literature review alone may not yield an accurate representation of the experiences of the target population [90]. The generalizability of a study's findings must be considered in light of the methodology used to select attributes, as well as the method of choice set presentation (e.g. tablets, cards, drawings), sampling methodology and sample size [12]. Fourth, sample sizes varied widely (25 to 2,090 participants) and were not accompanied by a sample size justification or power description in 90% of studies. Power and minimum sample size can be difficult to calculate in stated-preference studies without precisely knowing the attributes, levels and initial estimates of the parameter values [89,91]. Nevertheless, minimum sample size can be estimated and doing so is important so that non-significant findings can be assessed in context of whether the study had sufficient power to detect a certain outcome in the first place [91]. Finally, over half of studies in our review did not report using probability sampling, which is important to acquire preference estimates that are representative of the population. However, probability sampling may not be feasible or desirable for studying key populations such as female sex workers, MSM or people who inject drugs.
Our review has strengths and limitations. Our systematic approach enabled us to assess the extent and nature of stated-preference studies in the HIV field as well as areas for future research. We did not search grey or non-English literature which may have provided additional articles. However, we did not find additional articles through our review of the references of included articles which supports the comprehensiveness of our search. We also did not summarize measures of preference heterogeneity or other more nuanced and key outcomes (e.g. probability of uptake, willingness-to-pay, utility scores), as the objective of this scoping review was to examine the extent and nature of studies rather than synthesize their notably heterogeneous findings particularly for readers who are less familiar with stated-preference research. Finally, this scoping review was a large undertaking and our results are only current up to February 2018.

Conclusion
Stated-preference research is emerging in the HIV field as evidenced by the increasing frequency of published studies over time. These studies cover diverse areas relating to HIV prevention, HIV counselling and testing, HIV care and service delivery, and ART. However, few studies were conducted in Sub-Saharan Africa or included key populations, which represent priorities for future research. These reviews can help researchers, policy makers, program implementers, and health economists to better understand the various applications of statedpreference research methods in the field of HIV.
Supporting information S1