Are client and provider preferences for HIV care coordination programme features concordant? Discrete choice experiments in Ryan White part A-funded New York City care coordination programmes

Published 2 weeks ago• 27 minute read

⁴Bureau of Hepatitis, HIV, and Sexually Transmitted Infections, New York City Department of Health and Mental Hygiene, Queens, New York, USA
⁵Division of Disease Control, New York City Department of Health and Mental Hygiene, Queens, New York, USA

Correspondence to Rebecca Zimba; rebecca.zimba{at}sph.cuny.edu

The New York City (NYC) HIV Care Coordination Programme (CCP) is designed to help people with HIV (PWH) overcome barriers to care and treatment engagement. We assessed preferences for CCP components among programme enrollees (’clients’) and providers. Our objective is to compare client and provider preferences, which were previously analysed separately.

We used a discrete choice experiment to assess preferences for four CCP features (‘attributes’): Help with Adherence to Antiretroviral Therapy (ART), Help with Primary Care Appointments, Help with Issues other than Primary Care and Where Programme Visits Happen. Each of these attributes had 3–4 variants (‘levels’). In the original surveys, levels within Where Programme Visits Happen varied by participant type (client vs provider). We recoded the levels by visit location (VL) or by travel time (TT) to make them comparable and report results from both approaches.

25 Ryan White Part A-funded NYC CCPs participated.

152 providers and 181 clients completed the survey.

Preferences were quantified using the relative importance of the attributes and utility of the levels.

From January 2020 to March 2021, 152 providers and 181 clients completed the survey. Most of the providers (52%) were <40 years, while most of the clients (60.2%) were ≥50 years. Almost half of the providers identified as Hispanic, whereas two-thirds of the clients (66.9%) identified as Black non-Hispanic. Most of the providers and clients identified as women (68.4% and 55.3%, respectively). In both the VL and TT analyses, clients were most influenced by Help with Adherence to ART (relative importance (RI) 30.5%, 95% CI 28.5% to 32.4% and 29.4%, 95% CI 27.5% to 31.4%, respectively), preferring medication reminders via phone or text, and Where Programme Visits Happen (RI 26.8%, 95% CI 25% to 28.6% and 32%, 95% CI 30.1% to 33.8%, respectively), preferring visits via phone or video chat. In the VL analysis, providers were most influenced by Help with Issues other than Primary Care (RI 26.9%, 95% CI 25.3% to 28.6%), valuing connections to specialty medical care, and by Help with Adherence to ART (RI 25.5%, 95% CI 23.5% to 27.5%), valuing directly observed therapy most highly. In the TT analysis, providers were most influenced by Where Programme Visits Happen (RI 28.2%, 95% CI 26.6% to 29.9%), preferring longer travel times, and Help with Issues other than Primary Care (RI 24.5%, 9%% CI 22.9% to 26.1%), again preferring connections to specialty medical care.

Client and provider preferences clearly diverged regarding CCP service intensity: in the aggregate, clients tended to prefer lower-intensity services, whereas providers endorsed higher-intensity services. These results highlight the importance of engaging clients as partners in decisions about programme services to facilitate alignment with client values.

Deidentified DCE data collected for this study can be made available on reasonable request to the first author at [email protected].

http://creativecommons.org/licenses/by-nc/4.0/

This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Ryan White Part A-funded (RWPA) HIV Care Coordination Programmes (CCPs) in New York City (NYC) aim to help the most vulnerable persons with HIV (PWH) to consistently engage in care and adhere to their HIV treatment regimens. The CCP incorporates outreach, case management, interdisciplinary case conferences, patient navigation, antiretroviral therapy (ART) adherence support and structured health promotion. These programmes are either co-located with primary medical care or formally affiliated with primary medical care providers. Originally launched in 2009 by the NYC Department of Health and Mental Hygiene (NYC Health Department),1 the CCP underwent revisions in 2018 to reduce challenges to implementation, improve client engagement and enhance the programme’s delivery and impact. Specifically, the revised model included changes to better reach those who could most benefit from the programme, facilitate client participation in the programme, simplify reimbursement for services by adopting a fee‐for‐service payment structure and permit more tailored care based on clients’ needs.2 3

That same year, the NYC Health Department and the City University of New York Institute for Implementation Science in Population Health launched the Program Refinements to Optimise Model Impact and Scalability Based on Evidence (PROMISE) study to evaluate the implementation and impact of revisions to the CCP (ClinicalTrials.gov ID NCT03628287).2 As part of the PROMISE study, we conducted discrete choice experiments (DCEs) among CCP clients and service providers to better understand their preferences for different components of the programme. In DCEs, participants are presented with a series of choices between sets of hypothetical programmes, services or products made up of predefined features that vary within and across choice sets. We have separately reported findings from the client and provider DCEs.3–5

Other studies have found discordance between the preferences of physicians and patients (or patients’ caregivers/guardians) for the treatment of various conditions, including asthma, diabetes, cancer and psoriasis.6–11 However, there is a dearth of research comparing the preferences of providers and clients of HIV care coordination or medical case management programmes for the support and services they provide or receive, respectively. Discordant client–provider preferences may indicate diverging opinions about the value of care coordination components, which could impact client satisfaction, provider–client communication and treatment adherence.12 13 Here we aim to compare CCP clients’ and providers’ preferences as measured through DCEs.

The target population and recruitment methods have been described elsewhere.3 4 14 Briefly, we conducted a census of NYC RWPA providers in the core CCP roles of patient navigators/health educators, care coordinators/case managers and programme directors or other administrators at the 25 CCP-implementing agencies. We invited all of these providers to participate using the survey URL and a unique identification code. CCP clients were eligible for the client DCE if they were aged ≥18 years, spoke English, Spanish or French as their primary or secondary language and were enrolled in the CCP at one of six partnering agencies. Clients were recruited by staff of the partnering CCP programme in which they were being served. Using information reported by CCP staff to the NYC Health Department, lists of eligible clients were generated and refreshed by the NYC Health Department during data collection and shared with CCP staff. Refreshing the lists ensured that eligible clients more newly enrolled in the CCP were able to be recruited and that clients who were no longer eligible or enrolled were not recruited. CCP staff provided clients with unique identification codes if they agreed to participate. All participants electronically provided informed consent and received a US$25 gift card upon survey completion. The NYC Health Department Institutional Review Board approved the study.

Our process for designing and implementing the DCE has been described elsewhere.3 4 14 In summary, we conducted focus groups with providers and clients to come up with a list of possible CCP features that could be included in the DCEs for both populations to facilitate comparison. The study team reviewed notes and transcripts from the focus groups internally and with the PROMISE Study Advisory Board. Our final DCEs included four features or ‘attributes’ in DCE design: Help with Adherence to ART, Help with Primary Care Appointments, Help with Issues other than Primary Care and Where Programme Visits Happen. Each attribute included 3–4 ‘levels’. Levels were illustrated with black-and-white graphics to facilitate comprehension and comparison across the choice alternatives. We customised the wording of the levels as needed for the client and provider audiences. See online supplemental file 1, figure S1 and table S1.

We designed the survey using Lighthouse Studio V.9.8.1 (Sawtooth Software, Provo, Utah, USA) and deployed the survey via Sawtooth’s online hosting platform. The DCE included 10 comparison tasks, with two hypothetical programme options per task. We used Sawtooth’s Balanced Overlap method15 16 to generate random tasks in which each level within an attribute appeared about the same number of times as the other levels within that attribute (ie, level balance), the same level within an attribute could sometimes appear in both hypothetical options in a single task and levels in different attributes appeared independently of one another (ie, orthogonality). The survey was deployed in English for providers and in English, Spanish and French for clients.

In each choice exercise, we prompted providers and clients: ‘Imagine that you had to choose between two programmes with the features below. Select the one that you would prefer’. Though the DCE presents mutually exclusive options in the DCE task format, all levels are supported by the CCP. Relatedly, we did not include a ‘None’ option, since everyone who took the DCE was either a client or provider of the CCP and would necessarily engage with some or all of the features present in each alternative. Additional descriptive and clinical data were collected after the choice exercises or merged from contract documents or other NYC Health Department records, including site-specific variables such as whether the site was co-located in a medical facility.

To calculate minimum sample sizes, we used n ≥(500c)/(ta), where n is the number of participants, 500 is the minimum number of times each main-effect level appears in the design, c is the maximum number of levels among all of the attributes, t is the number of choice tasks, and a is the number of options per task.15 16 Given a maximum of four levels, 10 choice tasks and two alternatives per task (c=4, t=10 and a=2), the minimum sample size is 100. We aimed to obtain 150 responses from providers and 200 responses from clients.

In early January 2020, we emailed personalised links to 227 eligible providers. The provider survey was closed in early March 2020. The client DCE remained open from late January 2020 until March 2021 due to challenges recruiting participants during the COVID-19 pandemic. A total of 884 clients at participating sites were eligible for the DCE over the course of the recruitment period.

We consider both clients and providers to be members of the public. As summarised above, clients and providers participated in focus groups, which helped us determine which CCP features or services should appear in the DCEs. Advisory board members, who are providers, also gave feedback on the design of the DCEs. Providers helped recruit participants for the client DCE, and many providers were themselves participants in the provider DCE. While the primary purpose of the DCEs was to obtain feedback from clients and providers to refine the design and conduct of the CCP, in which they are key stakeholders, the public was not involved in any other aspects of the design, conduct, report or dissemination plans of this research.

Aligning attributes and levels

The attributes and levels in both DCEs needed to be as similar as possible to facilitate comparison between the provider and client results. In the original DCEs, the levels were comparable for all attributes except Where Programme Visits Happen, which conflated travel time and visit location. Only one of the four levels in the Where Programme Visits Happen attribute, meeting by phone or video chat, was directly equivalent across populations. For the other three levels, the provider version assumed providers originated at the programme location and either remained at the programme location (no travel) or travelled to the clients’ homes (30 or 60 min). In contrast, the other three levels in the client DCE assumed clients started at their homes and either remained at home (no travel) or travelled to the programme location (also 30 or 60 min). We conducted two comparison analyses to address the lack of universal equivalence. As we were unable to find previous work to guide these analyses, we developed our comparisons in consultation with Sawtooth Software support staff.

In one analysis, we retained the original focus of the attribute (visit location) and combined the levels with the same location: the clients’ home, by phone or video chat and the programme location. In the other analysis, rather than focusing on visit location, we mapped levels across surveys according to travel time, that is, 30 or 60 min to meet at the client’s home (for providers) corresponding to 30 or 60 min to meet at the programme (for clients). Meeting at the programme (for providers) and meeting at home (for clients) mapped to no additional travel. The level for meeting by phone or video chat required no adjusting. See online supplemental file 1, figure S2.

Statistical analyses

We have described our analytical approaches elsewhere.3–5 Briefly, using a hierarchical Bayesian multinomial logit model in the Sawtooth Software Lighthouse Studio Analysis Module, we estimated part-worth utilities, which can be interpreted as measures of preference for levels within an attribute; utilities were effects-coded and zero-centred.17 18 Additional analyses were conducted in SAS V.9.4 (SAS Institute, Cary, North Carolina, USA). We calculated the relative importance of attributes, which are measures of attributes’ influence on choice behaviours, as the range of utilities for levels within that attribute divided by the sum of ranges of utilities for all attributes, expressed as a percentage.17 We summarised the ranges of relative importance across attributes, expressed in percentage points (pp); narrower ranges suggest that respondents were influenced similarly by all attributes, whereas wider ranges suggest respondents were influenced more strongly by some attributes than others. To assess the statistical significance of differences across study populations, we used the independent t-test for continuous variables and the χ² test or Fisher’s exact test for categorical variables, where applicable. Confidence limits were computed as the sample mean±(1.96×SE).

Sensitivity analyses

On 20 March 2020, New York State Governor Andrew Cuomo signed the ‘New York State on PAUSE’ executive order, and many of our partnering agencies transitioned to telehealth at or around that time.19 As a sensitivity analysis, we restricted the data to responses completed before the suspension of in-person services. We also examined differences in utilities and relative importance between the original analysis and two sensitivity analyses excluding poor-quality responses as indicated through straightlining (picking the right-hand or left-hand option for each task) and speeding (completing the choice tasks at a rate faster than 10 or 5 s per task).20 21 Results of all sensitivity analyses are reported in online supplemental file 1 (senstivity analyses section and figures S3–S10).

Software and data availability

Importances and utilities were estimated using Sawtooth Software’s Lighthouse Studio V.9.8.1. Additional analyses were done using SAS (release V.9.4) and R (V.4.3.2). We completed the Strengthening the Reporting of Observational Studies in Epidemiology checklist for cross-sectional study reporting (online supplemental file 2).22 Deidentified DCE data collected for this study can be made available on reasonable request to the authors.

Surveys were completed by 181 clients (20% of 884 eligible clients, January 2020–March 2021) and 152 providers (67% of 227 eligible providers, January–March 2020). The basic demographic characteristics for both groups are presented in table 1; all characteristics were statistically significantly different between populations. Most of the providers (52%) were aged under 40 years, while most of the clients (60.2%) were 50 or older. Almost half of the providers identified as Hispanic, compared with 28.2% of clients, and two-thirds of the clients (66.9%) identified as Black non-Hispanic, compared with 33.6% of providers. Most of the providers and clients identified as women (68.4% and 55.3%, respectively). All 152 providers and 140 clients (77%) completed the DCE before the suspension of in-person services due to the COVID-19 pandemic. Additional population-specific characteristics are summarised in our earlier papers3–5; notably, 92.3% of clients had been enrolled in the CCP for more than 1 year, 78.5% were virally suppressed, 68.5% were stably housed, and 26.5% had received directly observed therapy (DOT, including virtual DOT) services within the CCP to support adherence to daily oral ART regimens.

Table 1

Demographic characteristics of clients and providers

The average relative importances for each attribute for providers and clients are compared in figure 1. Panel A shows the relative importances when levels in the Where Programme Visits Happen attribute were combined by location. All the relative importance comparisons between provider and client responses were statistically significant. The most important attribute for clients was Help with Adherence to ART (30.5%, 95% CI 28.5% to 32.4%), followed by Where Programme Visits Happen (visit location) (26.8%, 95% CI 25% to 28.6%), Help with Issues other than Primary Care (24.1%, 95% CI 22.4% to 25.7%) and Help with Primary Care Appointments (18.7%, 95% CI 17.5% to 20%). The most important attribute for providers was Help with Issues other than Primary Care (26.9%, 95% CI 25.3% to 28.6%), closely followed by Help with Adherence to ART (25.5%, 95% CI 23.5% to 27.5%), Help with Primary Care Appointments (25.3%, 95% CI 23.8% to 26.8%) and Where Programme Visits Happen (visit location) (22.3%, 95% CI 20.7% to 23.9%). The range of importance for clients (18.7%–30.5%, 11.8 pp) was larger than for the providers (22.3%–26.9%, 4.6 pp). Help With Primary Care Appointments had the biggest difference in relative importance between clients and providers (6.6 pp lower among clients, −8.6 pp to −4.7 pp).

Figure 1

Attribute relative importances by analysis and population. (A) Where Programme Visits Happen combined by location. (B) Where Programme Visits Happen mapped by travel time. ART, antiretroviral therapy.

Panel B in figure 1 shows the average relative importances when levels in the Where Programme Visits Happen attribute were mapped by travel time. All the attribute importance comparisons between provider and client responses were again statistically significant. Among clients, the most important attribute was Where Programme Visits Happen (travel time) (32%, 95% CI 30.1% to 33.8%), followed by Help with Adherence to ART (29.4%, 95% CI 27.5% to 31.4%), Help with Issues other than Primary Care (20.8%, 95% CI 19.5% to 22.2%) and Help with Primary Care Appointments (17.8%, 95% CI 16.6% to 19%). Among providers, Where Programme Visits Happen (travel time) was also the most important attribute (28.2%, 95% CI 26.6% to 29.9%), followed by Help with Issues other than Primary Care (24.5%, 95% CI 22.9% to 26.1%), Help with Adherence to ART (24.2%, 95% CI 22.4% to 26.1%) and Help with Primary Care Appointments (23.1%, 95% CI 21.8% to 24.3%). The range of importance for clients (17.8%–32%, 14.2 pp) was again larger than for providers (23.1%–28.2%, 5.1 pp). Help With Primary Care Appointments again had the biggest difference in relative importance between clients and providers (5.3 pp lower among clients, −7 pp to −3.5 pp).

The part-worth utilities for clients and providers are shown in figure 2. Panel A shows the utilities under the scenario in which levels in the Where Programme Visits Happen attribute were combined by location. Comparisons of the utilities between providers and clients were all statistically significant, with several differences in both the magnitude and direction of preference between populations.

Figure 2

Part-worth utilities by analysis and population. (A) Where Program Visits Happen combined by location. (B) Where Program Visits Happen mapped by travel time. ART, antiretroviral therapy.

Within the Help with Adherence to ART attribute, clients preferred Medication reminders via phone or text (31.9, 95% CI 25.5 to 38.3), followed by Adherence assessment (10.8, 95% CI 4.9 to 16.7); they had a negative preference for DOT (−42.6, 95% CI −50.9 to –34.3). In contrast, DOT was the most preferred level for providers within the Help with Adherence to ART attribute (28, 95% CI 20.8 to 35.1), with lower preferences for Medication reminders via phone or text (−5.3, 95% CI −11.3 to 0.7) and Adherence assessment (−22.7, 95% CI −30.3 to –15.1). Within the Help with Primary Care Appointments attribute, clients preferred Appointment reminders and transportation (12.5, 95% CI 7.6 to 17.4), followed by Appointment reminders (0.9, 95% CI −3.9 to 5.6), with the lowest preference being for Appointment reminders and accompaniment (−13.4, 95% CI –18.5 to –8.3). However, Appointment reminders and accompaniment was the most preferred level for providers (21.4, 95% CI 15.4 to 27.3), followed by Appointment reminders and transportation (20.3, 95% CI 15.3 to 25.4) and Appointment reminders (−41.7, 95% CI −47.8 to –35.6). For the Help with Issues other than Primary Care attribute, Housing and food was the clients’ most preferred level (19, 95% CI 12.4 to 25.7), followed by Specialty medical care (6.2, 95% CI 0.6 to 11.8), Mental health and well-being (−12.4, 95% CI −17.7 to –7.1) and Insurance, benefits and paperwork (−12.8, 95% CI −17.8 to –7.9). In comparison, providers’ most preferred level was Specialty medical care (29, 95% CI 23.4 to 34.6), followed by Mental health and well-being (18.8, 95% CI 13.8 to 23.9), Insurance, benefits and paperwork (0.2, 95% CI −5.2 to 5.6) and Housing and food (−48, 95% CI −52.9 to –43). Finally, for Where Programme Visits Happen, clients preferred Meet via phone or video chat (19.3, 95% CI 12.9 to 25.8), followed by Meet at home (1.2, 95% CI −5.6 to 8.1) and Meet at programme location (−20.6, 95% CI −28.2 to –13). Providers had the highest preference for Meet at home (19.4, 95% CI 13.2 to 25.5), followed by Meet at programme location (5.3, 95% CI −0.9 to 11.5) and the lowest preference for Meet via phone or video chat (−24.7, 95% CI −30.2 to –19.1). DOT had the largest utility gap between clients and providers (70.6 lower among clients, 95% CI −81.6 to –59.6), followed by Housing and food (67 higher among clients, 95% CI 58.7 to 75.3) and Meet via phone or video chat (44 higher among clients, 95% CI 35.5 to 52.5).

The patterns of preference and utility estimates in Panel B of figure 2 are similar to those in Panel A, with the exception of the Where Programme Visits Happen attribute, here mapped by travel time. Meet via phone or video chat remained the most preferred level for clients (22.2, 95% CI 15.3 to 29), but the next most preferred level was Travel 30 min to meeting (13.2, 95% CI 7.8 to 18.5), then No travel (6.6, 95% CI −0.6 to 13.8) and lastly Travel 60 min to meeting (−41.9, 95% CI −49.7 to –34.2). For providers, Travel 60 min to meeting was the most preferred level (19.4, 95% CI 10.1 to 28.8), Travel 30 min to meeting was the next most preferred level (7.8, 95% CI 2.4 to 13.2), followed by No travel (1.3, 95% CI –5.4 to 7.9) and Meet via phone or video chat (−28.5, 95% CI −34.0 to –22.9). DOT again had the largest utility gap between clients and providers (66.4 lower among clients, 95% CI −77.2 to –55.5), followed by Travel 60 min to meeting (61.3 lower among clients, 95% CI −73.4 to –49.3) and Housing and food (60 higher among clients, 95% CI 52.9 to 67.1).

To our knowledge, this is the first study that has used a DCE to assess service provider–client agreement on non-clinical ART adherence support services. Other studies have largely focused on attributes of ART rather than ART adherence supports, and/or patient–physician comparisons rather than client-medical case management provider comparisons.23–25 A recent survey of preferences for ART maintenance services among providers (including clinicians) and HIV-positive clients of healthcare facilities in Thailand used a standard multiselect survey question format.26 These authors measured preferences for location, provider type and the frequency of ART refills, viral load testing, STI testing and psychosocial support, finding broad agreement for the most frequently selected category of each of the measures. This high level of agreement may derive from the type of survey question and the fact that respondents could pick all that apply; the focus on the evaluation of preferences for clinical services; and/or cultural context. Another study focused on the degree of agreement between physicians and patients for characteristics of ART (eg, side effects and dosing flexibility).23 Notably, these authors used a DCE to assess patient preferences for these characteristics and to assess what physicians thought patient preferences were (rather than physicians’ own preferences), and they also found a high degree of agreement for the ART attributes included in the survey.

Our study contributes to the literature by focusing on CCP providers and recipients and evaluating the concordance of their self-reported preferences for features of non-clinical adherence support services. In the present study, client and provider preferences for CCP service intensity clearly diverged: clients tended to prefer lower-intensity services such as medication reminders via phone or text and virtual visits, whereas providers endorsed higher-intensity services such as home visits and accompanying clients to primary care appointments. While the magnitude of the preferences changed depending on the focus of the analysis, the direction of the preferences remained consistent across analyses within each group of respondents for attributes that aligned across both groups. Notably, in the location-based analysis, clients’ preference for programme-based visits was strongly negative. However, in the travel-time-based analysis, clients had a positive utility for 30 min of travel and a strong negative utility for 60 min of travel, suggesting that clients did not mind the programme location, per se, but longer travel time. In contrast, providers appeared to be more influenced by location and willing to travel regardless of duration if it meant they could meet clients in their homes. As part of the ongoing evolution of the CCP, administrative budget allowances have recently changed so that CCPs can pay for cars (ie, taxis or rideshares) to get clients to their clinics, which may help alleviate the travel time burden for clients.

There are several potential reasons for the discrepancies identified between client and provider preferences. Taking the survey, participating providers likely considered the range of clients’ needs across their caseloads and may have been enacting a belief that higher-intensity services are more likely to help their most vulnerable clients achieve desired health outcomes. Home visits, in particular, provide valuable information to staff about clients’ living conditions and how those may promote or inhibit adherence to treatment. In contrast, clients who participated in the DCE were asked to consider only their own needs. Participating clients had been enrolled in the CCP longer than their non-participating peers4 and may not have perceived a need for such support for themselves or may have felt that higher-intensity services such as DOT or home visits were intrusive or inconvenient. Providers may also have been motivated to endorse features of the CCP that they perceived as clearly distinguishing the CCP from other forms of medical case management by design, such as home visits. Though clients and providers statistically significantly differed on age group, race/ethnicity and self-identified gender, it is unclear to what extent these differences might explain each population’s preferences.

The CCP allows for flexibility in practice based on client expectations; if a client prefers to meet somewhere other than their home, their patient navigator would accommodate this preference. Nevertheless, preference discordance could impact client satisfaction with services, retention in care and ART adherence,12 13 27–30 which could negatively affect both individual- and population-level HIV-related health outcomes. Providers aware of preference discordance may experience some degree of tension between their beliefs about what would be best for their clients and their belief in the importance of client autonomy.31 32 Providers may also experience job dissatisfaction if their clients prefer not to receive services that providers believe would benefit them or that are expectations of the CCP, like home visits.33 This could contribute to high rates of staff burnout and/or turnover and, relatedly, interruptions in client–CCP staff rapport and trust. However, the present study only examined aggregate preferences and did not evaluate the occurrence of preference discordance or potential consequences in client–provider dyads.

It is important to stress that this analysis compared aggregated preferences across populations and did not compare preferences within subgroups across populations. In our earlier work, latent class analyses identified subsets of clients who had higher preferences for higher-intensity services and providers who had higher preferences for lower-intensity services.4 5 Our sample sizes required that these segmented analyses be exploratory in nature, which is why we did not explicitly consider heterogeneity in the present comparisons. However, those results alongside the present analysis support the 2018 revisions to the CCP, which allow for differentiated service delivery, increasingly recognised as central to improving client retention and adherence.34 35 These findings may inform future revisions to the CCP as part of periodic contract re-solicitation, following review by the NYC Health Department and HIV Planning Council of multiple data sources, policy considerations and perspectives relevant to programme updates.

One strength of this study is our use of DCEs to model the complexity of real-world decision-making by asking participants to choose between sets of attributes and levels relevant to the CCP, rather than focusing on each programme feature in isolation. This study’s strengths also include the iterative and participatory approach to DCE design, involving both clients and providers of CCP services, as well as an advisory board and quality management and process improvement staff from the NYC Health Department; the careful consideration of the comparability of levels across the study populations; and the implications of each approach and the demonstration of how participant preference comparisons may highlight barriers or facilitators to intervention implementation.

We recognise the following limitations of our study. The prompts used in the client and provider DCEs were the same: ‘Imagine that you had to choose between two programmes with the features below. Select the one that you would prefer’. For clients, this prompt straightforwardly pertained to their preferences for themselves. For providers, it may have been unclear whether they were meant to make choices based on what programme they would prefer to deliver or what programme they think would most benefit their clients (ambiguity that has been observed in concordance analyses in the context of other conditions as well33–35). However, as discussed in our previous analysis of provider preferences, we interpret the findings from the provider DCE as an endorsement of the higher-intensity features of the CCP regardless of which perspective was motivating providers’ choices.12 Future studies should carefully calibrate the instructions to each audience to make sure the desired framing is clear for each type of respondent.

In addition, the preferences of participating NYC CCP clients and providers are not expected to be generalised to other jurisdictions. Nevertheless, we believe our results support differentiated service delivery, which has cross-cutting relevance, regardless of regional or other (eg, urban-rural) differences in specific population needs or characteristics. Lastly, the extent to which stated preferences, as measured via surveys such as DCEs, agree with revealed preferences, as indicated by actual choices and behaviours, is an area of ongoing inquiry.36–38 Future work analysing NYC Health Department records could add to the literature on this topic by measuring concordance between clients’ and providers’ stated preferences and actual service activity and identifying predictors of agreement or divergence between stated and revealed preferences.

Discordant preferences between clients and providers could potentially limit both populations’ engagement in the programme, detract from client–provider rapport and interfere with communication about care and treatment goals, which is central to effective medical case management care planning. These results highlight the importance of engaging clients as partners in decisions about supportive services to ensure such services are aligned with client values.

Deidentified DCE data collected for this study can be made available on reasonable request to the first author at [email protected].

Not applicable.

This study involves human participants and was approved by New York City Department of Health Institutional Review Board protocol number 18-009. Participants gave informed consent to participate in the study before taking part.