Preferences for transitional HIV care among people living with HIV recently released from prison in Zambia: a discrete choice experiment

Abstract Introduction No studies from sub‐Saharan Africa have attempted to assess HIV service delivery preferences among incarcerated people living with HIV as they transition from prisons to the community (“releasees”). We conducted a discrete choice experiment (DCE) to characterize releasee preferences for transitional HIV care services in Zambia to inform the development of a differentiated service delivery model to promote HIV care continuity for releasees. Methods Between January and October 2019, we enrolled a consecutive sample of 101 releasees from a larger cohort prospectively following 296 releasees from five prisons in Zambia. We administered a DCE eliciting preferences for 12 systematically designed choice scenarios, each presenting three hypothetical transitional care options. Options combined six attributes: (1) clinic type for post‐release HIV care; (2) client focus of healthcare workers; (3) transitional care model type; (4) characteristics of transitional care provider; (5) type of transitional care support; and (6) HIV status disclosure support. We analysed DCE choice data using a mixed logit model, with coefficients describing participants’ average (“mean”) preferences for each option compared to the standard of care and their distributions describing preference variation across participants. Results Most DCE participants were male (n = 84, 83.2%) and had completed primary school (n = 54, 53.5%), with 29 (28.7%) unemployed at follow‐up. Participants had spent an average of 8.2 months in the community prior to the DCE, with 18 (17.8%) reporting an intervening episode of re‐incarceration. While we observed significant preference variation across participants (p < 0.001 for most characteristics), releasees were generally averse to clinics run by community‐based organizations versus government antiretroviral therapy clinics providing post‐release HIV care (mean preference = –0.78, p < 0.001). On average, releasees most preferred livelihood support (mean preference = 1.19, p < 0.001) and HIV care support (mean preference = 1.00, p < 0.001) delivered by support groups involving people living with HIV (mean preference = 1.24, p < 0.001). Conclusions We identified preferred characteristics of transitional HIV care that can form the basis for differentiated service delivery models for prison releasees. Such models should offer client‐centred care in trusted clinics, provide individualized HIV care support delivered by support groups and/or peer navigators, and strengthen linkages to programs providing livelihood support.

can be provided to PLHIV in prisons, resulting in clinical benefits comparable to those seen in the community [9][10][11]. Unfortunately, such benefits are short-lived for PLHIV after release ("releasees") due to problems with post-release care continuity posed by multiple psychosocial, health system and structural barriers [6,[12][13][14][15]. Although data to quantify the extent of the problem in SSA are scarce, available reports suggest that one-third or more of releasees fail to link to community care or experience HIV treatment interruption post-release [16,17].
Despite these HIV care disruptions, no studies from SSA have attempted to assess HIV service delivery preferences among incarcerated PLHIV as they transition from prisons into the community, and specific recommendations about interventions to promote post-release HIV care continuity are lacking [18,19]. Descriptions of North American transitional care programs highlight service delivery elements that may be applicable to African settings, such as psychosocial support and treatment of co-morbid substance use disorders [20][21][22][23]. However, client preferences for similar services in SSA are, as of yet, unstudied.
We report here the findings of the first DCE, to our knowledge, conducted with releasees living with HIV in SSA. The aim of this study was to characterize releasee preferences for transitional HIV care services to inform development of a differentiated transitional care model to promote HIV care continuity for this population in Zambia.

M E T H O D S
This study represents a sub-study of the Releasee Care Continuum (RCC) study (#NCT02905162). Key elements of this sub-study are described below; additional details are presented in appendices (Appendix S1). Study activities were approved by institutional review boards of the University of North Carolina (#16-0276), University of Zambia (#001-02-16), James Cook University (#H6896) and University of South Carolina (#Pro00076701). All RCC participants provided written informed consent.

Study setting and population
To be eligible for the RCC study, participants had to be: incarcerated at one of five prisons in Lusaka or Central Province, Zambia; adults ≥18 years with documented HIV infection; scheduled for prison release within ∼30 days of study screening; enrolled in the national HIV treatment program [42,43]; and, if on treatment, receiving antiretroviral therapy (ART) for ≥30 days. RCC participants were recruited, screened and enrolled prior to release, and underwent one baseline study visit prior to release and one follow-up visit approximately 6 months post-release. Baseline socio-demographic and clinical information was collected at enrolment. The DCE survey was administered to consecutive RCC participants during the follow-up visit. Our DCE sampling frame was limited to 125 participants who had not died, moved out of the study area, become lost to follow up or previously completed a follow-up study visit during the DCE data collection period.

Discrete choice experiment
The DCE method is based on two theoretical foundations, Lancaster's theory [27] and RUT [25,26]. Lancaster's theory postulates that a person chooses a good or service based on its intrinsic characteristics or attributes. The combination of attributes gives rise to utility or value. RUT assumes that utility is a latent, unobservable variable; an individual's choice of a good or service follows a stochastic process in which the utility of the chosen alternative matches or exceeds that of all alternate goods or services considered. A DCE simulates real-world choice situations by asking participants to choose between systematically designed combinations of attributes. An analysis of participants' choices provides estimates of the contribution of each attribute to utility (value). A positive contribution to utility is consistent with a preference for a specific attribute. The DCE presented here was used to elicit the relative preferences for modifiable attributes of transitional HIV care services for releasees living with HIV in Zambia.

Attributes and levels
A list of preference-relevant attributes of transitional HIV care services was generated based on the extant literature [19,20,22,[44][45][46][47][48][49][50][51][52][53][54][55], preliminary results of a survey conducted with RCC participants about their transitional care experiences and results of five in-depth interviews (IDIs) exploring RCC participants' post-release needs and experiences. A review of IDI results and expert judgement by the study team were used to prioritize six attributes for inclusion in the DCE. Each attribute was described by two to four feasible values commonly referred to as "levels" (Figure 1). Attribute levels were described verbally and using graphical depictions [56]. Pre-tests with ∼50 individuals familiar with the transitional HIV care context for releasees in Zambia were used to iteratively refine the presentation of attributes and levels. Pre-tests helped refine the descriptions of transitional care services (including translations) and led to the exclusion of unfeasible combinations of transitional HIV care characteristics. Unfeasible combinations included, for example, a pairing of not being linked to a transitional care model, but receiving transitional care services nonetheless (Appendix S1).

Experimental design and choice format
After excluding unfeasible combinations, the attribute levels depicted in Figure 1 yielded 296 feasible transitional care options, including the "standard of care" (SOC). The SOC, available to all releasees living with HIV in Zambia, was defined as: (1) referral to a government ART clinic with (2) healthcare workers who focus on the health needs of PLHIV generally, (3) no linkage to a transitional care provider and no receipt of transitional care services and (4) no assistance with HIV status disclosure to partners at release. The experimental design of a DCE represents the subset of choice tasks (selected from 43,660 potential choice tasks containing two feasible transitional care options and the SOC) that is used to estimate preference parameters with the smallest possible error [57]. Ngene software (ChoiceMetrics 2017) version 1.12b [58] was used to identify a D-efficient design that was optimized for analysis using a mixed multinomial logit model with effects-coded, normally distributed priors (Appendix S1). The direction and relative magnitude of priors were informed by IDIs and expert judgement by the study team. The final design included 84 choice tasks split into seven blocks with 12 tasks each. Participants were randomized across blocks, with the order of choice tasks randomized within participants and the order of alternatives randomized within tasks. Following a best-best preference elicitation format [59], participants were asked in each task to identify their most preferred and next-most preferred transitional care options. The DCE survey was translated into the two most commonly spoken languages in Lusaka, Nyanja and Bemba, and programmed into a custom-built system for DCE administration on iOS tablet devices (Comet suite; Selway Labs 2019). A sample choice task is shown in Figure 2.

Survey administration
Trained local research assistants administered surveys in the participant's preferred language (English, Nyanja or Bemba) at a mutually agreed upon time and location. A paper survey was used to characterize participants' post-release experiences with healthcare and transitional care services (Data S1), while tablet devices were used to administer the DCE. First, participants were asked to read two short sentences to assess literacy [60], and visual acuity was assessed an E-test [61]. Next, participants were provided with verbal and graphical descriptions of attributes and levels. For each attribute, participants were asked to rank the respective levels according to their preference. To ensure participant comprehension of DCE choice tasks, the results of these unconditional rankings of attribute levels were used to dynamically generate one training task with one dominant alternative (combining the participant's preferred levels of all attributes) and one dominated alternative (combining less preferred levels of all attributes). Finally, each participant completed 12 DCE choice tasks. Additional questions assessed the difficulty of the survey and relative preferences for other support services not included in the DCE (for mental health, alcohol and drug use disorders).

Sample size
Our target sample size of n = 100 was based on a commonly used rule of thumb for estimating DCE sample size requirements [62,63]. The formula n ≥ 500 c∕ta combines the number of choice tasks t, alternatives per task a and the maximum number of levels per attribute c to derive a minimum sample size n. A priori, based on t = 12 choice tasks, a = 3 alternatives and c = 7 combinations of attributes C and D, we expected that ≥97.2 participants were required to characterize releasees' average transitional care preferences.

Statistical analysis
Participants' characteristics and unconditional rankings of attribute levels were analysed using descriptive statistics. DCE choice data, reflecting trade-offs between different combinations of attribute levels, were analysed using a mixed logit model [64], with the choice of an alternative as the binary dependent variable and characteristics of the transitional care option (i.e. attribute levels) as independent variables whose coefficients were assumed to be normally distributed. Coefficient estimates from the mixed logit model represent estimates of participants' average preferences for each transitional care characteristic relative to its reference level, which described the SOC. Coefficients' estimated standard deviations describe the variation in preferences across participants [65,66]. Standard deviations that are significantly different from zero indicate significant preference heterogeneity across releasees. Finally, individual participants' choices were combined with information on the distribution of preferences across participants to derive individual-level preference estimates ("posterior betas") for each attribute level [64,67,68]. The range of the estimated individual-level coefficients within attributes was compared across attributes to derive individual-level estimates of relative attribute importance. A preliminary analysis of systematic preference heterogeneity involved the estimation of a series of mixed logit models in which the main effect for each attribute level was iteratively re-estimated as fixed (instead of random), but with the model including an additional interaction between the attribute level and an observable characteristic. Statistical analyses were performed using STATA v16 (StataCorp, College Station, TX, USA).    Table 2 presents participants' preferred transitional care characteristics against characteristics of care they actually received post-release. Participants generally preferred referrals to government ART clinics (82.2%) and livelihood support (74.3%), and to be linked to either a support group (48.5%) or a peer navigator (48.5%), relative to no transitional care provider (3%). Substantial heterogeneity was observed in participants' rankings of characteristics of the transitional care provider, the focus of healthcare workers and disclosure assistance options. More releasees expressed a preference to have formerly incarcerated PLHIV (50.5%) be their transitional care providers rather than PLHIV without an incarceration history (30.7%), formerly incarcerated persons without HIV infection (1%) or persons without either an incarceration or HIV history (17.8%). When asked about their preferences for services other than HIV care and livelihood support services, mental health services were the most commonly preferred (40.6%), followed by support for alcohol (34.6%) and drug use (21.8%) disorders. Nearly half of participants (46.5%) preferred no assistance with voluntary HIV status disclosure to their partners, while 43.6% preferred assistance from a healthcare worker.

Direct assessment of transitional care preferences and receipt
Asked about their actual transitional care experiences postrelease, most participants (86.1%) reported having linked to a government ART clinic, primarily clinics that did not focus specifically on the healthcare needs of formerly incarcerated people (60.9%). Few participants reported receiving support from a peer navigator (11.9%) or support group (1%), and fewer still received livelihood support (5%) or support for hazardous alcohol use (2%) or other mental health issues (1%). Of 74 participants who had disclosed their HIV status to their partner(s), 45 did so without support from a healthcare worker. There was evidence of significant preference variation across participants, with the standard deviations for 9 of 11 attribute-level coefficients significantly different from zero. For example, while the coefficient for healthcare workers' focus on releasees living with HIV was not statistically significant, its standard deviation indicates significant variation in individual preferences within the sample, with some individuals perceiving healthcare workers' focus on releasees living with HIV strongly negatively, while others viewing it strongly positively. Similar heterogeneity was observed for HIV status disclosure assistance from transitional care providers. The distributions of releasees' preferences are shown graphically in Figure 4. Most releasees were averse to referrals to a CBO-run ART clinic (vs. a government ART clinic), preferred livelihood support or HIV care support provided by a support group (vs. no support) and preferred transitional care providers who were either living with HIV or formerly incarcerated and living with HIV (vs. providers from the general community). Notably, negative correlations among individuallevel preference estimates for peers versus support groups (ρ = -0.07 and ρ = -0.09 for HIV care and livelihood support, respectively, results not shown) suggest differentiated preferences for these alternative transitional care models across participants. There was little evidence of systematic variation in preferences by participant characteristic, with only a hand-    ); SD, standard deviation. a Coefficients describe the "average utility" derived from each transitional care characteristic relative to the respective reference category, confidence intervals describe the precision with which the average is estimated. b Standard deviations describe the variability around the average utility derived from each transitional care characteristic across participants. The corresponding p-values characterize the statistical significance of preference heterogeneity. c Attributes C (transitional care model) and D (type of support provided) were included in the experimental design and analysis as a compound attribute due to their inseparable nature (i.e. transitional care services can only be provided if PLHIV are linked to a transitional care provider). d The utility from both HIV care support and livelihood support was not significantly different from HIV care support only (p = 0.301 for peers; p = 0.062 for support groups), therefore, only the main effects for the two types of support were included in the final model. ful of interaction terms found to be statistically significant (Appendix S1). Figure 5 summarizes the relative importance of transitional care attributes. The type and characteristics of the transitional care provider and the types of support received accounted jointly for nearly two thirds (62%) of the potential utility gains associated with different transitional care configurations. Referral destination, focus of healthcare workers and disclosure assistance, each accounted for 10-15% of potential utility gains. There was large variation in the relative importance of individual attributes across participants. Answers to the debriefing question assessing the difficulty of the DCE indicate that 4 participants (4%) found the DCE very difficult, 28 (27.7%) somewhat difficult and 69 (68.3%) not difficult (results not shown).

D I S C U S S I O N
Findings from this DCE highlight the unmet need for transitional HIV care services in Zambia and opportunities for providing differentiated HIV care, livelihood and other support services to releasees living with HIV. Among the releasees we surveyed, virtually all transitional care characteristics evaluated were preferable to the SOC. A strong preference emerged for referrals to government ART clinics and services that help releasees access livelihood and HIV care support in their communities. The model of transitional care delivery also featured prominently, with support groups and peer nav-igators being preferred over current, non-existent modalities. Experience living with HIV was the most preferred characteristic for transitional care providers. These results notwithstanding, we observed substantial variation in participants' preferences, suggesting a role for differentiating services to meet the individual needs of releasees. While, on average, livelihood assistance was the most preferred type of support, fitting with the needs of a study population in which many participants were unemployed, HIV care support for accompaniment to ART clinic visits and encouragement of medication adherence were also strongly preferred over the SOC. For both livelihood and HIV care support, a slight preference emerged for service delivery through a support group model, which is a common modality for providing psychosocial support to PLHIV in Zambia and SSA [69,70]. Although support delivered by peer navigators was strongly favoured over no support, preferences were less strong relative to support groups. This may reflect the fact that peer navigators are not as commonly encountered in Zambia, or that peer providers are associated with back-tocare services typically reserved for patients who become lost to clinical follow up. Interestingly, participants who had not linked to HIV care in the community had a greater preference (compared to those who had linked) for livelihood support delivered by peer navigators, which may suggest a new role for peer navigators beyond their typical singular focus on HIV care support.
Participants stated a preference for continuing their HIV care in government ART clinics over clinics managed by CBOs catering to formerly incarcerated populations. This observation requires further study through our forthcoming qualitative analyses, and may reflect a number of explanatory factors, including: general community perceptions about the quality of health services in government ART clinics, a lack of ART services offered by most CBOs, a desire for anonymity and avoidance of stigma associated with services for formerly incarcerated people or the convenience and familiarity of collecting ART through government health facilities [41]. Interestingly, while releasees preferred referrals to government ART clinics for continuing HIV care once in the community, a substantial number of participants expressed a preference for receiving care from healthcare workers conversant with the unique needs of releasees living with HIV. Such a result may reflect a preference for client-centred care more generally, or a specific desire for services responsive to the unique needs of releasees, particularly in the face of dual stigma posed by their HIV status and incarceration history [71]. While assistance with partner disclosure did not feature strongly in participant preferences, helping releasees voluntarily disclose their HIV status may help overcome barriers to family and community re-integration. Through the scale up of community index testing, voluntary assisted partner notification services are increasingly becoming a major part of HIV testing services in Zambia [72], and may offer a natural platform for family counselling and partner HIV status disclosure for releasees.
Examining participants' mean preferences and preference heterogeneity simultaneously reveals substantial individual variation in what releasees want from transitional care services, and argues against a "one size fits all" approach to transitional HIV care. That said, several overarching observations emerged that can inform the design of a future transitional HIV care intervention. First, there were clear releasee preferences for receiving HIV care and treatment at government ART clinics, which can serve as the foundation for transitional care programming and help marshal existing government human and clinical resources to enhance the health of releasees. Second, transitional care models involving support groups or peer navigators were generally strongly preferred over the SOC. With peer navigation emerging as a strategy to support HIV care continuity for key populations [73,74], including for incarcerated people and releasees living with HIV [75], a hybrid approach involving options for both support groups and peer navigators may provide a differentiated means of meeting the heterogeneous preferences of this population. Third, the strong preference for livelihood support speaks to the need for transitional care interventions that go beyond HIV care support alone and that more holistically address basic needs like food, housing and employment. Finally, releasees expressed a preference for mental health services, including for mitigating hazardous alcohol and drug use, which are scarce in the current Zambian HIV treatment program.
We acknowledge several limitations. First, the use of hypothetical transitional care characteristics may have introduced measurement error, potentially biasing estimates of releasee preferences. Second, stated preferences may have varied from revealed preferences (i.e. participants' actual decisions during and after release) due to multi-level factors and constraints [76], such as health system barriers to community HIV care, family and partner relationship dynamics, and the demands of meeting basic needs. We attempted to mitigate these biases by concurrently administering a transitional care questionnaire grounded in participants' actual post-release experiences and by having research staff familiar with the study population administer the DCE. Third, given the virtually non-existent literature on the post-release experience of incarcerated PLHIV in SSA, there was limited evidence from the region to guide the selection of attributes, which may have led to the omission of some important attributes. However, the attributes selected align with the major themes identified through our formative IDIs. Fourth, given the considerable resources required for following participants after release, and high rates of recidivism in Zambia [11], we were unable to enrol a larger DCE sample. This may have resulted in selection bias, and limited the generalizability and precision of our estimates. Finally, most of our sample was in HIV care at the time of the DCE, and, as such, our results may not fully capture the preferences of releasees experiencing prolonged care disengagement.

C O N C L U S I O N S
Improving HIV programming for key populations, including formerly incarcerated PLHIV, requires partnering with these populations and rigorously documenting their HIV care needs and preferences. In the first DCE with releasees living with HIV in SSA, we identified preferred characteristics of transitional HIV care that can form the basis for differentiated service delivery models for this population. Such models should aim to provide longitudinal and individualized HIV care support to releasees, offer client-centred treatment in trusted health facilities and strengthen linkages to programs and organizations providing livelihood and mental health support.

A C K N O W L E D G E M E N T S
We thank the study participants, as well as the field staff of the Zambia Correctional Service, the Zambian Ministry of Health and the Releasee Care Continuum study who made this work possible. We acknowledge the support of the senior leadership of the Zambia Correctional Service, including the Health Director, Mr. Yotam Lungu, and the Commissioner General, Dr. Chileshe Chisela. We appreciate the guidance and mentorship provided by Dr. Stewart Reid, Dr. David Wohl and Dr. Carol Golin.

F U N D I N G
Research reported in this publication was supported by the Fogarty International Center of the National Institutes of Health under Award Number K01TW010272 (MEH). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.