Incidence of Health Facility Switching and Associations With HIV Viral Rebound Among Persons on Antiretroviral Therapy in Uganda: A Population-based Study

Abstract

Background

A substantial proportion of persons on antiretroviral therapy (ART) considered lost to follow-up have actually transferred their human immunodeficiency virus (HIV) care to other facilities. However, the relationship between facility switching and virologic outcomes, including viral rebound, is poorly understood.

Methods

We used data from 40 communities (2015–2020) in the Rakai Community Cohort Study to estimate incidence of facility switching and viral rebound. Persons aged 15–49 years with serologically confirmed HIV who self-reported ART use and contributed ≥1 follow-up visit were included. Facility switching and virologic outcomes were assessed between 2 consecutive study visits (ie, index and follow-up visits, interval of approximately 18 months). Those who reported different HIV treatment facilities between index and follow-up study visits were classified as having switched facilities. Virologic outcomes included viral rebound among individuals initially suppressed (<200 copies/mL). Multivariable Poisson regression was used to estimate associations between facility switching and viral rebound.

Results

Overall, 2257 persons who self-reported ART use (median age, 35 years; 65% female, 92% initially suppressed) contributed 3335 visit-pairs and 5959 person-years to the analysis. Facility switching was common (4.8 per 100 person-years; 95% confidence interval [CI], 4.2–5.5) and most pronounced in persons aged <30 years and fishing community residents. Among persons suppressed at their index visit (n = 2076), incidence of viral rebound was more than twice as high in persons who switched facilities (adjusted incidence rate ratio = 2.27; 95% CI, 1.16–4.45).

Conclusions

Facility switching was common and associated with viral rebound among persons initially suppressed. Investments in more agile, person-centered models for mobile clients are needed to address system inefficiencies and bottlenecks that can disrupt HIV care continuity.

Universal antiretroviral therapy (ART) provision through a “Treat All” approach has improved human immunodeficiency virus (HIV) outcomes throughout Africa. In the “Treat All” era, ART coverage in settings with generalized HIV epidemics such as Uganda has exceeded 90%, accelerating gains in population-level viral load suppression (VLS) [1–3]. Given the success of mass HIV treatment scale-up efforts, focus has been redirected toward identifying and addressing the unsuppressed fraction of persons with HIV, including those experiencing HIV treatment interruptions or defaulting from care altogether. A growing body of literature has attempted to characterize care attrition in persons with HIV, deemed lost to follow-up from HIV services [4]. This broadly includes individuals who miss appointments or ART refills and are not successfully traced by the health facility.

Nevertheless, some studies have shown that sizeable numbers of these untraced persons have not defaulted from HIV treatment but rather transferred care to other health facilities [5]. In 2 South African studies, more than a third of clients who transferred clinics were misclassified as lost to follow-up [6, 7]. Available evidence suggests that facility switching, both planned and unplanned, is motivated by numerous factors, including stigma/privacy concerns, care quality, travel distances, and migration [8–10]. Whether formal (through clinic-facilitated transfer processes) or informal (through client-initiated self-transfers unregistered by the health system, also called “silent transfers”), facility switching has the potential to disrupt HIV care continuity, with some studies attributing prolonged treatment gaps to clinic transfers [11–13].

Despite the frequency of facility transfers in many contexts and their potential to destabilize care, few studies have examined HIV clinical outcomes in persons who switch facilities, and none have leveraged population-level data. The available literature presents mixed evidence on whether facility switching improves or worsens virologic outcomes. In Malawi, 3-year survival among patients on ART who transferred their care to other facilities was comparable to that of the nontransferring population [14]. Conversely, other studies of South African clinical cohorts have reported suboptimal VLS outcomes following facility transfers [15, 16]. Importantly, these studies reported on facility switches documented by the health system and include time periods that predate “Treat All” scale-up. Contemporary population-level data, which capture both formal and silent facility transfers, can better illuminate switching dynamics across independent facility networks and how these dynamics relate to virologic outcomes in the “Treat All” era.

We nested a study of health facility switching in a prospective HIV surveillance cohort in southern Uganda to examine the longitudinal relationship between facility switching and viral load status among persons who self-reported ART use. Estimating the incidence of facility switching and its relationship to longitudinal virologic outcomes can guide differentiation of HIV services to clients who move between facilities, optimizing their treatment outcomes.

METHODS

Design

Data were derived from the Rakai Community Cohort Study (RCCS), an open, population-based HIV surveillance cohort in 40 communities in and around Rakai District, Uganda. Located in south-central Uganda, the RCCS surveillance area consists of rural (agrarian) and semiurban (trading) inland communities with moderate HIV prevalence (9%–26%) as well as hyperendemic fishing communities (HIV prevalence, 37%–43%) along Lake Victoria [17]. RCCS communities are also characterized by substantial in- and out-migration, including mobility within the study area as well as migration to and from communities outside Rakai [18].

Individuals aged ≥15 years across censused communities complete structured interviews approximately every 18 months for assessment of household characteristics, sexual behaviors, and HIV service utilization. HIV status is ascertained among those aged 15–49 years using a validated algorithm of 3 rapid tests and confirmatory enzyme immunoassay testing [19]. Viral load testing is then performed on stored plasma of persons with serologically confirmed HIV using the Abbott RealTime HIV-1 assay (Abbott Molecular, Inc, Des Plaines, IL).

For this study, inclusion was restricted to persons with HIV who self-reported ART use over ≥2 consecutive visits during the following 3 survey intervals: round 17 (February 2015 to September 2016), round 18 (October 2016 to May 2018), and round 19 (June 2018 to October 2020).

Measures

The unit of analysis was a visit-pair, defined as 2 consecutive study visits during the observation period. Each visit-pair consisted of an index visit and a follow-up visit. Participants could, therefore, contribute up to 2 visit-pairs to the analysis, where an individual’s follow-up visit in the first visit-pair served as the index visit in their second visit-pair.

Longitudinal Virologic Outcomes

Four categorical viral load outcomes were identified within visit-pairs, each measuring the following longitudinal virologic patterns over visit intervals: durable suppression (VLS across visits), new suppression (VLS at follow-up visit only), viral rebound (VLS at index visit only), and persistent viremia (VLS at neither visit). A VLS cut point of <200 HIV RNA copies/mL was used to maximize detection of persons who exhibited lower-level viremia and minimize misclassification of persons who experienced episodic, inconsequential viremic “blips” as unsuppressed [20]. ART use (current, previous, never) is self-reported in the RCCS and is a valid proxy for treatment coverage relative to assay-based drug detection methods (sensitivity, 77%; specificity, 99%) [21].

Health Facility Switching

In Uganda, HIV care is provided at level 2 facilities (ie, parish level) and higher. Clinics or hospitals attended for HIV care, both within and outside the study area, were enumerated at each study visit for persons currently on ART. Participants were considered to have switched facilities if they reported HIV care enrollment at 2 distinct health facilities over the visit-pair. By comparison, individuals were classified as not having switched facilities (“stable attendance”) if their designated health facility remained constant over the visit-pair.

Distance From Households to Health Facilities

Global positioning system (GPS) coordinates (latitude and longitude) were captured for censused households and enumerated health facilities within the RCCS surveillance area using handheld GPS devices. Health facilities outside the study area were identified and geocoded using a combination of sources, including the Uganda Ministry of Health’s National Health Facility Master List [22], a spatial database of public health facilities in sub-Saharan Africa [23], and Google Earth Pro version 9 (Google LLC, Mountain View, CA). The distance (in kilometers) from an individual’s household to their designated HIV care facility was measured using a geodesic distance calculation, which estimates the shortest spherical distance between 2 points along the earth’s surface. Although road network distance calculations measure travel distances more accurately than geodesic distance, numerous households in the study area were not connected to existing road networks and, thus, could not be mapped onto existing transit infrastructure [24]. Geodesic distances do not vary substantially from travel distances in this context and are, thus, appropriate proxies for travel distances from households to facilities [25].

To estimate the relative increase or decrease in distances to health facilities that resulted from facility switching, distances (in kilometers) between households and health facilities estimated at follow-up (d₂) were subtracted from distances between households and facilities at the index visit (d₁). The final travel distance function (d₁ – d₂) corresponded to the change in travel distance attributed to switching, with positive values (d₁ – d₂ > 0) representing switches to facilities closer to households and negative values (d₁ – d₂ < 0) representing switches to facilities further away from households.

Sociodemographics

Other measured independent variables included age, sex, marital status, educational attainment, primary occupation, religion, household wealth (ascertained from a locally validated index [26] of asset ownership), in-migration (immigration into the study area since the last census), community type, and intercommunity mobility (residence in a different community at follow-up). Among females, pregnancy was ascertained at the index and follow-up visits through self-report and urine dipstick screening. Excluding intercommunity mobility and pregnancy, sociodemographic variables were assigned time-invariant values from the participant’s index visit.

Statistical Analyses

Data were managed and analyzed using Stata/IC 15.1 (StataCorp LLC, College Station, TX). Facility switching incidence rates were estimated, overall and stratified by age, sex, and community type, by dividing the total number of switching events by the sum of total person-years contributed within each visit-pair (accrued person-time between index and follow-up visit, represented by visit dates). Events were assumed to occur at the midpoint of visit-pairs.

Next, Poisson regression with generalized estimating equations, independent covariance matrices, and robust standard errors (accounting for nonindependence of visit-pairs contributed by the same individual) estimated associations of facility switching with new VLS among initially viremic persons and viral rebound among initially suppressed persons, respectively, reported as adjusted incidence rate ratios (adjIRR) with 95% confidence intervals (CIs). Visit-pairs that exhibited persistent viremia and durable VLS served as the reference groups for new suppression and viral rebound, respectively. A best-fitting working correlation structure was selected using the quasi-likelihood under the independence model criterion [27] (Supplementary Table 1). Statistical models were first fit for the overall sample, then stratified by sex and community type. Multivariable models adjusted for sociodemographics were associated with facility switching and VLS at the P < .1 level. To examine the potential effect of the coronavirus disease 2019 (COVID-19) pandemic on results, multivariable models were reanalyzed, excluding all follow-up visits after March 2020.

To address selection biases induced by excluding participants with <2 consecutive visits on ART, inverse probability of selection weights were calculated using marginal probabilities of study inclusion, derived from logistic regression (Supplementary Table 2, Supplementary Figure 1) and included in multivariable analysis.

Ethics

The Johns Hopkins University School of Medicine Institutional Review Board and the Uganda Virus Research Institute Research and Ethics Committee approved the study protocol. Adults (aged ≥18 years) provided written informed consent prior to study procedures. Written assent and parental consent were obtained for unemancipated minors aged 15–17 years.

RESULTS

Overall, 33 219 individuals had ≥1 RCCS visit(s) from February 2015 to October 2020, of which 5814 (17.5%) were serologically confirmed to be diagnosed HIV (Figure 1). Of these, 913 (15.7%) persons without a follow-up visit during the observation period were excluded, and the remaining 3080 (62.8%) had ≥2 visits during the observation period. An additional 823 individuals were excluded due to ART nonuse (n = 257) or nonsequential study visits on ART (n = 566), yielding a final analytic sample of 2257 individuals contributing 3335 visit-pairs and 5959 person-years to the analysis.

Table 1 presents descriptive participant characteristics at their first visit during the observation period (n = 2257), stratified by sex. The median age was 35 years (interquartile range [IQR], 30–40), and most participants were female (64.7%) and exhibited VLS (92.0%) at their first visit. Relative to the analytic cohort, individuals lost to follow-up (<2 consecutive visits) were significantly (P < .001) younger (<30 years, 36.0% vs 23.2%) and more likely to recently in-migrate into the study area (31.2% vs 14.6%) and exhibit HIV viremia (14.5% vs 8.0%) at their initial visit (Supplementary Table 2).

Incidence and Predictors of Facility Switching

Overall, 275 facility switching events were identified in 254 participants (11.3%) over the observation period, yielding an incidence rate of 4.8 per 100 person-years (95% CI, 4.2–5.5; Supplementary Table 3). Figure 2 displays age-stratified incidence rates of facility switching by sex and community type. Facility switching rates were comparable among males and females. Across community types, facility switching incidence was highest among individuals aged 25–29 years (inland, 5.1 per 100 person-years; 95% CI, 3.2–8.1 and fishing, 10.4 per 100 person-years; 95% CI, 7.7–14.0); however, facility switching was higher in fishing than inland communities across age strata.

Table 2 summarizes sociodemographic and clinical predictors of facility switching at the visit-pair level (n = 3335). Younger people (<30 years, 29.4% vs 19.1%), households in the poorest wealth quartile (56.0% vs 40.5%), fishing community residents (68.7% vs 44.6%), in-migrants (22.2% vs 9.2%), and individuals who moved to new communities within the study area (7.3% vs 2.0%) were significantly (P < .001) more likely to switch facilities. Facility switching was also significantly more common among individuals who exhibited viremia at their index study visit than in suppressed persons (20.4% vs 5.8%, P < .001). Among females, facility switching was unassociated with pregnancy at the index or follow-up visit (15.3% vs 13.5%, P = .491).

Distance From Households to Health Facilities

At index visits, the median distance from households to health facilities was 24.5 km (IQR, 14.4–33.4; Table 1). Distance from households to health facilities, however, was unassociated with subsequent facility switching (21.3 vs 24.4 km; P = .146; Table 2). The largest fraction of facility switches were within 5 km of participants’ households (37.3%), followed by switches to facilities located ≥5 km from their households (33.2%; Supplementary Figure 2). The median travel distance difference (d₁ – d₂) between origin and destination clinics was 3.8 km (IQR, −11.5  to 9.5), indicating that persons tended to switch to facilities closer to their homes.

Facility Switching, New HIV Suppression, and Viral Rebound

Most visit-pairs exhibited durable VLS (90.5%) during the observation period (Supplementary Table 4); however, durable VLS was substantially less common in visit-pairs where facility switching was observed (75.1% vs 91.9%).

Table 3 presents results from multivariable analysis of facility switching and new suppression and viral rebound, respectively. While new VLS was unassociated with facility switching in the overall sample (adjIRR = 1.17; 95% CI, .76–1.78), individuals in inland communities who exhibited viremia initially were twice as likely to achieve VLS at follow-up if they switched facilities (adjIRR = 2.15; 95% CI, 1.27–3.62).

Conversely, individuals who exhibited VLS at their index visit were twice as likely to exhibit viremia at follow-up if they switched facilities (adjIRR = 2.27; 95% CI, 1.16–4.45). A similar, albeit more pronounced, effect size was observed among inland community residents (adjIRR = 4.35; 95% CI, 1.76–10.73). Findings were similar in weighted analyses (Table 3) and sensitivity analyses restricted to visits through March 2020 (Supplementary Table 5), indicating limited effects of the COVID-19 pandemic restrictions on study findings. Travel distance differences from households to health facilities were also unassociated with virologic outcomes in visit-pairs where facility switching was observed (Supplementary Figure 3).

DISCUSSION

In this population-based study, facility switching was common and significantly associated with HIV viremia, specifically among individuals who exhibited VLS prior to facility switching. We found that facility switching was associated with a 2-fold rate increase in rebounding viremia among initially suppressed persons, reaffirming how facility switching can destabilize HIV care continuity [11–13]. Loss of transfer paperwork, delays in processing transfers, and denial of services to “silently transferring” clients are documented mechanisms through which HIV treatment interruptions can occur during facility switching [8, 12]. Furthermore, the relationship between facility switching and viral rebound was most pronounced in inland communities, where facility switching is less common and, thus, a potentially more disruptive event than in fishing communities. Fishing community residents, whose mobility within and outside the study area can be intermittent and episodic [28, 29], may be more experienced than inland community residents at navigating the HIV care system, particularly during periods of mobility. Addressing systemic bottlenecks to issuing and administering client transfers, including retroactive processing of clients who self-transferred to facilities, is imperative to maintaining care continuity for persons who switch clinics.

Facility switching prevalence (approximately 8%) and incidence (approximately 5 per 100 person-years) in this study population were comparable to estimates from other studies in high-prevalence settings [6, 7, 15, 30]. Consistent with other studies in sub-Saharan Africa [8, 9, 31], mobility was a critical driver of facility switching in the present study. Facility switching was significantly concentrated in highly mobile groups (ie, young people, in-migrants, fishing community residents) who already shoulder a disproportionate burden of HIV viremia in the overall population [32]. Likewise, durable VLS was substantially less common in the context of facility switching (approximately 75%), revealing suboptimal virologic outcomes among persons on ART with facility switching histories. Expanding access to multimonth ART dispensing for viremic persons (who are ineligible for extended prescription durations in most settings) could mitigate care retention challenges for individuals who switched facilities.

Among individuals who switched health facilities, there was substantial heterogeneity in household proximity to health facilities after switching. Although persons switched predominantly to facilities closer to their households, comparable fractions of switching occurred when individuals transferred their care to facilities further from their households, suggesting facility switching may be motivated by factors beyond travel distance. Importantly, distances from households to health facilities were unassociated with facility switching, reaffirming that individuals may prefer receiving HIV services from clinics located further from their homes [25, 33, 34]. The proliferation of differentiated service delivery models (DSDMs) has also expanded treatment access by reducing travel time and clinic visit frequency [35]. Sustained DSDM scale-up could continue rendering travel distances to clinics negligible, minimizing the burden of facility transfers for clients and health systems alike.

This study has some key limitations. First, identified associations of facility switching with viral rebound remain temporally ambiguous. Viral rebound is a plausible consequence of facility switching, but facility switching may also be a response to rebounding viremia. Poor-quality care and mistreatment by providers, for instance, are known predictors of poor virologic outcomes motivating facility switching [36–38]. Future studies should seek to temporally disentangle the relationship between VLS and facility switching. Second, only RCCS participants observed over ≥2 rounds in the cohort were eligible for inclusion; thus, highly mobile persons with greater propensity for facility switching were likely excluded, inducing selection bias, though this was potentially attenuated through inverse probability weighting. Third, facility switching and viral load status were ascertained only at the start and end of visit-pairs, likely underestimating switching events or changes in viremia that occurred during the interval. Likewise, pregnancy ascertainment between approximately 18-month study intervals likely underestimates pregnancies in the study population and their potential relationship with facility switching. Pregnant persons may be diagnosed with HIV during antenatal care and initiate ART at one facility, then switch to another facility postpartum, which has been documented in other settings such as South Africa [6, 39]. Fourth, facility switching was ascertained by self-report and is, thus, susceptible to response and recall biases. Fifth, despite widespread DSDM implementation in Uganda [40], this study did not measure the proportion of persons enrolled in community-based DSDMs who may not necessarily need to visit their designated health facilities for ART refills. Sixth, unlike other clinic-based studies, this study could not discern whether observed facility switching events were formal transfers or self-transfers, which could further clarify the relationship between facility switching and viremia. Last, unmeasured clinical factors likely associated with virologic outcomes (eg, ART use duration, ART regimen, HIV drug resistance) were unavailable.

CONCLUSIONS

In this population-based study of persons on ART in southern Uganda, the frequency of facility switching over a 5-year period was high and significantly associated with loss of VLS. Although facility switching may benefit persons on ART when desired by clients themselves, it can place strain on underresourced health systems with limited capacity to surveil client movements between clinics. Investments in more agile, person-centered models for mobile and complex clients, including retroactive transfer processing and expedited reenrollments for self-transfers, could address system inefficiencies and bottlenecks that disrupt HIV care continuity.

Supplementary Data

Supplementary materials are available at Clinical Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.

Notes

Acknowledgments. The authors thank all Rakai Community Cohort Study (RCCS) participants for their time and contributions to the present study. They also acknowledge the dedication and efforts of all RCCS data collection, laboratory, service linkage, and data management staff.

Disclaimer. The contents presented here are solely the responsibility of the authors and do not necessarily reflect the views or policies of the funding agencies, nor do mentions of trade names, commercial products, or organizations imply endorsement by the US government.

Financial support. This work was supported by the National Institute of Allergy and Infectious Diseases (U01AI100031, U01AI075115, R01AI110324, R01AI128779, R01AI123002, R01AI143333, R01AI155080, T32AI102623), the National Institute of Mental Health (R01MH105313), the National Cancer Institute (75N91019D00024), and the US President’s Emergency Plan for AIDS Relief through the Centers for Disease Control and Prevention under the terms of NU2GGH00081. J. G. R. and K. B. R. were supported by the National Institute of Mental Health (F31MH126796, K01MH129226).

References

Author notes