The differential impacts of PEPFAR transition on private for-profit, private not-for-profit and publicly owned health facilities in Uganda

While transition of donor programs to national control is increasingly common, there is a lack of evidence about the consequences of transition for private health care providers. In 2015, President’s Emergency Plan for AIDS Relief (PEPFAR) identified 734 facilities in Uganda for transition from PEPFAR support, including 137 private not-for-profits (PNFP) and 140 private for-profits (PFPs). We sought to understand the differential impacts of transition on facilities with differing ownership statuses. We used a survey conducted in mid-2017 among 145 public, 29 PNFP and 32 PFP facilities reporting transition from PEPFAR. The survey collected information on current and prior PEPFAR support, service provision, laboratory services and staff time allocation. We used both bivariate and logistic regression to analyse the association between ownership and survey responses. All analyses adjust for survey design. Public facilities were more likely to report increased disruption of sputum microscopy tests following transition than PFPs [odds ratio (OR) = 5.85, 1.79-19.23, P = 0.005]. Compared with public facilities, PNFPs were more likely to report declining frequency of supervision for human immunodeficiency virus (HIV) since transition (OR = 2.27, 1.136-4.518, P = 0.022). Workers in PFP facilities were more likely to report reduced time spent on HIV care since transition (OR = 6.241, 2.709-14.38, P < 0.001), and PFP facilities were also more likely to discontinue HIV outreach following transition (OR = 3.029, 1.325-6.925; P = 0.011). PNFP facilities' loss of supervision may require that public sector supervision be extended to them. Reduced HIV clinical care in PFPs, primarily HIV testing and counselling, increases burdens on public facilities. Prior PFP clients who preferred the confidentiality and service of private facilities may opt to forgo HIV testing altogether. Donors and governments should consider the roles and responses of PNFPs and PFPs when transitioning donor-funded health programs. This study is nested within a mixed methods evaluation of the PEPFAR GP in Kenya (Rodríguez et al., 2018) and Uganda (Wilhelm et al., 2018). The parent study has three components: documentation of PEPFAR GP process; quantitative analysis of trends in HIV and maternal, neonatal and child health (MNCH) services, staffing and health systems; and in-depth qualitative research through case studies. Prior findings from the parent study identified impacts of transition on human resources (HIV supervision, training, worker time allocation for HIV and non-HIV care), service delivery (HIV outreach, facility in-charges perceptions of quality of care) and laboratory networks [disruption of viral load (VL) and sputum tuberculosis (TB) testing]. In Uganda, we conducted a cross-sectional facility survey in July and August of 2017, roughly 9 months after the median transition date and 4 months after the official end of the GP process reported by United States Agency for International Development (USAID). For logistical reasons, we limited the sampling area for this survey to 42 districts (out of the 112 districts that existed in 2014): 40 districts in Northern and Eastern Uganda, as well as two urban districts, Kampala and Wakiso, in Central Uganda. Kampala and Wakiso contain more than half of the PFPs transitioning from PEPFAR. We constructed the sample frame from a list supplied by USAID of PEPFAR-supported facilities. Only facilities supported by USAID-contracted Implementing Partners (IPs) were included in the survey. We also excluded all facilities identified for scale-up (i.e. increased support). From the sample frame, we selected districts using a stratified random sampling design with three strata: (1) 100% selection of all districts containing transitioning HC IVs and/or Hospitals as well as Kampala and Wakiso, (2) random sampling of 11 out of 18 remaining districts that were designated for transition or maintenance and (3) random sampling of 6 out of 14 Scale-Up districts in our sampling region, which also contain some facilities designated for transition on the basis of having ‘low volume’ of HIV services. We sampled all facilities within selected districts that were identified as maintenance (constant support) or transition (withdrawal of support), except for Kampala/Wakiso, where we selected a 40% sample of transition facilities due to the large number of PFP facilities in these districts. This sampling methodology was guided by the goal of the parent study to achieve a 2:1 ratio of transition to maintenance facilities from across Northern, Eastern and Central Uganda. The 2:1 ratio was selected to provide enough power for comparisons between transition facilities (e.g. by ownership) as well as between transition and maintenance facilities. Using this process, we selected 275 facilities. We assumed a 9% non-response rate to achieve a final sample of ∼250. Two facilities that were selected for longitudinal case studies by the parent study but not randomly selected for the facility survey—one PNFP and one PFP—were purposively added to the survey sample and weighted accordingly. Enumerators were able to complete surveys at 262 facilities. Of the 15 facilities that could not be surveyed, 9 had closed permanently, 2 were closed for construction, 2 facilities were identified as duplicate records, 1 (a PFP facility) refused to participate in the survey and 1 was not accessible due to road conditions. Of surveyed facilities, 206 reported having been transitioned, 20 reported continuing to receive PEPFAR support and 36 claimed to have had no PEPFAR support within the past 3 years. This was contrary to what was expected, due both to the 36 sites claiming to have no recent PEPFAR support and the larger than expected proportion of sites reporting transition. From follow-up interviews with IPs and USAID, we determined that as many as 60 of the transitioned facilities were experiencing a break in support between IPs lasting for about 12 months. As the objective of this study was to study facilities’ responses to loss of support, we decided to use facility-reported transition status, and included the roughly 60 facilities that were designated for maintenance and may have been in an extended break in funding as transition. In addition, these facilities reported similar processes to those intended for transition. To examine differences between transitioned PFPs, PNFPs and public facilities, we restrict the analysis in this paper to the 206 facilities reporting transition. In smaller facilities, survey interviews were conducted with facility in-charges or their representatives. In larger facilities, multiple respondents (e.g. facility in-charge, head of the HIV clinic, head of the maternity ward and a financial officer) contributed to different components of the survey. Enumerators asked about conditions before and after the facility’s transition date in the areas of service delivery, laboratory, commodities, human resources and finances. In addition, enumerators sought 1–3 staff that provide HIV care (including potentially the primary respondent) present on the day of the survey to answer a short individual questionnaire on changes in worker time allocation and receipt of incentives (bonuses/salary top-ups, outreach allowances or other support). These individual interviews were conducted in private to improve confidentiality. A total of 429 health worker interviews were collected from transition facilities (304 in public, 71 in PNFPs and 54 in PFPs). Given the large number of potential differences by ownership type and the increased risk of making type I errors with multiple comparisons, we pre-specified 17 hypotheses (Supplementary Table S1). Rather than using the Bonferroni correction approach (i.e. dividing alpha = 0.05 by the number of hypothesis tests), which is conservative when outcomes are correlated, we use the Benjamini–Hochberg (BH) method (Benjamini and Hochberg, 1995) to control the false detection rate to 5%. Compared with public facilities, we hypothesized that transitioned PNFPs would be more likely to lose supervision, access to training and access to lab networks. Public facilities are more closely linked to national support structures, including referral labs, training programs and DHTs, whereas PNFPs may have been more reliant on PEPFAR IPs to provide or facilitate access to these resources. However, we expected that PNFPs, which have a strong mandate to provide care to all in need, and greater flexibility than public facilities to manage staff and deploy funding, would continue to offer HIV services at a level comparable to public facilities. We also expected PFPs to lose more support than transitioned public facilities, even though many PFPs reported never having received supervision. We hypothesized that PFPs are more profit-motivated and have less of a social mission than PNFPs, thus as they lose subsidized inputs for HIV services this may diminish their profits leading them to respond to loss of PEPFAR support by disengaging from HIV services (particularly HIV testing, which accounts for the bulk of their HIV services), following transition. We aimed to isolate differences in outcomes that are due to ownership by controlling for other factors. Public, PFP and PNFP facilities lost differing amounts of support in transition. We constructed a ‘transition impact index’ that sought to quantify the amount of support lost by an individual facility as a data reduction tool to limit the number of variables required to control for differences in transition strength. In constructing the index, we used four ordinal variables: (1) the number of types of IP support for HIV reported lost (for training, supervision, outreach, ART, laboratory), (2) the number of HIV services (HIV testing and counselling, PMTCT, ART, outreach) for which the in-charge identified the PEPFAR IP as primary source of support prior to transition but not after transition, (3) the change in frequency of HIV supervision since transition (−1 decreased, 0 same and 1 increased), with facilities reporting no previous support imputed to ‘same’ and (4) The types of non-salary incentives provided by the IP to at least one worker in the facility prior to transition (0 = none provided, 1 = bonuses or outreach allowances provided and 2 = both provided), almost all of which was lost during transition. We attempted to include loss of salaries paid by the IP, both as a proportion of workers and as a binary variable for any salaries lost; however, salaries had a high degree of uniqueness and added little to the index. Principle component analysis with a polychoric correlation matrix resulted in a single factor model that explained 50.7% of variance. We used exploratory factor analysis to determine factor loadings and create an index score for each facility (Supplementary Table S2). The index can be used to identify transition status with a high degree of accuracy (AUC = 0.981). The diversity of scores for transition facilities is large, suggesting a considerable variation in amounts of support lost (Figure 1). Among transition facilities, the impact index was independently associated with several outcomes, including discontinuation of outreach and workers reporting less time on HIV care. However, the index is only a rough measure of the loss of support during transition. Since the index contains information on supervision frequency, we omit it from the analysis of supervision frequency. Histogram of transition impact index scores. In addition to the transition index, we include other covariates, including facility level (health centre—HC II or clinic, HC III, HC IV or hospital), number of HIV workers prior to transition, number of months since transition and an index of transition preparedness. Ten districts were selected by PEPFAR for transition of all facilities during GP, regardless of their volume (‘Central Support Districts’). These districts commonly had low PEPFAR presence and are mostly located in the sparsely populated Karamoja region. New districts, those created within 10 years of the survey, tend to have less capacity than more established districts. We adjust for district status (new vs established) in the analysis. We created the preparedness index by taking an unweighted average of 14 questions about the facility’s preparedness for transition in domains of communication (to facility, to patients, between facilities), consistency (of HIV and MNCH services, reporting systems and outreach to key populations) before and after transition, and capacity (of facility, management, staff) for transition, each rated on a 5-item Likert scale, excluding don’t know/not applicable. Higher scores indicate a higher level of self-rated preparedness. We compare responses across ownership categories and perform pairwise weighted chi-square tests of the significance of differences in proportions. We also use logistic regression to compare outcomes across facility ownership types, adjusting for covariates. The two methods are complementary. The unadjusted proportions provide perspective on changes taking place among transition facilities as a whole, while logistic regression better isolates the effect of ownership. In both methods, we accounted for survey design using stratification, clustering, finite population correction and sampling weights. All analyses were performed using Stata 15 (StataCorp, 2017).