Stuart RM et al. Journal of the International AIDS Society 2018, 21:e25068 http://onlinelibrary.wiley.com/doi/10.1002/jia2.25068/full | https://doi.org/10.1002/jia2.25068 RESEARCH ARTICLE The City of Johannesburg can end AIDS by 2030: modelling the impact of achieving the Fast-Track targets and what it will take to get there Robyn M Stuart1,2 , Nicole Fraser-Hurt3, Cliff C Kerr2,4, Emily Mabusela5, Vusi Madi5, Fredrika Mkhwanazi6, Yogan Pillay7, Peter Barron8, Batanayi Muzah7, Thulani Matsebula3, Marelize Gorgens3 and David P Wilson2,9 Corresponding author: Robyn M. Stuart, Universitetsparken 5, København Ø 2300, Denmark. Tel: +45 2878 5222. (robyn@math.ku.dk) Abstract Introduction: In 2014, city leaders from around the world endorsed the Paris Declaration on Fast-Track Cities, pledging to achieve the 2020 and 2030 HIV targets championed by UNAIDS. The City of Johannesburg – one of South Africa’s metropoli- tan municipalities and also a health district – has over 600,000 people living with HIV (PLHIV), more than any other city worldwide. We estimate what it would take in terms of programmatic targets and costs for the City of Johannesburg to meet the Fast-Track targets, and demonstrate the impact that this would have. Methods: We applied the Optima HIV epidemic and resource allocation model to demographic, epidemiological and beha- vioural data on 26 sub-populations in Johannesburg. We used data on programme costs and coverage to produce baseline pro- jections. We calculated how many people must be diagnosed, put onto treatment and maintained with viral suppression to achieve the 2020 and 2030 targets. We also estimated how treatment needs – and therefore fiscal commitments – could be reduced if the treatment targets are combined with primary HIV prevention interventions (voluntary medical male circumci- sion (VMMC), an expanded condom programme, and comprehensive packages for female sex workers (FSW) and young females). Results: If current programmatic coverage were maintained, Johannesburg could expect 303,000 new infections and 96,000 AIDS-related deaths between 2017 and 2030 and 769,000 PLHIV by 2030. Achieving the Fast-Track targets would require an additional 135,000 diagnoses and 232,000 people on treatment by 2020 (an increase in around 80% over 2016 treatment numbers), but would avert 176,000 infections and 56,500 deaths by 2030. Assuming stable ART unit costs, this would require ZAR 29 billion (USD 2.15 billion) in cumulative treatment investments over the 14 years to 2030. Plausible scale-ups of other proven interventions (VMMC, condom distribution and FSW strategies) could yield additional reductions in new infections (be- tween 4 and 15%), and in overall treatment investment needs. Scaling up VMMC in line with national targets is found to be cost-effective in the medium term. Conclusions: The scale-up in testing and treatment programmes over this decade has been rapid, but these efforts must be doubled to reach 2020 targets. Strategic investments in proven interventions will help Johannesburg achieve the treatment targets and be on track to end AIDS by 2030. Keywords: Fast-Track targets; ending AIDS; Johannesburg; HIV modelling; allocative efficiency Additional Supporting Information may be found online in the Supporting information tab for this article. Received 26 April 2017; Accepted 18 December 2017 Copyright © 2018 The Authors. Journal of the International AIDS Society published by John Wiley & sons Ltd on behalf of the International AIDS Society. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. 1 | INTRODUCTION their HIV status, 90% of all people with diagnosed HIV infec- tion will receive antiretroviral therapy (ART), and 90% of all In 2014, city leaders from around the world endorsed the people receiving ART will be virally suppressed (90-90-90 Paris Declaration on Fast-Track Cities, pledging to achieve the targets), with these percentages increasing to 95% by 2030 2020 and 2030 HIV targets championed by UNAIDS [1]. The (95-95-95 targets). The 90-90-90 and 95-95-95 targets are Fast-Track targets, now ubiquitous in the HIV field, state that associated with epidemiological milestones: approximately by 2020, 90% of all people living with HIV (PLHIV) will know 80% to 90% reductions in new infections and AIDS-related 1 Stuart RM et al. Journal of the International AIDS Society 2018, 21:e25068 http://onlinelibrary.wiley.com/doi/10.1002/jia2.25068/full | https://doi.org/10.1002/jia2.25068 deaths by 2030, often considered synonymous with the goal how well the network of clinics is working together as a of ending AIDS [2]. whole. Mathematical models have proven useful in answer- The City of Johannesburg became a signatory to the Paris ing questions of this type, and a multitude of different declaration in March 2016 along with 19 other municipalities modelling frameworks have been developed in response, in South Africa, joining other cities around the world [3]. The both specifically for South Africa [16] and more generally crucial role that cities will play in achieving the Fast-Track tar- [17]. gets [4] is particularly relevant in South Africa; both Johannes- To understand what it will take for the City of Johannes- burg and Durban metropolitan municipalities are estimated to burg to achieve the Fast-Track targets, we adapted the have more than 500,000 PLHIV, which would qualify them for Optima HIV epidemic and resource allocation model in order positions in the top 25 countries in the world according to to capture the key aspects of the HIV care and treatment HIV burden if they were counted alongside nations (2013 cascade [18]. Although the Optima HIV model has been suc- UNAIDS estimates [5]; Figure 1). In this study, our focus is on cessfully applied in many countries to assess the impact and the City of Johannesburg. optimize the allocation of HIV programme spending [19], it Notwithstanding the considerable political will that Johan- has yet to be used for a detailed analysis of the care and nesburg’s leaders demonstrated in signing the Paris Declara- treatment cascade, largely due to a lack of comprehensive tion, it remains uncertain how to translate the Fast-Track data. For this analysis, we sought data from a number of targets into programmatic or financial stratagems. The transla- sources, including clinic-level data, cohort studies, national tion step calls for an analysis of the HIV care and treatment reports, and a novel record-linkage analysis providing compre- cascade. Over the past 5 years, the HIV care and treatment hensive viral load and CD4 data for Johannesburg. Using cascade has been established as a useful framework for these data in our specially tailored model, we estimate what it assessing the gaps in accessing the full range of diagnostic, would take in terms of programmatic targets and investments care and treatment services available for PLHIV [6-8]. Under- for the City of Johannesburg to meet the Fast-Track targets, standing and minimizing blockages and leakages along the cas- and then estimate the epidemiological impact that this would cade is the only way that the Fast-Track targets can be have. attained [9,10]. The HIV response within South Africa as a whole is heav- 2 | METHODS ily focused on the achievement of the Fast-Track targets, with many service delivery modalities in place to improve 2.1 | Model structure diagnosis, linkage to and retention in care, and treatment initiation, monitoring and adherence. Despite the recogni- We adapted the deterministic compartmental epidemic model tion of the importance of cities in achieving the Fast-Track structure of Optima HIV [18], making several modifications in targets, most studies tend to focus on analysing the care order to better capture aspects of the care and treatment and treatment cascade within countries [7,9,11-13] or cascade (Figure 2a). Each population group included in the within clinics[14,15]. Aside from the strategic importance model has 37 possible health states: 7 treatment-related of cities, an analysis of a city’s cascade can shed light on states (susceptible, undiagnosed, diagnosed, in care, receiving Figure 1. Estimated number of PLHIV in 2013 (UNAIDS [5]). Johannesburg and Durban have been disaggregated from the rest of South Africa and are highlighted in pink. Together, the two cities make up 3% of the estimated global burden of HIV, with Johannesburg alone accounting for around 9% of South Africa’s HIV burden. 2 Stuart RM et al. Journal of the International AIDS Society 2018, 21:e25068 http://onlinelibrary.wiley.com/doi/10.1002/jia2.25068/full | https://doi.org/10.1002/jia2.25068 ART and not virally suppressed, receiving ART and virally 2.2 | Model transitions suppressed, lost-to-follow-up), with all infected stages further disaggregated into 6 CD4-related states (acute HIV infection, The movement of people along the cascade is modelled by >500 cells/lL, 350 to 500 cells/lL, 200 to 350 cells/lL, 50 the black arrows depicted in Figure 2a. Each arrow is associ- to 200 cells/lL, <50 cells/lL). In addition, we disaggregated ated with a probability of transitioning to a different stage of the total population of the City of Johannesburg (estimated the care continuum within a fixed interval of time, and each is at 4.9 million in 2016) into 26 sub-populations: males and specific to a particular sub-population and CD4-related state. females aged 15 to 60 stratified into 5-year age bands, plus Following the surveillance study undertaken by Howard infants aged 0 to 2, children aged 3 to 14, males and females University in conjunction with the President’s Emergency Plan aged 60+, female sex workers (FSW), clients of female sex For AIDS Relief (PEPFAR) and the National Department of workers, men who have sex with men, and people who inject Health, we consider someone to be linked to care if they have drugs. been enrolled in a facility for continuum of HIV care and have (a) (b) Figure 2. (a) The compartmental structure of the Optima HIV epidemic model, adapted to account for the care and treatment cascade; (b) Progress towards the Fast-Track targets over 2005–2016 (first 3 columns), and model estimates of the scale-up required to 2020 and 2030 (final 2 columns). 3 Stuart RM et al. Journal of the International AIDS Society 2018, 21:e25068 http://onlinelibrary.wiley.com/doi/10.1002/jia2.25068/full | https://doi.org/10.1002/jia2.25068 completed and received the results of an initial ART eligibil- 2.5 | Model analysis ity assessment within 3 months of diagnosis [20]. The dif- ferent transition probabilities associated with different CD4 We estimated the HIV testing and treatment scale-up counts are primarily intended to distinguish the behaviour required to meet the Fast-Track targets under a range of of people with <200 CD4 cells/lL (AIDS-stage infection) assumptions about coverage levels of other programmes. As and also capture differences in mortality rates. We assume the baseline scenario, we assume that coverage of all other that people with AIDS-stage infection have a symptomatic programmes would remain at latest reported levels. The test- testing rate above the standard testing rate for their popu- ing and treatment scale-up required under this baseline sce- lation cohort, which gives us an estimated median CD4 nario is shown in Figure 2b. We then constructed alternative count at diagnosis of 340 cells/lL in 2015, approximately scenarios to describe plausible programmatic scale-ups, on par with other studies [21,22]. We assume that treat- informed by South Africa’s National Strategic Plan (NSP) for ment spots are first taken by those with lower CD4 counts, 2017 to 2022 [24] and the South African National Sex which give us a mean CD4 count at time of treatment initi- Worker HIV Plan for 2016 to 2019 [25]. ation of 190 cells/ll, again comparable to other studies In modelling the impact of the 2017 to 2022 NSP, we [21]. In addition to the transitions along the cascade, we restrict our attention to Goal 1 (of a total 8 goals), which aims also take into account transitions between different sub- to reduce new HIV infections through combination prevention populations (for example, mixing between adult female and interventions, and more specifically to sub-objective 1.1.2, the FSW populations) and disease states (as stratified by CD4 component of Goal 1 that outlines an approach for the provi- counts). sion of targeted biomedical HIV prevention services. (Sub- objectives 1.1.1 and 1.1.3 relate to education programming, which is outside of the scope of our modelling framework; 2.3 | Data sub-objective 1.1.4 relates to the provision of PrEP, which we The data used to inform the model’s transition probabilities model separately as part of the National Sex Worker Plan; are summarized in Table 1. Details on the model equations and sub-objective 1.1.5 relates to PMTCT, which we include are provided in the Supplementary Material, and full details of in the treatment scale-ups associated with the Fast-Track tar- the model parameters are provided in the Optima HIV param- gets.) We omit Goal 2 because it relates to the attainment of eters sources compendium [23]. Estimating the current state the Fast-Track targets (which we already model separately). of the care and treatment cascade relied heavily on the We omit Goal 3, which aims to reach all key and vulnerable Howard University surveillance study [20]. populations with customized and targeted interventions, because concrete targets are yet to be formulated for many of the indicators in the NSP’s Monitoring and Evaluation 2.4 | Model calibration Framework. Finally, we omit Goals 4–8 because these relate We initialized the model in 2000 and produced projections to addressing the social and structural drivers of the epi- from 2000 to 2030. We fitted the model to data on popula- demics, grounding the responses in a human rights approach, tion sizes, HIV prevalence and the number of people on ART promoting leadership, mobilizing resources and strengthening by adjusting a subset of the model’s parameters in order to strategic information, and while these are essential compo- minimize the mean absolute percentage error between the nents of the strategy, our modelling framework is best suited model’s estimates and the data, and then subjecting the pro- to analysing the impact of biomedical and behavioural inter- jections produced by the model to scrutiny and validation by ventions designed to directly impact on one of the proximal the district, province and national health departments determinants of HIV transmission or mortality. involved in the study. For the purposes of fitting, we speci- For each scenario, we calculated the coverage and esti- fied which of the model’s parameters could be adjusted, as it mated investment levels required to attain the Fast-Track tar- is neither feasible nor desirable to allow all parameters to gets, as well as the impact. vary freely. Since our focus is on modelling the care and treatment cascade, we selected parameters to match the 3 | RESULTS cascade transitions depicted in Figure 1. Specifically, we fit- ted: [1] the initial HIV prevalence in each sub-population; [2] 3.1 | Progress towards Fast-Track targets parameters controlling the probability of infection; [3] test- ing rates (within 10% of reported values); [4] the percentage We estimate that in 2016 there were 616,000 PLHIV in of people linked to care within 3 months (within 10% of Johannesburg, with 73% aware of their status, 77% of those reported values); [5] the average time taken from treatment diagnosed linked to care (with 57% of newly diagnosed people initiation to viral suppression, within the ranges given in linked to care within 3 months), 80% of those who were ini- Table 1; [6] the proportion of people virally suppressed at tially linked to care retained in care (84% of whom are receiv- their last test (within 10% of reported values); [7] the pro- ing treatment) and 54% of those receiving treatment with portion of people not returning to their clinic after 90 days viral suppression. Using the UNAIDS cascade with PLHIV as and [8] all CD4 progression and recovery rates, within the denominator (which translates the Fast-Track targets to 90%- ranges given in Table 1. The key outputs of the calibration 81%-72%), the achievement of the three 90s in 2016 was are included in the Supplementary Material. The model’s 73%-48%-26%. The care and treatment cascade improved sig- estimates of prevalence, infections, deaths, number of peo- nificantly over the past 11 years across all stages for which ple on ART and PLHIV are generally aligned to available we have data (Figure 2b). The proportion of people aware of data. their status was estimated to increase from 50% to 73%, 4 Stuart RM et al. Journal of the International AIDS Society 2018, 21:e25068 http://onlinelibrary.wiley.com/doi/10.1002/jia2.25068/full | https://doi.org/10.1002/jia2.25068 Table 1. Key parameters used to inform the transitions for the epidemiological model. Transitions Data types Data availability Value Cascade Infection Sexual behavioural data (number of acts per year & National Time-varying & population-specific probability of condom use with regular, casual and commercial partners) Injecting behavioural data (number of injections National Time-varying & population-specific per year & probability of syringe sharing) Intervention uptake (% of people accessing Municipal Time-varying & population-specific PrEP, circumcision, ART, OST & PMTCT) Per-act transmission probabilities Literature See Supplementary Materials Efficacy of interventions Literature See Supplementary Materials Partnership formation patterns National Time-varying & population-specific Diagnosis % of population tested for HIV in the last 12 months National Time-varying & population-specific Linkage to care % of people linked to care within 3 months of diagnosis Municipal Time-varying & population-specific Treatment initiation Matched to available data on the number Municipal Time-varying & population-specific of people on ART Viral suppression Average time taken from treatment initiation to viral Municipal Time-varying suppression; frequency of viral load monitoring Treatment failure % of those who were virally suppressed Municipal Time-varying at their last VL test Loss to follow-up % of people not returning to their clinic after 90 days Clinic Time-varying CD4 change CD4 progression Duration of acute infection [23] 0.24 [0.10, 0.30] years Time to move from CD4 > 500 to 350 < CD4 < 500 [23] 0.95 [0.62, 1.16] years Time to move from 350 < CD4 < 500 to [23] 3.00 [2.83, 3.16] years 200 < CD4 < 350 Time to move from 200 < CD4 < 350 to 50 < CD4 < 200 [23] 3.74 [3.48, 4.00] years Time to move from 50 < CD4 < 200 to CD4 < 50 [23] 1.50 [1.13, 2.25] years CD4 recovery on Time to move from 350 < CD4 < 500 to CD4 > 500 [23] 2.20 [1.07, 7.28] years suppressive ART Time to move from 200 < CD4 < 350 to [23] 1.42 [0.90, 3.42] years 350 < CD4 < 500 Time to move from 50 < CD4 < 200 to 200 < CD4 < 350 [23] 2.14 [1.39, 3.58] years Time to move from CD4 < 50 to 50 < CD4 < 200 [23] 0.66 [0.51, 0.94] years Time from treatment initiation to achieve viral suppression [23] 0.20 [0.10, 0.30] years CD4 progression & % moving from CD4 > 500 to 350 < CD4 < 500 per year [23] 2.60 [0.50, 27.50]% recovery on % moving from 350 < CD4 < 500 to CD4 > 500 per year [23] 15.00 [3.80, 88.50]% non-suppressive ART % moving from 350 < CD4 < 500 to [23] 10.00 [2.20, 87.00]% 200 < CD4 < 350 per year % moving from 200 < CD4 < 350 to [23] 5.30 [0.80, 82.70]% 350 < CD4 < 500 per year % moving from 200 < CD4 < 350 to [23] 16.20 [5.00, 86.90]% 50 < CD4 < 200 per year % moving from 50 < CD4 < 200 to [23] 11.70 [3.20, 68.60]% 200 < CD4 < 350 per year % moving from 50 < CD4 < 200 to CD4 < 50 per year [23] 9.00 [1.90, 72.30]% % moving from CD4 < 50 to 50 < CD4 < 200 per year [23] 11.10 [4.70, 56.30]% Population transitions Risk Average length of time spent as sex worker National 12 years Average length of time spent as client of sex worker National 15 years Age Defined by width of age bins N/A The transitions between CD4 categories were determined following extensive literature review and data synthesis; full details are contained in the Supplementary Materials. The transitions between cascade stages were informed by local data. 5 Stuart RM et al. Journal of the International AIDS Society 2018, 21:e25068 http://onlinelibrary.wiley.com/doi/10.1002/jia2.25068/full | https://doi.org/10.1002/jia2.25068 attributable to significant increases in testing programme cov- 3.3 | The impact of achieving the Fast-Track erage throughout the entire country [26]. In particular, the targets proportion of MSM aware of their status was estimated to increase from 30% to 63%, and in FSW the increase was even If the status quo were maintained across the care and treat- more marked (from 30% to 70%, consistent with other esti- ment cascade (i.e. if current programmatic coverage were mates [25,27]). Expanded treatment eligibility and major maintained such that the proportions of those diagnosed, in investments in treatment increased the proportion of diag- care, receiving treatment and virally suppressed remained nosed people receiving treatment from 8% to 64% (although constant), we estimate that Johannesburg could expect treatment coverage remained low among FSW, at 23% 303,000 new infections and 96,000 AIDS-related deaths [26,27]). between 2016 and 2030, such that there would be around 769,000 PLHIV by 2030 (Figure 3; base case). In contrast, achieving the 2020 and 2030 Fast-Track targets would avert 3.2 | What will it take to achieve the Fast-Track 177,000 infections and 56,500 deaths by 2030 (reductions of targets? around 58%), leading to a 26% reduction in the number of We estimate that achieving the Fast-Track targets would PLHIV in 2030 compared to baseline and a 2.2%point reduc- require an additional 135,000 HIV diagnoses and 232,000 tion in prevalence compared to baseline (Figure 3). While the people on treatment by 2020 relative to 2016 (an increase in treatment scale-up that would be required in order to meet around 80% over 2016 treatment numbers; see Figure 2b). the 90-90-90 targets is significant, it is on par with the treat- Using unit cost estimates from the South African HIV and TB ment scale-up that took place over the first half of the decade Investment Case [28] in combination with the model’s esti- (Figure 2b). Furthermore, achieving the 90-90-90 targets by mates of the programme coverage targets required to reach 2020 would mean that 95-95-95 targets were within reach. the first two stages of the 90-90-90 and 95-95-95 targets, we broadly estimated total investment requirements of ZAR 3.4 | Scaling up primary prevention programmes 6.94 billion (USD 0.51 billion) over the 4 years to 2020, and reduces the cost of reaching the Fast-Track targets: ZAR 32.1 billion (USD 2.37 billion) over the 14 years to 2030 impact of biomedical prevention service scale-up (Tables 2 and 3). Taking into account the proportion of the population living in Johannesburg, our estimates of Johannes- We considered the impact of implementing the targeted burg’s investment requirements to 2020 are commensurate biomedical prevention services outlined in sub-objective 1.1.2 with those provided in the 2017 to 2022 NSP[24]. Around of South Africa’s National Strategic Plan (NSP) for 2017 to 10% of total investments would be required for testing pro- 2022. This sub-objective outlines a comprehensive approach grammes and the remainder for funding significant expansions for the implementation of targeted biomedical prevention ser- to treatment and care programmes. vices, with 5 components related to the provision of HIV Table 2. Coverage and investment levels required to achieve the first two 90 targets and the first two 95 targets under different assumptions about prevention programme coverage. Annual average 2016 2017 2018 2019 2020 2021–2030 Total 2017–2030 Achieving 90% aware by 2020 and 95% aware by 2030 Adult testing rates 51% 58% 65% 72% 80% 80% – Target population (millions) 3.8 3.9 4.0 4.1 4.2 4.3 59.2 Number of tests required (millions) 1.9 2.3 2.6 3.0 3.4 3.4 45.3 Investment required (ZAR millions) 183 213 245 278 317 183 2,883 Requirements to achieve 90% on treatment by 2020 and 95% by 2030 without scale-up in other prevention programmes Current prevention programme coverage maintained Number required on ART (millions) 0.299 0.368 0.419 0.474 0.530 0.568 7.474 Investment required (ZAR millions) 1,161 1,426 1,626 1,838 2,055 2,205 28,999 Requirements to achieve 90% on treatment by 2020 and 95% by 2030 with VMMC scale-up Number required on ART (millions) 0.299 0.366 0.416 0.468 0.521 0.556 7.331 ART investments required (ZAR millions) 1,160 1,422 1,612 1,814 2,020 2,158 28,445 Requirements to achieve 90% on treatment by 2020 and 95% by 2030 with VMMC and condom distribution scale-up Number required on ART (millions) 0.299 0.367 0.416 0.468 0.521 0.549 7.265 ART investments required (ZAR millions) 1,160 1,422 1,614 1,816 2,020 2,132 28,188 Requirements to achieve 90% on treatment by 2020 and 95% by 2030 with VMMC and condom distribution scale-up + FSW strategy Number required on ART (millions) 0.299 0.367 0.416 0.468 0.519 0.542 7.189 ART investments required (ZAR millions) 1,160 1,422 1,614 1,814 2,016 2,103 27,893 6 Stuart RM et al. Journal of the International AIDS Society 2018, 21:e25068 http://onlinelibrary.wiley.com/doi/10.1002/jia2.25068/full | https://doi.org/10.1002/jia2.25068 Table 3. Summary of the total investments required to achieve the first two 90 targets and the first two 95 targets, as well as the savings made under different assumptions about prevention programme scale-up and the impact in terms of infections averted. Current prevention VMMC and condom programme coverage VMMC VMMC and condom distribution scale-up + maintained (base) scale-up distribution scale-up FSW strategy Investment in ART & HTC to achieve 31,882 31,328 31,071 30,776 90/90 & 95/95, 2017–30 (ZAR m) Cost of prevention programme – 266 1,321 Not estimated scale-up, 2017–30 (ZAR m) Savings in ART & HTC programmes – 554 811 1,106 relative to base, 2017–30 (ZAR m) Net savings relative to base, 2017–30 (ZAR m) – 288 À510 Not estimated Infections averted relative to base, 2017–30 – 4% 8% 14% (a) (b) (c) (d) Figure 3. Key epidemic indicators assuming current programme coverage maintained and Fast-Track targets are met by scaling up HIV diag- nosis, treatment and viral suppression, with all other programmes maintained at their latest reported coverage levels. testing and counselling (HTC), 2 related to the voluntary med- programmes in line with national targets, which aim for 2.5 ical male circumcision (VMMC) programme, and a final compo- million medical circumcisions over 2017 to 2020 and the nent promising the provision of male and female condoms annual distribution of 850 million male condoms and 40 mil- (plus compatible lubricant) in all public and private health facil- lion female condoms. Assuming that the national targets are ities, in secondary schools, tertiary institutions, non-traditional proportional to population size, we translate these to targets community settings[24]. Since the Fast-Track targets subsume for Johannesburg of 220,000 medical circumcisions over the HTC targets, we do not model this separately here. 2017–2020, and annual provision of 68 million male condoms Rather, we model the impact of scaling up the VMMC pro- and 3.2 million female condoms. On programme efficacy, we gramme and the male and female condom distribution assume that circumcision reduces the probability of HIV- 7 Stuart RM et al. Journal of the International AIDS Society 2018, 21:e25068 http://onlinelibrary.wiley.com/doi/10.1002/jia2.25068/full | https://doi.org/10.1002/jia2.25068 negative males acquiring infection by 58% [23] and that 4 | DISCUSSION reaching the condom distribution targets would result in con- doms being used in 80% of acts between casual partners The results presented in this paper highlight the tremendous (compared to 50% in 2012; see Table S1 and Figure S2 in impact that achieving the Fast-Track targets would have on Supplementary Material for calculations). the HIV epidemic in the City of Johannesburg, while also We calculate that reaching the VMMC target would require demonstrating the still-crucial role that primary prevention an investment of ZAR 266 million (USD 19.7 million) over programmes play in both reducing HIV transmission and in 4 years, while reaching the condom distribution targets would reducing the future financial burden of achieving the Fast- require annual investment of ZAR 94 million (USD 6.9 million) Track targets. We focused on estimating the investments that (see Table S1 in Supplementary Material for calculations). In would be required in order to get 90% of people diagnosed combination, attaining these two targets by 2020 and there- by 2020 and 90% of those diagnosed onto treatment (with after maintaining them would reduce cumulative new infec- those percentages increasing to 95% by 2030) under different tions between 2017 and 2030 by a further 9% compared to a assumptions around prevention programme scale-up. We also base case scenario in which coverage of these two pro- identified key leakages along the cascade that currently pre- grammes is maintained at current levels. This would imply sent barriers to achieving the Fast-Track targets, with key ZAR 811 million (USD 59.9 million) in savings to the cumula- problem areas being the rate of linkage to care within tive treatment budget to 2030 (Table 3). Investing in the 3 months and the rate of viral suppression among those VMMC programme in particular is cost saving: we find that receiving treatment. The state of the care and treatment cas- scaling up VMMC alone would deliver a ZAR 554 million cade in Johannesburg has improved greatly over first half of (USD 40.1 million) saving to the overall treatment budget out this decade, but there is still significant work to be done if the to 2030, and thus would deliver modest net savings even Fast-Track targets are to be achieved, and these areas should after accounting for the cost of scale-up. be particular focal points. Johannesburg’s Implementation Plan for 2016/17 is a testa- ment to the health authorities’ commitment to addressing the 3.5 | Impact of implementing the National Sex programmatic scale-up implied by the Fast-Track targets. HTC Worker HIV Plan will be promoted in many different ways: through campaigns We consider the impact of the South African National Sex in high-transmission areas and among key populations, by Worker HIV Plan 2016 to 2019 [25]. In light of the signifi- making HTC access more convenient (e.g. alternative opening cant body of evidence demonstrating the impact of structural times of services, mobile provision) and integrated (e.g. rein- interventions on the health service use and uptake of vigorated provider-initiated counselling and testing, HTC in biomedical services among sex workers[29,30], we would like clinic waiting areas, improved index testing), through to capture the impact of the structural elements of the pack- approaches to diagnose more men (e.g. targeting factories, age, notably the Human Rights Package (comprising law transport hubs, tertiary institutions) and by focusing on yield reform, decriminalization, legal literacy and legal service use), through continuous monitoring and adjustment of efforts. the Social Capital Building Package (comprising community Counsellors will be distributed according to need and clinics’ empowerment, and collectivization), and the Economic HTC targets, and the electronic ART patient record system Empowerment Package (comprising skill building, career-path will be expanded to include HTC data. Equally important is defining, and participation in cooperatives and education the authorities’ scale-up efforts to link diagnosed HIV cases to interventions). Full-scale financing and implementation of care and treatment (which will help with the estimated need these structural packages in combination with the implemen- of putting an additional 232,000 people on treatment by tation of the Health Care package is associated with out- 2020). Given that linkage to care is especially challenging if comes of: [1] 95% condom use in FSW/client commercial HTC is provided at community level, various providers are acts in 2020, and [2] provision of PrEP to 3000 HIV-negative tasked with facilitating this process, using the Department of FSWs in 2016, with coverage then extended to all HIV-nega- Health’s ward based teams and civil society organizations. tive FSWs. Assuming that the national targets are propor- Adequate supply of HIV drugs and prevention of stock-outs tional to population size, we translate the latter target to using the new “Stock Visibility Solution” infrastructure are also mean that 240 HIV-negative FSWs in Johannesburg would in the plan. In order to support adherence, differentiated care be provided with PrEP in 2016 (4% coverage), scaling up to options are being or have been introduced, including decen- 4800 by 2020 (80% coverage). If implemented in addition to tralized medicine delivery schemes, adherence clubs, fast the scale-up of HIV diagnosis and treatment programmes queuing, tracing of those lost to care, and enhanced adher- needed to attain the Fast-Track targets and the planned ence counselling. This will be supported by improved data sys- scale-up to VMMC and condom distribution services, we esti- tems to identify patients in need of laboratory monitoring, mate that implementing the SW strategy would reduce drug refill and additional adherence support. In parallel, scale- cumulative new infections between 2016 and 2030 by a fur- up plans are in place for other core HIV prevention services, ther 7%, and would mean that the Fast-Track targets would including the National Sex Worker HIV plan, plans for be attainable with ZAR 295 million (USD 21.9 million) less enhanced condom distribution based on evidence of need and required for total treatment investment (Table 3). Unfortu- stock management, and increasing involvement of general nately we are unable to determine cost estimates for the practitioners to provide free VMMC services. structural elements of the package due to a lack of cost and We note several limitations to the analyses presented. First, effectiveness data, so we are unable to determine the magni- limitations in data availability and reliability can lead to uncer- tude of savings overall. tainty surrounding projected results, and these uncertainties 8 Stuart RM et al. Journal of the International AIDS Society 2018, 21:e25068 http://onlinelibrary.wiley.com/doi/10.1002/jia2.25068/full | https://doi.org/10.1002/jia2.25068 were not quantified. Second, many of the Optima HIV model and South Africa has also been successful in initiating large parameters (most notably those related to transmission proba- numbers of people onto ART. However, significant work bilities, disease progression and programme efficacy) were remains to be done on a national scale, particularly with sourced from clinical and research studies, and may differ respect to improving retention and adherence rates. from the values that would be observed in Johannesburg. Enhanced analysis of the HIV care cascade is crucial for Third, our estimates of the impact of biomedical prevention guiding the large investments that will be required to achieve service scale-up assume that the planned investments in con- the Fast-Track targets. In particular, translating broad political dom and VMMC scale-up would be met with increased con- targets into actionable stratagems requires detailed knowl- dom use and increase rates of medical circumcision. edge of the current status of the care and treatment cascade, Particularly with respect to condom usage, this may be an as well as an understanding of the likely future trajectories of optimistic assumption as behaviour changes of this type can the HIV epidemic. There is scope for mathematical models to be difficult to attain in practice. Fourth, our analysis of the help with the latter, and for improved data collection and anal- costs assumes that all programmes continue to operate at cur- ysis to help with the former. Conducting analyses at a city rent levels of efficiency, and does not consider the effect of level means that goals can be mapped against operational potential efficiency gains. There are various studies that have budgets, and translated into targets for the clinical and service highlighted the potential for efficiency gains in the implemen- provision networks. Such analyses would be best conducted tation of prevention programmes [31], including specific stud- within a broader national context, in order to ensure coher- ies on HTC [32], VMMC [33], condom distribution programme ence across different jurisdictional levels. Our modelling [34], and sex worker programme [35] efficiency in various con- methodology would be well-suited for many other cities, espe- texts in sub-Saharan Africa. More functional integration of HIV cially those who have committed to achieving the Fast-Track services, as is currently happening in South Africa within the targets, and could be used to answer similar questions about Integrated Chronic Disease Management and Chronic Treat- resource needs and impact in a range of settings. ment Adherence guidelines, creates economies of scope and may also improve the client experience. Technical efficiency gains, in areas where large volumes of patients are to be 5 | CONCLUSIONS served such as Johannesburg, are feasible to gain economies of scale and reduce unit costs (e.g. South Africa’s task shifting Enhanced analysis of the HIV care cascade is crucial for guiding to lower level health cadres, price reductions of drugs and the large investments that will be required in order to achieve diagnostic tests). Also, efficiency gains may be made through the goal of ending AIDS by 2030. By providing annual denomina- better targeting of programmes, the use of different service tors of people living with HIV and requiring services of HIV test- delivery modalities, or cutting back on indirect programme ing, linkage, treatment initiation and treatment maintenance, this costs. It is therefore worth investigating whether such effi- analysis hopes to inform target setting for the Fast-Track ciency gains may be made here. However, while there may be response and as a baseline for UTT. For Johannesburg, getting further cost-savings possible in the Johannesburg HIV pro- HIV service delivery right – guided by ruthless data tracking and gramme in the future, some are likely offset by the effort enhanced analysis – in an era of prolonged and increasing urban- required to identify additional and harder-to-reach cases and ization and rapidly growing ART programme costs is vital both for get them linked to HIV care and virally suppressed. HIV testing the individuals who live there, and for the economy as a whole. yield illustrates this challenge of case identification well: in South Africa, six people needed to be tested in 2005 to find AUTHORS’ AFFILIATIONS one new HIV case in 2005; in 2015 this increased to testing 1 Department of Mathematical Sciences, University of Copenhagen, Copenhagen, 18 people [36]. Another factor potentially offsetting the savings Denmark; 2Burnet Institute, Melbourne, Australia; 3The World Bank, Washington gained through efficiencies is the need to switch more ART cli- DC, USA; 4School of Physics, University of Sydney, Sydney, Australia; 5Depart- ents to second line HIV treatment, which remains more costly. ment of Health, Gauteng Province, Johannesburg, South Africa; 6Department of Health, Johannesburg Health District, Johannesburg, South Africa; 7National The importance to Johannesburg of meeting the Fast-Track Department of Health, Pretoria, South Africa; 8School of Public Health, Univer- targets–thereby averting an estimated 177,000 infections and sity of the Witwatersrand, Johannesburg, South Africa; 9Monash University, 56,500 deaths by 2030 and reducing the number of people Melbourne, Australia living with HIV by a third – cannot be overemphasized. Ensur- ing scale and quality of the HIV treatment programme is vital COMPETING INTERESTS for the city’s economic prosperity and for South Africa as a The authors declare no competing interests. whole. Johannesburg’s commitment towards the Fast-Track targets echoes the national dedication to achieving the targets AUTHORS’ CONTRIBUTIONS and ending AIDS by 2030. “Universal test and treat” (UTT) became national policy in September 2016, with accelerated RMS and NF-H wrote the manuscript. Analyses were carried out by RMS and NF-H, with substantial technical inputs and data provided by CCK, EM, VM, FM, efforts being made for HIV clinic decongestion, down-referral BM and TM. Supervision, oversight and guidance was provided by YP, PB, MG of HIV clients, decentralization of services and community- and DPW. based monitoring to facilitate UTT. Substantial investments continue to be made throughout South Africa to support HIV ACKNOWLEDGEMENTS testing, linkage to care, pre-ART care, and treatment initiation, The authors are grateful for the collaborative efforts of Lihle Ngubane (District maintenance and adherence. The focus on diagnosing HIV- of Johannesburg), Francis Akpan, Khudugo Letsoalo, Eunice Sithole, and Nom- positive individuals over recent years has paid off with large buso Madonsela (Gauteng Province Department of Health), Hasina Subedar increases in the proportion of people aware of their status, (the National Department of Health of South Africa) and the World Bank (Paolo 9 Stuart RM et al. Journal of the International AIDS Society 2018, 21:e25068 http://onlinelibrary.wiley.com/doi/10.1002/jia2.25068/full | https://doi.org/10.1002/jia2.25068 Belli). Andrew Scheibe and Maxim Berdnikov provided invaluable comments on 21. Siedner MJ, Ng CK, Bassett IV, Katz IT, Bangsberg DR, Tsai AC. Trends early results from this study. in CD4 count at presentation to care and treatment initiation in sub- Saharan Africa, 2002-2013: a meta-analysis. Clin Infect Dis. 2015;60 (7):1120–7. FUNDING 22. Govender S, Otwombe K, Essien T, Panchia R, de Bruyn G, Mohapi L, et al. Funding for this study was provided by the World Bank Group and National CD4 counts and viral loads of newly diagnosed HIV-infected individuals: implica- Health and Medical Research Council. tions for treatment as prevention. PLoS ONE. 2014;9(3):e90754. 23. Stuart RM. The optima consortium for decision science. Volume VI: Para- meter data sources. In: Optima HIV User Guide 2018. [Cited 2018 Jan 16] Available from: optimamodel.com/http://optimamodel.com/user-guide. REFERENCES 24. SANAC. National Strategic Plan on HIV, STIs and TB 2017-2022. 2017. 1. UNAIDS. Fast-Track Cities: Ending the AIDS epidemic. 2014. 25. SANAC. The South African National Sex Worker HIV Plan 2016 - 2019. 2. UNAIDS. On the Fast-Track to end AIDS. 2016. 2016. 3. About the Fast-Track Cities Initiative. 2016. [Cited 2018 Jan 16]. Available 26. SANAC. South Africa Global AIDS Response Progress Report. 2015. from: http://www.iapac.org/cities/. 27. UCSF AHIW. South African Health Monitoring Study (SAHMS), Final 4. UNAIDS. The Cities Report. 2014. Report: The Integrated Biological and Behavioural Survey among Female Sex 5. UNAIDS. The Gap Report. 2014. Workers, South Africa 2013-2014. San Francisco: UCSF; 2015. 6. Gardner EM, McLees MP, Steiner JF, Del Rio C, Burman WJ. The spectrum 28. Department of Health SA, and South African National AIDS Council. South of engagement in HIV care and its relevance to test-and-treat strategies for African HIV and TB Investment Case - Summary Report Phase 1; 2016 March. prevention of HIV infection. Clin Infect Dis. 2011;52(6):793–800. 29. Shannon K, Strathdee SA, Goldenberg SM, Duff P, Mwangi P, Rusakova M, 7. Sibanda EL, Weller IV, Hakim JG, Cowan FM. The magnitude of loss to fol- et al. Global epidemiology of HIV among female sex workers: influence of struc- low-up of HIV-exposed infants along the prevention of mother-to-child HIV tural determinants. Lancet. 2015;385(9962):55–71. transmission continuum of care: a systematic review and meta-analysis. AIDS. 30. Vassall A, Chandrashekar S, Pickles M, Beattie TS, Shetty G, Bhattacharjee 2013;27(17):2787–97. P, et al. Community mobilisation and empowerment interventions as part of HIV 8. Barnighausen T, Chaiyachati K, Chimbindi N, Peoples A, Haberer J, Newell prevention for female sex workers in Southern India: a cost-effectiveness analy- ML. Interventions to increase antiretroviral adherence in sub-Saharan Africa: sis. PLoS ONE. 2014;9(10):e110562. a systematic review of evaluation studies. Lancet Infect Dis. 2011;11 31. Bautista-Arredondo S, Sosa-Rub ı SG, Opuni M, Kwan A, Chaumont C, Coet- (12):942–51. zee J, et al. Assessing cost and technical efficiency of HIV prevention interven- 9. Kilmarx PH, Mutasa-Apollo T. Patching a leaky pipe: the cascade of HIV care. tions in sub-Saharan Africa: the ORPHEA study design and methods. BMC Curr Opin HIV AIDS. 2013;8(1):59–64. Health Serv Res. 2014;14(1):1–11. 10. MacCarthy S, Hoffmann M, Ferguson L, Nunn A, Irvin R, Bangsberg D, 32. Sharma M, Ying R, Tarr G, Barnabas R. Systematic review and meta-analysis et al. The HIV care cascade: models, measures and moving forward. J Intern of community and facility-based HIV testing to address linkage to care gaps in AIDS Soc. 2015;18(1):19395. sub-Saharan Africa. Nature 2015;528(12):S77–85. 11. Ortego C, Huedo-Medina TB, Llorca J, Sevilla L, Santos P, Rodriguez E, 33. Awad SF, Sgaier SK, Tambatamba BC, Mohamoud YA, Lau FK, Reed JB, et al. Adherence to highly active antiretroviral therapy (HAART): a meta-analy- et al. Investigating voluntary medical male circumcision program efficiency gains sis. AIDS Behav. 2011;15(7):1381–96. through subpopulation prioritization: insights from application to Zambia. PLoS 12. Pin ~eiru a A, Sierra-Madero J, Cahn P, Guevara Palmero RN, Mart ınez Bui- ONE. 2015;10(12):e0145729. trago E, Young B, et al. The HIV care continuum in Latin America: challenges 34. Sandøy IF, Zyaambo C, Michelo C, Fylkesnes K. Targeting condom distribu- and opportunities. Lancet Infect Dis. 2015;15(7):833–9. tion at high risk places increases condom utilization-evidence from an interven- 13. Medland NA, McMahon JH, Chow EP, Elliott JH, Hoy JF, Fairley CK. The tion study in Livingstone, Zambia. BMC Public Health. 2012;12(1):1–12. HIV care cascade: a systematic review of data sources, methodology and com- 35. Siapka M, Remme M, Obure CD, Maier CB, Dehne KL, Vassall A. Is there parability. J Intern AIDS Soc. 2015;18:20634. scope for cost savings and efficiency gains in HIV services? A systematic review 14. Tea B€ arnighausen. Interventions to increase antiretroviral adherence in of the evidence from low- and middle-income countries. Bull World Health sub-Saharan Africa: a systematic review of evaluation studies. Lancet Infect Dis. Organ. 2014;92:499–511. 2011;11(12):942–51. 36. Williams B. Ending AIDS: Progress and prospects for the control of HIV 15. Yehia BR, Stephens-Shields AJ, Fleishman JA, Berry SA, Agwu AL, Metlay and TB in South Africa. Regional Consultation on HIV Testing Services in East JP, et al. The HIV Care Continuum: changes over time in retention in care and and Southern Africa. 2015. viral suppression. PLoS ONE. 2015;10(6):e0129376. 16. Johnson L. THEMBISA version 1.0: A model for evaluating the impact of HIV/AIDS in South Africa. 2014. 17. Eaton JW, Menzies NA, Stover J, Cambiano V, Chindelevitch L, Cori A, et al. Health benefits, costs, and cost-effectiveness of earlier eligibility SUPPORTING INFORMATION for adult antiretroviral therapy and expanded treatment coverage: a com- bined analysis of 12 mathematical models. Lancet Glob Health. 2014;2(1): e23–34. Additional Supporting Information may be found in the online 18. Kerr CC, Stuart RM, Gray RT, Shattock AJ, Fraser-Hurt N, Benedikt C, version of this article: et al. Optima: a model for HIV epidemic analysis, program prioritization, Figure S1. shows the model estimates of HIV prevalence in and resource optimization. J Acquir Immune Defic Syndr. 2015;69(3):365–76. each sub-population compared to available data. 19. Stuart RM, Grobicki L, Haghparast-Bidgoli H, Skordis-Worrall J, Keiser O, Estill J, et al. How should HIV resources be allocated? Lessons learnt from Table S1. Calculations to determine the cost of scaling up the applying Optima HIV in 25 countries. 2017. VMMC and condom distribution programmes in Johannesburg 20. Fomundam H. Surveillance of newly diagnosed HIV-positive persons in South Africa: What have we learnt about the second 90? Evidence from the dis- tricts. International AIDS Conference; July 21st 2016; Durban 2016. 10