Trends in knowledge of HIV status and efficiency of HIV testing services in sub-Saharan Africa, 2000–20: a modelling study using survey and HIV testing programme data

Summary Background Monitoring knowledge of HIV status among people living with HIV is essential for an effective national HIV response. This study estimates progress and gaps in reaching the UNAIDS 2020 target of 90% knowledge of status, and the efficiency of HIV testing services in sub-Saharan Africa, where two thirds of all people living with HIV reside. Methods For this modelling study, we used data from 183 population-based surveys (including more than 2·7 million participants) and national HIV testing programme reports (315 country-years) from 40 countries in sub-Saharan Africa as inputs into a mathematical model to examine trends in knowledge of status among people living with HIV, median time from HIV infection to diagnosis, HIV testing positivity, and proportion of new diagnoses among all positive tests, adjusting for retesting. We included data from 2000 to 2019, and projected results to 2020. Findings Across sub-Saharan Africa, knowledge of status steadily increased from 5·7% (95% credible interval [CrI] 4·6–7·0) in 2000 to 84% (82–86) in 2020. 12 countries and one region, southern Africa, reached the 90% target. In 2020, knowledge of status was lower among men (79%, 95% CrI 76–81) than women (87%, 85–89) across sub-Saharan Africa. People living with HIV aged 15–24 years were the least likely to know their status (65%, 62–69), but the largest gap in terms of absolute numbers was among men aged 35–49 years, with 701 000 (95% CrI 611 000–788 000) remaining undiagnosed. As knowledge of status increased from 2000 to 2020, the median time to diagnosis decreased from 9·6 years (9·1–10) to 2·6 years (1·8–3·5), HIV testing positivity declined from 9·0% (7·7–10) to 2·8% (2·1–3·9), and the proportion of first-time diagnoses among all positive tests dropped from 89% (77–96) to 42% (30–55). Interpretation On the path towards the next UNAIDS target of 95% diagnostic coverage by 2025, and in a context of declining positivity and yield of first-time diagnoses, disparities in knowledge of status must be addressed. Increasing knowledge of status and treatment coverage among older men could be crucial to reducing HIV incidence among women in sub-Saharan Africa, and by extension, reducing mother-to-child transmission. Funding Steinberg Fund for Interdisciplinary Global Health Research (McGill University); Canadian Institutes of Health Research; Bill & Melinda Gates Foundation; Fonds the recherche du Québec—Santé; UNAIDS; UK Medical Research Council; MRC Centre for Global Infectious Disease Analysis; UK Foreign, Commonwealth & Development Office.


Research in Context
household survey data about the proportion of adults ever tested for HIV and HTS program data on the 51 total annual number of HIV tests performed among adults using a mathematical model of testing 52 behaviors, this study is the first to systematically and comprehensively assess how KOS and HTS 53 efficiency evolved in SSA over 20 years, and with stratification by sex, age, and region. 54

Implications of the available evidence 55
The last two decades witnessed remarkable increases in KOS across SSA, but stark sex, age, and regional 56 disparities remain, even in countries that have met the 90% target overall. Concomitant decreases in 57 median time to diagnosis, HIV testing positivity, and proportion of new diagnoses among all positive 58 tests highlight one of the major challenges faced by testing programs -targeting of HTS to achieve 59 greatest yield of new diagnoses as the undiagnosed population shrinks and diagnosis delays are reduced. 60 With national HIV control programs now contemplating how to reach the next UNAIDS target of 95% 61 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020. 10.20.20216283 doi: medRxiv preprint number of HIV tests conducted and number of HIV diagnoses, are routinely collected, but reports are 93 often not deduplicated and rates of retesting and re-diagnosis can be high. 13,14 Household surveys provide 94 cross-sectional data about testing history by HIV status at intervals roughly every five years in most 95 countries, but only a few surveys directly ask respondents if they are aware of their HIV status, a sensitive 96 question that has high potential for non-disclosure. [15][16][17] These challenges are compounded by imprecise 97 estimates for the number of new infections by age, sex and geographical area, and by incomplete 98 ascertainment of mortality among the previously diagnosed and undiagnosed population. 99 Nearing the UNAIDS' interim 2020 target deadline, we sought to evaluate progress towards the 'first 90' 100 HIV diagnosis target in SSA, describe the impact of HTS programs on knowledge of HIV positive status 101 (KOS) and timeliness of HIV diagnosis over the 2000-2020 period, and identify remaining gaps in who 102 is being reached by HTS. We synthesized data from 40 SSA countries about HIV testing history from 103 population-based surveys, HTS program data, and HIV epidemic indicators using a validated 104 mathematical model specifically designed to estimate KOS. 13 In addition to trends in KOS and diagnosis 105 gaps, we estimated time from HIV infection to diagnosis, probability of getting tested within one year of 106 infection or before reaching a CD4 cell count threshold lower than 350 CD4 cells per µL, positivity 107 (proportion of HIV-positive tests among all tests), diagnosis yield (proportion of new diagnoses among 108 all tests), and proportion of new diagnoses among positive tests. 109

110
Overview 111 We previously developed and validated a compartmental deterministic mathematical model (named 112 Shiny90), to synthesize multiple data sources into a coherent framework to longitudinally estimate KOS. 113 This model has been described in detail elsewhere. 13 Briefly, Shiny90 models the transition of individuals 114 aged ≥15 years between six stages: 1) HIV-susceptible who have never been tested, 2) HIV-susceptible 115 ever tested, 3) PLHIV who have never been tested, 4) PLHIV unaware who have ever been tested, 5) 116 PLHIV aware (not on ART), and 6) PLHIV on ART. Household surveys and HTS program data are used 117 to estimate the rates of HIV testing among adults not living with HIV and those living with HIV, where 118 HIV testing rates vary with calendar time, sex, age, previous HIV testing status, awareness of status, and, 119 for PLHIV, CD4 cell count category (as a marker of risk of AIDS-related symptoms motivating care-120 seeking and HIV testing). 13 In this way, the proportion of PLHIV who know their status estimated by 121 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020. 10.20.20216283 doi: medRxiv preprint Shiny90 is bound by ART coverage (minimum) and the proportion of PLHIV who have ever been tested 122 and received the results (maximum). 123

Data sources and model calibration 124
Shiny90 uses inputs for HIV incidence, mortality, and ART coverage estimated and reported by national 125 governments using the UNAIDS-supported Spectrum modeling software and its Estimation and 126 Projection Package. 12 The Spectrum model calculates epidemic statistics stratified by age, sex, CD4 cell 127 count category, and ART status. Parameter estimates for HIV disease progression and mortality, as well 128 as demographic rates, are also informed by Spectrum. 129 Two main data sources are used for estimation of HIV testing rates during model calibration: 130 1) The proportion of individuals (³15 years old) who self-report having ever been tested for 131 Assessments (PHIA; https://phia-data.icap.columbia.edu/files), and other country-specific surveys 137 ( Figure 1). The model was calibrated to data on the proportion ever tested for HIV, but we did not 138 calibrate to self-reported awareness of status data, due to evidence of non-disclosure. 15 For the analyses reported here, we used the Shiny90 country files submitted to UNAIDS in 2020 144 (www.unaids.org/en/dataanalysis/datatools/spectrum-epp), including Spectrum, surveys, and program 145 data. Additional programme data sources are listed in appendix (pp 47-50). 146 Sub-saharan African countries with at least one available survey stratified by HIV sero-status, or 147 countries with surveys not stratified by HIV sero-status but having at least one HTS program data set 148 including total number of positive tests between 2000-2019 were included in analyses. These were the 149 minimal set of survey and HTS program data that were required to calibrate the model for a given 150 country. Countries with a population under 250,000 people, or without available survey data, or with 151 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.20.20216283 doi: medRxiv preprint only survey data not stratified by HIV sero-status and no HTS program data between 2000-2019 were 152 excluded from analyses. 153 For each country, the model estimates rates of HIV testing by sex, age, HIV status, and testing and 154 treatment history were estimated from the household survey and HTS program data in a Bayesian 155 framework. The mode of the posterior distribution was estimated via optimisation with the Broyden-156 Fletcher-Goldfarb-Shanno algorithm 18 and the posterior density was approximated via Laplace 157 approximation around the posterior mode. 13 Conceptually, the HTS program data inform rates of HIV 158 testing in the population, while changes in the proportion ever tested by HIV status, sex, and age, 159 alongside estimates of HIV incidence and mortality, inform the proportion of tests conducted among 160 those being HIV tested or diagnosed for the first time versus repeat testing. 13  can be stratified by sex and age group, and aggregated to regional level by weighting each country's 171 indicator by the number of estimated PLHIV from Spectrum for that calendar year. 172

Estimating time to diagnosis 173
From the annual sex-, age-, HIV testing history-, and CD4 cell count-specific testing rates, we 174 calculated several cross-sectional indicators using period life table methods 19  to experience that year's HIV testing rates by age and CD4 category for their remaining lifetime. Details 182 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.20.20216283 doi: medRxiv preprint of the calculations are presented as supplementary materials (appendix p 1). For each calendar year 183 between 2000 and 2020, we estimated these indicators for the 16 age/sex/testing history strata separately. 184 They were then aggregated to the desired demographic or geographic level (e.g., age, sex, country, region 185 [Western, Central, Eastern, and Southern Africa]) by weighting each stratum by the estimated number of 186 new HIV infections in that stratum for that year (obtained from Spectrum). 187

Uncertainty 188
We obtained uncertainty intervals by drawing 1,000 samples from the posterior distribution of the testing 189 rates estimated by Shiny90. We summarized all indicators using the median, 2.5 th and 97.5 th percentile 190 of their posterior distribution. We performed analyses using R version 3.5.1 and the Rcpp packages. 20 191 The code for Shiny90 is available on a public repository (www.github.com/mrc-ide/first90release). We

Role of the Funding Source 201
The funders of the study played no role in study design, data collection, data analysis, data interpretation, 202 or writing of the report. The corresponding author had full access to all the data in the study and had final 203 responsibility for the decision to submit for publication. 204

205
A total of 40 countries, 183 population-based surveys (>2.7 millions surveyed individuals), and 315 206 country-years of HTS program data reports informed our model ( Figure 1). Four SSA countries (Cabo 207 Verde, Central African Republic, Guinea-Bissau, Mauritius) were excluded from the analyses due to 208 insufficient data inputs for model calibration, and one (Sao Tome and Principe), because of high 209 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.20.20216283 doi: medRxiv preprint uncertainty in epidemic statistics for small population sizes (<250,000 people). Results of the Shiny90 210 model calibration are presented in Text S2 (appendix pp . 211 Across SSA, the proportion of adults 15 years and older (both those living and not living with HIV) 212 estimated to have been tested increased by 48 percentage points from 2000 to 2020 (Table 1). Testing 213 coverage was highest in Southern Africa with 85% (95%CrI: 83 to 88%) of adults projected to have ever 214 been tested in the region in 2020. 215 The proportion of adult PLHIV with KOS increased steadily from 5.7% (95%CrI: 4.6 to 7.0%) in 2000 216 to 84% (95%CrI: 82 to 86%) in 2020 in SSA (Table 1) Our results also suggest disparities in KOS by sex and age. Across SSA in 2020, men had lower KOS 225 (79%, 95%CrI: 76 to 81%) than women (87%, 95%CrI: 85 to 89%), and 15-24 year-olds were the least 226 likely to know their status (65%, 95%CrI: 62 to 69%; Figure 2B-C, Table S4: appendix p 53). Such 227 disparities were also observed among the 12 countries projected to achieve at least 90% of KOS overall 228 in 2020. Of these countries, only six are projected to achieve 90% of KOS among men, and none are 229 projected to do so among the 15-24 year-olds. 230 While the proportion of PLHIV aware of their status was lower among younger adults, the absolute 231 number of PLHIV was also lower. Consequently, in absolute numbers, the largest group of undiagnosed 232 PLHIV in SSA were men aged 35-49 years, with >700,000 left undiagnosed and 305,000 diagnoses 233 needed to reach 90% awareness of status ( Figure 4,  Figure 5A). That is, if projected HIV testing rates in 2020 persisted into the future, 236 50% of people infected in 2020 would be diagnosed (or, with small probability, suffer AIDS-related 237 mortality) within 2.6 years of seroconverting. National trends are presented in Figure S3 (appendix p 45). 238 Consistent with the estimated decreases in median time to diagnosis, the probability of receiving an HIV 239 test within 1 year following infection or before reaching a CD4 count threshold lower than 350 cells per 240 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.20.20216283 doi: medRxiv preprint µL increased respectively by 31 and 52 percentage points from 2000 to 2020 in SSA (   (Table 1). That is, we 246 project that 58% of PLHIV undergoing testing in 2020 will have been previously diagnosed with HIV. 247 For each of the previous outcomes, five-yearly estimates are presented by sex and age stratification and 248 by region in Tables S4 to S8 (appendix pp 53-57). 249

250
Across SSA, impressive gains were achieved in KOS with 84% (95%CrI: 82 to 86%) of PLHIV being 251 aware of their HIV positive status, and 12 countries and the region of Southern Africa that are projected 252 to reach the 90% KOS target in 2020. Concomitant with these improvements, we estimated that median 253 time from HIV acquisition to diagnosis would be reduced to 2.6 years (95%CrI: 1.8 to 3.5) over that 254 period. 255 Despite this progress, our results highlight substantial regional, national, sex, and age disparities in KOS 256 in SSA. KOS was consistently lower in Western and Central Africa than Eastern and Southern Africa. 257 In those regions, HIV prevalence is lower but key populations -including sex workers, men who have 258 sex with men, and people who inject drugs -account for a higher HIV burden. In all four SSA regions, and consistent with previous studies, men are less likely to know their HIV status 274 compared to women. [24][25][26][27][28] Overall, the diagnosis gap is such that there is a 8% point difference between 275 men and women in 2020. Large differences in KOS are also observed between age groups, with the 276 lowest proportion diagnosed being among PLHIV aged 15-24 years. Importantly, all countries have yet 277 to reach 90% KOS in this younger group. This gap between age groups is the natural consequence of 278 HIV transmission dynamics. HIV incidence is highest and average time since infection is short -and 279 thus cumulative exposure to testing is lower -in this age group compared to older ones. 12 To achieve 280 90% KOS among 15-24 years-old would require a simultaneous increase in testing with greater 281 investment in HIV prevention to increase coverage of high impact prevention interventions. 282 While we found that KOS was proportionally the lowest among men aged 15-24 years old, the largest 283 group of undiagnosed PLHIV was men aged 35-49 years, with >700,000 projected to be undiagnosed in 284 2020. Lower uptake of HIV testing among men may be explained by fewer opportunities for testing as 285 well as other social and system-wide barriers such as harmful gender norms 7,29 and inaccessible or 286 unfriendly services. 30 Engaging men in HIV prevention efforts is critically important, not only for their 287 own needs, but also for their sexual partners. An increase in KOS and of treatment coverage among older 288 men could be critical to reduce HIV acquisition rates among women, and by extension, reducing mother-289 to-child transmission. Among different testing approaches, community-based testing, door-to-door HTS, 290 home-based couples testing, workplace programs, mobile testing services, social network interventions, 291 incentives to test, self-testing, and partner notification have shown success in increasing diagnostic 292 coverage among men. 31 As part of these efforts, facilitating linkage and retaining men with HIV care 293 remains a key challenge for further progress towards HIV testing and treatment targets. 294 Despite improvements, especially in Southern Africa where the median time to diagnosis (or AIDS-295 death) was estimated at 1.5 years (95%CrI: 0.9 to 2.3) in 2020, we projected that across SSA and at 296 current testing levels, 50% of PLHIV will not be diagnosed within 3 years following their infection, and 297 29% will not get tested before reaching a CD4 count threshold lower than 350 cells per µL in 2020. These 298 diagnostic delays impede rapid ART initiation at high CD4 counts which, in its absence, contribute to 299 increased HIV morbidity and onward HIV transmission. 1,[32][33][34] Reducing diagnostic delays on their own 300 will likely not be enough to improve individual and population health outcomes. Earlier diagnosis should 301 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.20.20216283 doi: medRxiv preprint be accompanied by rapid ART linkage and long-term adherence to ART ¾ these are crucial to 302 minimizing morbidity and reducing HIV incidence. 32,33,35 303 As the undiagnosed population shrinks and diagnosis delays are reduced, targeting of HTS to achieve 304 greatest yield of new diagnoses is one of the major challenges faced by testing programs. 36 Although we 305 noted an ecological correlation between a country's testing volume with respect to its population of 306 reproductive age and KOS, we also estimated a decline in positivity and in the yield of new diagnoses. 307 Such declining yields are an inevitable consequence of reaching saturation in testing programs ¾ as long 308 as testing rates are lower in previously-diagnosed individuals than in undiagnosed, we can expect yields 309 to decline as KOS increases. Our analyses also highlight substantial retesting of PLHIV already aware 310 of their status. We projected that 58% of positive tests will be performed on previously diagnosed PLHIV 311 in SSA in 2020. In previous studies conducted in SSA between 2004 and 2018, retesting among PLHIV 312 with known HIV status was also common, ranging from 13% to 68%. 14,37-41 Retesting can be motivated 313 by multiple factors, one of them being the ability to confirm the accuracy of the initial test result. 41-43 314 Another important driver of retesting may be avoiding disclosing prior knowledge of HIV positive status 315 due to societal stigma or denial. A recent study conducted among persons undergoing HIV testing at a 316 health facility in South Africa found that 50% of patients testing HIV-positive had previously been in 317 HIV care (and hence previously diagnosed). Among these, half did not disclose prior knowledge of HIV 318 status to their health care provider. 14  was incorrectly counted as separate HIV diagnoses, which our model would not be able to identify from 332 routinely reported data. Fifth, we also assumed that self-reporting of HIV testing histories was accurate 333 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.20.20216283 doi: medRxiv preprint but social desirability and recall biases could result in underestimation of the proportion ever tested and, 334 ultimately, of KOS. 44 However, validation of self-reported HIV testing histories by mean of antiretroviral 335 biomarkers data from PHIA surveys from eSwatini, Malawi, Tanzania, and Zambia using Bayesian latent 336 class model suggest that, self-reported HIV testing history being highly sensitive, underestimation of the 337 proportion ever tested and of KOS should be low (Xia et al. preprint). 45 Sixth, earlier estimates of 338 diagnosis delays are informed by relatively few population-based survey estimates and HTS program 339 data. Given the cross-sectional nature of these metrics, they could be more sensitive to the elicited 340 model's prior distributions in early years. Finally, the impact of measures taken to prevent the spread of 341 COVID-19 in some countries could have affected both HIV incidence and HTS. 46 Such unacounted 342 factors could potentially lead to slightly lower KOS estimates than those projected in 2020, although a 343 notable decrease would be unlikely since already diagnosed PLHIV would remain so. 344 Although previous studies examined HIV testing uptake or self-reported KOS at community or country 345 level, the present analysis is believed to be the first to systematically and comprehensively assess how 346 HTS efficiency evolved in SSA over 20 years. By using a unified framework to compare HTS metrics, 347 consistency and comparability of results between the different outcomes, countries, and regions is 348 improved. A second strength is the large number of surveys and program data used for triangulation, 349 improving the precision and robustness of our results. Third, in assessing time to diagnosis (or AIDS-350 related death) and other related metrics, we provide valuable information to help programs optimize HTS 351 efficiency. 47 With clear individual and population-health benefits of early treatment initiation, reducing 352 diagnostic delays and improving linkage to care will contribute towards the ultimate goal to end AIDS 353 epidemics by 2030. 354

355
In 2014, the world adopted the goal of achieving 90% HIV diagnosis by 2020. Sub-Saharan Africa, the 356 most affected region, is close to reaching this target and we project that 12 countries and one region, 357 Southern Africa, will have reached that goal among adults in 2020. However, reaching 90% diagnosis 358 coverage remains challenging and our results shed light on stark sex and age gaps in KOS. None of the 359 12 countries projected to reach the 90% target overall are projected to do so in all age and sex groups. 360 National HIV control programs are now contemplating how to reach the next UNAIDS target of 95% 361 diagnostic coverage by 2030 in a context of declining positivity, declining yields of "true" new diagnoses, 362 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 23, 2020.  . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 23, 2020. ;  . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.20.20216283 doi: medRxiv preprint . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.20.20216283 doi: medRxiv preprint . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 23, 2020. ; CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.20.20216283 doi: medRxiv preprint