The Impact of Scaling up Dolutegravir on Antiretroviral Resistance in South Africa

Background Rising resistance of HIV-1 to non-nucleoside reverse transcriptase inhibitors (NNRTIs) threat-ens the success of the global scale-up of antiretroviral therapy (ART). The switch to WHO-recommended dolutegravir (DTG)-based regimens could reduce this threat due to DTG’s high genetic barrier to resistance. We used mathematical modelling to examine the impact of the scale-up of DTG-based ART on NNRTI pre-treatment drug resistance (PDR) in South Africa, 2019-2040. results We adapted the MARISA (Modelling Antiretroviral drug Resistance In South Africa) model, an epidemiological model of the transmission of NNRTI resistance in South Africa. We modelled the introduction of DTG in 2019 under two scenarios: DTG as (cid:28)rst-line regimen for ART-initiators, or DTG for all patients, including patients on suppressive NNRTI-based ART. Due to safety concerns related to DTG during pregnancy, we assessed the impact of prescribing DTG to all men and in addition to i) women beyond reproductive age, ii) women beyond reproductive age or using contraception, and iii) all women. The model projections show that, compared to

The rollout of antiretroviral therapy in South Africa is estimated to have prevented 0.73 million 2 HIV infections between 2004 and 2013 as well as 1.72 million deaths between 2000 and 2014 3 [1,2]. However, the spread of non-nucleoside reverse transcriptase inhibitor (NNRTI) resistant 4 viruses is threatening this success [3]. An estimated 16% of AIDS-related deaths and 8% of ART 5 costs will be attributable to HIV drug resistance up to 2030 in the sub-Saharan African countries 6 that reached HIV pre-treatment drug resistance levels above 10% in 2016 [4]. 7 In Southern Africa, dolutegravir (DTG), an integrase inhibitor drug, will be introduced on a 8 large scale as part of xed-dose combinations of Tenofovir, Lamivudine, and Dolutegravir (TLD) 9 [5]. With a high genetic barrier to resistance, DTG has the potential to curb the spread of 10 antiretroviral resistance, as it is highly eective, well tolerated and aordable in resource-limited 11 settings [6,7,8,9]. Mathematical models explored the eectiveness and cost-eectiveness of 12 prescribing DTG to all ART-initiators [10]. These models found that the introduction of DTG 13 was cost-saving and reduced HIV mortality in people living with HIV who initiate ART [10]. 14 The introduction of DTG has been complicated by the increased risk of neural tube defects 15 (NTD) in women living with HIV using DTG at the time of conception [11] and other potential 16 side eects such as weight gain [12,9]. Concerns surrounding NTD risk have delayed the rollout of 17 DTG and, in some settings, led to recommending DTG-based regimens only for men and women 18 who are not at risk of pregnancy [13,14]. For South Africa, a mathematical modelling study 19 showed that DTG-based rst-line ART for all women of child-bearing potential would prevent 20 more deaths among women and more sexual HIV transmissions than either NNRTI-based ART 21 for women of child-bearing potential or women without contraception, but increase pediatric 22 deaths [15]. In its 2019 guidelines, the WHO recommends DTG in combination with nucleoside 23 reverse-transcriptase inhibitors (NRTI) for rst-line ART, with the proviso that women should 24 be provided with information about benets and risks to make an informed choice regarding the 25 use of DTG [16]. 26 It is likely that in many settings people living with HIV on NNRTI-based rst-line ART will be 27 switched to DTG-based ART, however, the rate of the transition will vary between countries and 28 settings. For second-line ART, WHO recommends DTG-based ART in people living with HIV 29 for whom a NNRTI-based rst-line regimen has failed [16]. Again, the rate of switching to DTG-30 based second-line ART will vary, inuenced by concerns about the development of dolutegravir 31 resistance in patients who switch with pre-existing resistance to NRTIs [17]. Taken together, it 32 is likely that for the foreseeable future a considerable fraction of people living with HIV, and 33 particularly women, may continue to rely on NNRTI-based ART regimens, even in the case when 34 with subsequent treatment-specic suppression (Supp compartment in Fig 1)   . CC-BY 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19010132 doi: medRxiv preprint Diagnosis rate, treatment initiation rate to NNRTI Calibrated by tting MARISA to Thembisa model (see [18]) from 2005 to 2016 (see S1 Text Section 2.1 and Table 3) NNRTI suppression and failure rates Estimated with individual epidemiological data from IeDEA-SA [19] (see S1 Text Table 1) PI suppression and failure rates Estimated with data from IeDEA-SA [19] (see S1 Text Table 1) Rate of switching from NNRTI to PI (before 2019) Estimated with data from IeDEA-SA and then adjusted (see S1 Text Section 2.1) Added rates DTG suppression and failure rates Assumption: same rate as for NNRTI (relaxed in S1 Text Section 5.1) DTG initiation rate (from 2019) Same DTG initiation rate as for NNRTI (for DTG-inel. Diagnosis rate from 2019 (distribution across DTG-eligibility classes) Rate of switching from NNRTI to PI (DTG-inel.) Rate of switching from DTG to PI Switching rate from NNRTI to DTG (maintenance therapy) Switching rate from NNRTI to DTG (switch therapy) DTG-ineligible women, the cascade of care remains unchanged after 2019 (Fig 1). 57 Calibration and extension of the MARISA model 58 We previously calibrated the MARISA model by combining dierent sources of data. Rates either 59 related to treatment response (NNRTI-or PI-based regimen) or disease progression (characterized 60 by CD4 counts) were estimated using clinical data from data from ve cohorts in South Africa 61 (Aurum Institute, Hlabisa, Khayelitsha, Kheth'Impilo and Tygerberg) that participate in the 62 IeDEA collaboration [19]. These cohorts provided longitudinal information for 54,016 HIV-63 infected adults. Other parameters were either estimated from the literature (e.g. resistance-64 related parameters) or tted to estimates from the Thembisa model (e.g. diagnosis rates and 65 treatment initiation rates). Thembisa is a demographic projection model on which the ocial 66 UNAIDS estimates for South Africa are based [20]. More details about the calibration procedure 67 can be found in [18] and in the supplementary materials (S1 Text, Section 1.2). 68 We added and modied parameters in order to model the introduction of DTG. We assumed that 69 the DTG initiation rate γ D→T 3 (t) is the same as the NNRTI initiation rate γ D→T 1 (t) from 2019. 70 Both NNRTI and DTG initiation rates increase until 2022, as a consequence of the Treat-All 71 policy that was implemented in 2017. From 2022 onwards, they are assumed to remain constant 72 (S1 Text, Section 2.1). Finally, we xed switching rates from NNRTI-to DTG-based regimens 73 for both eligible suppressed (see Scenarios) and all failing individuals (γ S 1 →S 3 (t) and γ F 1 →T 3 (t) 74 respectively) to 1 year −1 . Finally, we assumed that suppressed individuals would stay suppressed 75 when switching, while failing individuals would start DTG in the Treat init. compartment (see 76 Fig 1). The main parameters are summarized in Table 1. . CC-BY 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19010132 doi: medRxiv preprint The percentages of women eligible for DTG in b) and c) were determined by analyzing cohort 94 data from IeDEA, which show that 17.5% adult women on ART are 50 or older [19], and estimates 95 on the use of contraception from the World Bank [21](see S1 Text, Section 3.1). Of note, to model 96 . We thus assumed that in all scenarios DTG will be used in 98 second-line regimens for people failing NNRTI-based rst-line ART. 99 Additional analyses 100 We predicted the impact of dierent levels of DTG introduction on the level of NNRTI failure. 101 We considered the scenario where DTG was prescribed to ART initiators and those on NNRTI-102 We assessed the impact of dierent switching rates from NNRTI-to DTG-based regimens, xed 114 to 1 year −1 for both suppressed and failing individuals. We varied both rates γ S 1 →S 3 (t) and 115 γ F 1 →T 3 (t) within a range corresponding to a time to switch of between 0.5 and 10 years after 116 start of ART. For each analysis, the percentage of women who are DTG-eligible varied from 0% 117 to 100%. . CC-BY 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19010132 doi: medRxiv preprint The values of eight parameters were varied in the sensitivity analysis: three transmission-related 120 parameters (percentage of men who have sex with men (MSM), probability of male-to-male in-121 fection per sexual contact, and HIV prevalence ratio between MSM and heterosexuals), four 122 resistance-related parameters (resistance rates, reversion to wild-type rate and the eect of 123 NNRTI-resistance on NNRTI ecacy) and one parameter related to treatment (ecacy of DTG-124 based treatment). Multivariate uncertainty within specied ranges was assessed using Latin 125 hypercube sampling [22]. Each model estimate is reported with a 95% sensitivity range. Further 126 details are available in S1 Text, Section 3.2. In addition, we also investigated the impact of 127 lower treatment-initiation rates than suggested by the Treat-All policy (as suggested by [23]), 128 treatment interruption and higher ecacy of DTG on NNRTI PDR (S1 Text, Section 5). even when given to all men and women ( Fig 3A). 145 Restricting DTG-based ART to men to avoid the risk of DTG-associated neural tube defects 146 in newborns will not curb the increase in NNRTI resistance: the prevalence of resistance is 147 predicted to increase over the entire study period, reaching values of close to 50% by 2040 148 ( Fig 3B) both ART-initiators and individuals already on NNRTI-based ART (Fig 3). 155 Impact of switching rates 156 We calculated levels of NNRTI resistance for 2035 for dierent average switching delays and per-157 centages of women eligible for DTG-based ART. We considered the eect of a modied switching 158 rate both in suppressed individuals ( Fig 4A) and in individuals on a failing regimen (Fig 4B). 159 The predicted levels of NNRTI resistance range from 14    ART: increasing use of DTG-based regimens was the strategy with the greatest potential to curb 187 the spread of NNRTI resistance. The latter strategy will also lower the risk of virologic failure 188 in women who have to rely on NNRTI-based ART in the future. 189 While some countries, such as South Africa, rst considered to limit access to DTG to men, 190 menopausal women, and women using long-term family planning as a potential policy, the new 191 WHO guidelines state that women should not in principle be excluded from DTG-based ART, 192 October 22, 2019 8 . CC-BY 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity. will continue to rely on NNRTI-based ART. Therefore, even with the rollout of DTG, NNRTI 202 resistance will continue to be relevant for these women. Compared with other modelling work 203 that assessed risks and benets of DTG introduction (e.g. [15]), our model focused on its indirect, 204 population-level impact on NNRTI resistance. Rather than assigning a level of NNRTI resistance 205 that is xed over time, HIV care and disease stages (as in [15]), our model considered the dynamic 206 development of NNRTI resistance under relevant scenarios. 207 Our model also has several limitations. First, real-world data on the ecacy of DTG, especially 208 in resource-limited settings are scarce. Therefore, we conservatively assumed that DTG has a 209 similar ecacy as NNRTI. The eect of higher DTG ecacy, as suggested by several studies 210 [12,9], is investigated in supplementary analyses (see S1 Text, Section 5.3). Second, predictions 211 of levels of NNRTI resistance over the next twenty years are naturally uncertain, as reected by 212 the wide sensitivity ranges in Fig 3. However, despite the uncertainty, it is clear that the dierent 213 strategies of rolling out DTG-based ART inuenced the levels of NNRTI resistance. Finally, the 214 MARISA model includes some simplifying assumptions, e.g. we did not model prevention of 215 mother to child transmission (PMTCT), or treatment interruption. However, relaxing some of 216 these assumptions did not drastically change our conclusion (see S1 Text, Section 5). 217 Another limitation of this study is the fact that the MARISA model does not take into ac-218 count resistances to neither NRTI nor DTG. In the context of the introduction of DTG-based 219 ART, modelling of NRTI resistance is particularly relevant as individuals starting on DTG as a 220 October 22, 2019 9 . CC-BY 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint  functional monotherapy due to resistance to both NRTI backbones -tenofovir and lamivudine -221 experience higher risk of treatment failure [25]. As they are considerably less frequently trans-222 mitted [26] and revert back quickly [27,28], NRTI resistances might primarily be an issue for 223 ART-experienced individuals and more specically, in patients failing NNRTI-based regimens, 224 who often exhibit high levels of NRTI resistance [29]. These patients who are on non-suppressive 225 NNRTI-based regimens are expected to switch to DTG, either after identication of treatment 226 failure, following the new WHO guidelines, or blindly [17]. In the context of modelling the 227 DTG rollout, this consideration has two important implications. First, patients currently failing 228 NNRTI-based regimens are expected to have higher DTG failure rates, mainly due to previously 229 acquired NRTI resistance. Second, due to ongoing viral replication and due to pre-existing NRTI 230 resistance, they are at higher risk of accumulating resistance, which may also lead to the emer-231 gence of DTG resistance. So far, data on emergence of DTG resistance is primarily available 232 from patients in whom treatment failure was detected relatively early, which may not be the 233 case in African settings [9]. Therefore, to understand risk inherent in the emergence of DTG 234 resistance, adapting the MARISA model by extending its resistance dimension to both NRTI 235 and DTG resistances will be necessary. 236

Conclusion 237
In conclusion, our study indicates that giving access to DTG-based ART to all women not at risk 238 of pregnancy could limit the increase of NNRTI resistance, but even if all women receive DTG-239 October 22, 2019 10 . CC-BY 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19010132 doi: medRxiv preprint Anthony Hauser & al. based ART the level of NNRTI resistance will remain above 10% in South Africa. Our model 240 highlights the importance of a rapid switch of patients currently on NNRTI-based to DTG-based 241 ART in order to limit the increase in NNRTI resistance. Women who remain on NNRTI-based 242 ART will indirectly benet from a high level of DTG uptake due to a reduced risk of virologic 243 failure. . CC-BY 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19010132 doi: medRxiv preprint . CC-BY 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19010132 doi: medRxiv preprint