Routine programmatic data show a positive population-level impact of HIV self-testing: the case of Côte d’Ivoire and implications for implementation

Objectives: We estimate the effects of ATLAS's HIV self-testing (HIVST) kit distribution on conventional HIV testing, diagnoses, and antiretroviral treatment (ART) initiations in Côte d’Ivoire. Design: Ecological study using routinely collected HIV testing services program data. Methods: We used the ATLAS's programmatic data recorded between the third quarter of 2019 and the first quarter of 2021, in addition to data from the President's Emergency Plan for AIDS Relief dashboard. We performed ecological time series regression using linear mixed models. Results are presented per 1000 HIVST kits distributed through ATLAS. Results: We found a negative but nonsignificant effect of the number of ATLAS’ distributed HIVST kits on conventional testing uptake (−190 conventional tests; 95% confidence interval [CI]: −427 to 37). The relationship between the number of HIVST kits and HIV diagnoses was significant and positive (+8 diagnosis; 95% CI: 0 to 15). No effect was observed on ART initiation (−2 ART initiations; 95% CI: −8 to 5). Conclusions: ATLAS’ HIVST kit distribution had a positive impact on HIV diagnoses. Despite the negative signal on conventional testing, even if only 20% of distributed kits are used, HIVST would increase access to testing. The methodology used in this paper offers a promising way to leverage routinely collected programmatic data to estimate the effects of HIVST kit distribution in real-world programs.


Introduction
In 2019, up to 19% of people with HIV (PWH) worldwide were not aware of their HIV status [1]. In Western Africa, this proportion of undiagnosed PWH reached 33% in 2020 [2]. This is well below the Joint United Nations Programme on HIV/AIDS (UNAIDS) target to achieve <5% of PWH being undiagnosed by 2025. HIV testing is a crucial element of responses to HIV, as it is the first step to linkage to care and treatment. HIV testing is also key for prevention, as PWH on antiretroviral treatment (ART) and virally suppressed will not transmit HIV to their sexual partners [3].
The World Health Organization (WHO) recommends HIV self-testing (HIVST), which allows individuals to test themselves and learn their results when and how they want [4]. It is an innovative tool that has been demonstrated to be safe, accurate, empowering, and acceptable and to also consistently increase the uptake and frequency of HIV testing across settings and populations [5][6][7][8][9][10][11][12]. It is recommended that a reactive HIVST must be followed by a conventional test to confirm or disprove the results.
In Southern and East Africa, HIVST has been scaled up quickly, catalyzed by the Unitaid-funded Self-test Africa (STAR) initiative, which was started in 2015 [13]. However, before 2019, HIVST was offered only in West Africa through small-case pilot projects [14]. A mediumscale HIVST program was implemented in Côte d'Ivoire, Mali, and Senegal in 2018, with an effective distribution of kits through the ATLAS project funded by Unitaid and implemented by a consortium led by Solidarit e th erapeutique et initiatives pour la sant e (Solthis) and the French Research Institute for Sustainable Development (IRD) since 2019 [15]. From 2019 to 2022, together with national programs, ATLAS planned to deliver 400 000 HIVST kits (214 000 in Côte d'Ivoire). The ATLAS program had set a target for 90% of HIVST implementation to reach key populations (KPs) and their sexual partners, peers and clients. In West Africa, the epidemic remains concentrated in KPs, such as female sex workers (FSW), men who have sex with men (MSM), and people who use drugs (PWUD); and is partly drawn by some vulnerable group such as clients of FSW and their non-FSW female partners [16][17][18]. The remaining priority populations of the ATLAS HIVST implementation were people with sexually transmitted infections (STIs), their partners and partners of people living with HIV. ATLAS activities rely on both primary and secondary distribution channels. With primary distribution, HIVST kits are distributed by peer educators and frontline healthcare workers to primary contacts À MSM, PWH, STI patients, FSW, and PWUD À for their personal use. For secondary distribution, primary contacts are invited to redistribute HIVST kits to their peers, partners, clients and relatives. These secondary contacts are often members of key and vulnerable populations who often do not have easy access to the health system, including sexual partners of PWH or members of KPs. This specificity of HIVST kit distribution implies that HIVST beneficiaries (end-users) are not limited to primary contacts. ATLAS's program results have shown that HIVST can reduce stigma; preserve anonymity and confidentiality; reach KPs that do not access testing via other testing approaches; save opportunity costs for users and providers; empower users with autonomy and responsibility; and is noninvasive and easy to use [19][20][21][22].
Several programs have developed methods to assess the use of HIVST and test results, such as supervision by health workers, the return of used kits, messages or phone call reminders to return used samples, the electronic transmission of photographs of test results, or the use of Bluetooth sensors [23]. However, such tracking can be costly and counterproductive by limiting the use and distribution of HIVST and is not in line with the philosophy of HIVST, where users can anonymously decide when and where they are tested and if and to whom they want to report their results. The systematic tracking of HIVST through secondary distribution is logistically challenging and can also hinder the secondary distribution of HIVST, as primary contacts can be reluctant to redistribute an HIVST kit if they need to collect contact information. It could also be challenging for tracking HIVST at a large scale due to the logistics it might involve. To preserve the anonymity and confidentiality of those using HIVSTand not impede the use of HIVST, ATLAS decided not to systematically track distributed HIVST kits. Nevertheless, HIVST users can still, if they wish, obtain additional support by calling a peer educator or a national HIV hotline.
Without systematic and direct feedback regarding the use and results of HIVST and linkage to confirmatory testing and ART, it is challenging to estimate the populationlevel impacts of HIVST distribution [24]. In this paper, we aimed to circumvent this problem by using routinely collected programmatic data to estimate the effects of ATLAS's HIVST distribution on conventional HIV testing (i.e., self-testing excluded), HIV diagnoses, and ART initiations in Côte d'Ivoire.

Methods
Data sources ATLAS HIVST distribution in Côte d'Ivoire started during the third quarter of 2019 (Q3 2019) among individuals aged 16 years or older (minimum legal age for HIV testing without parental consent). All ATLAS implementing partners reported the number of HIVST kits distributed through ATLAS monthly by distribution site, delivery channel, age group and sex of primary contacts. Data were aggregated per health district and quarter of the year. In 2020, Côte d'Ivoire was divided into 33 health regions and 113 health districts.
Routine programmatic data for adults over 15 years of age were obtained from the President's Emergency Plan for AIDS Relief (PEPFAR) open-access public repository (https://data.pepfar.gov/). PEPFAR is the principal donor to the national AIDS program in Côte d'Ivoire. It collects programmatic data in the health districts where it intervenes, including the number of HIVST kits distributed through PEPFAR-funded activities; conventional testing (i.e. the number of 'individuals tested for HIV who received results'); HIV diagnoses (i.e. the number of 'individuals who newly tested positive for HIV'); and ART initiations (i.e. the number of 'people newly enrolled to receive ART').
For this study, we used these two sources of quarterly data aggregated at the health district level À harmonized according to the 2020 subdivision À from Q3 2019 to Q1 2021. Over this period, the PEPFAR data were only available for 78/113 (69%) Ivorian health districts. Only these districts were included in the analysis.
This study does not raise any ethical concern, as the data used are aggregated and completely anonymized. A secondary analysis of the ATLAS programmatic data is included in the associated research protocol approved by the WHO Ethical Research Committee, the National Ethics Committee for Life Sciences and Health of Côte d'Ivoire, the Ethics Committee of the Faculty of Medicine and Pharmacy of the University of Bamako, and the National Ethics Committee for Health Research of Senegal.

Modeling strategy
Our analysis considered three outcomes: the number of conventional HIV tests, HIV diagnoses, and ART initiations. These three outcomes were obtained directly from the PEPFAR datasets.
We used ecological time series regression to model the linear effect of the number of HIVST kits distributed through ATLAS for each outcome [25]. We first used linear mixed models with district-level random effects, as presented in Equation (1): where y i,t is the outcome of district i at time t. HIVST is the number of HIVST kits distributed through ATLAS for district i at time t. b 1 represents the effect of the latter variable on the outcomes. T t is calendar time, which captures conjectural effects in vector b 2 . Conjectural effects were modeled as a categorical variable of the quarter of the year to account for any nonlinear or non-polynomial trend [26]. Modeling time as a categorical variable is also equivalent to running a time fixed-effects model, which allows controlling for variables that are constant across districts but vary over time such as shocks that might occur over the time. d i is the district-specific random effect. District-level random effects were used to account for autocorrelation due to multiple observations and to produce standard errors adjusted for clustering.
Then, contextual effects were also taken into account by introducing the categorical variable of health regions (R i ), and related vector of coefficients b 3 , in the model defined by Equation (2): For each outcome, both Models (1) and (2) were run and results are presented in Supplemental Digital Content, http://links.lww.com/QAD/C586.
ATLAS activities were implemented in nine of the 78 districts covered by the PEPFAR dataset. We performed a sensitivity analysis by restricting the sample to these nine health districts. In addition, we assessed the robustness of our results by using cubic splines instead of a categorical variable for modeling time and compared the AIC (Akaike information criterion) of the models.
All analyses were performed in R (version 4.0.3) using the 'lme4' package for statistical models [27].  (Fig. 1a). In the 69 districts that were not covered by ATLAS (Fig. 1c), HIVST kits distributed through PEPFAR remained limited and largely insufficient to compensate for the reduction in conventional testing; only 13% of the tests in these districts were HIVST kits. In the nine ATLAS districts, HIVST kit distribution À mainly through ATLAS, but also partially through PEPFAR À has increased continuously since Q3 2019 (Fig. 1b), with a slow-down in Q2 2020, when governmental coronavirus disease 2019 (COVID-19) measures were introduced. Overall, the shock caused by the COVID-19 pandemic is observed in Q2 2020.

Descriptive statistics
HIV diagnoses and ART initiations remained relatively stable over time (Figs. 2a and b), with a catch-up effect observed in Q1 2020 after a slowdown in Q4 2019. Trends were similar in the ATLAS districts and the districts not covered by ATLAS.

Regression results
When adjusting for time and region (first columns of When restricting the analyses to the nine ATLAS districts, the estimated magnitudes of association were larger, though not statistically significant (second columns of Table 2). The sensitivity analyses modeling time with cubic splines instead of categorical variables showed very similar results (

Discussion
Using routinely collected programmatic data aggregated quarterly at the health district level, our analyses showed a significant positive effect of HIVST kits distributed through ATLAS on HIV diagnoses. Our results suggested a possible negative signal, albeit not statistically significant, on conventional testing, and no observable effect on the most distal outcome of ART initiations.
HIVST could lead to some substitution effects if HIVST kits are used by individuals who would have undergone a conventional HIV test in their absence, as observed in other studies [5,28]. Such effects may be concerning for policymakers, as gains in HIV testing coverage due to HIVST distribution may result in a reduction in the number of conventional HIV tests. Our results did show a negative 1874 AIDS 2022, Vol 36 No 13  [29], suggesting our 20% utilization rate assumption is conservative. Moreover, the descriptive data showed that a decrease in conventional testing occurred in all districts, including those not covered by ATLAS activities, this is linked to the fact that PEPFAR's testing strategies are revised annually and favoring more targeted approaches [30]. Our results suggest that ATLAS HIVST distribution help maintain access to HIV testing in its implementation districts despite the slowdown observed in Q2 2020 when governmental COVID-19 measures were introduced. In fact, a main takeaway from the ATLAS project is that HIVST distribution activities among KPs can be easily adapted, including in the context of the COVID-19 pandemic [21].
Assessing impact of HIV self-testing Simo Fotso et al. 1875 Due to its confidential nature, HIVST could overcome several structural barriers for HIV diagnoses -such as stigma and opportunity costs -and create new approaches to reach first-time testers and boost HIV retesting for KPs, therefore improving access to HIV testing overall. These results are in line with previous studies among KPs in East and South Africa [5,28,31].
If HIVST is appropriately used as a triage test and individuals with reactive self-tests are linked for confirmatory testing, HIVST distribution activities should lead to a higher number of positive tests in conventional testing. Several actors have expressed concern that HIVST could have a negative impact on new diagnoses [32]. In fact, at the beginning of the ATLAS project, key stakeholders, though recognizing the potential of HIVST to reach first-time testers, expressed some doubts regarding users' ability to accept a reactive test result. There were concerns that individuals with reactive tests would not seek confirmatory testing, which would limit the number of new diagnosed observed at health facilities [19]. Our results did not show any deleterious effect on HIV diagnoses but rather showed a significant positive effect. For the 99 353 HIVST distributed through ATLAS over the period Q3 2019 through Q1 2021, this could translate to 795 additional diagnoses. This is in line with some other studies, such as that by MacGowan et al. [33], who found that the number of HIV infections detected in their HIVSTarm was higher compared to the control arm in a randomized trial conducted among MSM.
Our model did not observe any effect of ATLAS HIVST kit distribution on ART initiations. The estimated effect was negative when all 78 districts were included and positive when the analysis was limited to the nine ATLAS districts. The analysis with nine districts could suffer from  a lack of power. However, in all instances, the effect size estimate was small. The PEPFAR datasets are not exhaustive for Côte d'Ivoire and cover only 78 out of the 113 health districts at the national level and 9 of the 12 ATLAS districts.
Using aggregated data rather than individual data implies a lower number of observation points and therefore lower statistical power, although these data allow us to make population-level estimates. Aggregated data is subject to ecological bias and statistical relationships must be interpreted with caution. In addition, it is not possible to completely rule out any 'contamination' effect, as individuals living on district borders could perform conventional testing in the neighboring district. However, we could assume that population movement at boundaries could happen in both directions, thus compensating for each other, or expect the observed effect to have been even stronger without a 'contamination' effect. The collected data did not allow us to distinguish between confirmatory tests following HIVST and classic conventional tests, but as the HIV prevalence is relatively low in the country, the former number might not be important. Finally, HIVST kits distributed through UNICEF or the Global Fund to fight AIDS, tuberculosis and malaria at district level and by yearly quarter are not available. Nevertheless, the volumes of tests distributed by these programs were very low, 6879 and 1373 kits, respectively, by 2020, representing less than 7% of all HIVST kits distributed in the country.
A strength of this study is that it specifically used only indicators that had already been routinely collected by countries, which means that the method could be easily replicated in other contexts and used by other countries to monitor the impact of their HIVST activities without any additional cost. Our analysis did not rely on any systematic tracking system or data collection process, which can be expensive and complex and are counter to the rationale for HIVST.
A core component of the ATLAS HIVST strategy in West Africa is the secondary distribution of HIVST kits, primarily distributed through activities targeting individuals in KPs, in particular FSW and MSM. It is therefore expected that many HIVSTusers would not self-identify as being in a KP and that those with a reactive test would not link to partner community facilities serving KPs for confirmatory testing but rather to more general public or private facilities, making it difficult to link specific records with the distribution of HIVST kits. In addition, records of prior HIVST kit use at health facilities are expected to be underestimated, as recognizing such use would mean the individual was a member of and/or in a network of a KP. By using data aggregated at the district level and covering all testing facilities, confirmatory tests prompted by reactive HIVST results are considered, regardless of where they occurred. By allowing programs to shift from systematic tracking for evaluation, such indirect evaluation would help to focus on and increase access to HIV testing for hard-toreach populations and first-time testers and allow largescale secondary distribution implementation.

Conclusion
To the best of our knowledge, this represents the first study estimating the impact of HIVST kit distribution at the population level in West Africa. Our results highlight that a social network-based HIVST distribution strategy, focusing on key population members as primary contacts but aiming to reach their partners and social contacts, does have a positive impact on diagnoses that is observable at the population level.
Such evaluation is pragmatic and could be performed with routinely collected programmatic indicators. The WHO recommends reporting on the number of HIVST kits distributed and estimating HIVST access and use through population-based surveys. Countries are burdened with multiple HIV reporting systems and numerous indicators. It could therefore be of considerable benefit to monitor the impact of self-testing through current data systems without introducing new indicators and further data collection. This method of triangulating available data provides further information on the population-level impacts of HIV selftesting to guide program use.