Estimated impact of tafenoquine for Plasmodium vivax control and elimination in Brazil: A modelling study

Background Despite recent intensification of control measures, Plasmodium vivax poses a major challenge for malaria elimination efforts. Liver-stage hypnozoite parasites that cause relapsing infections can be cleared with primaquine; however, poor treatment adherence undermines drug effectiveness. Tafenoquine, a new single-dose treatment, offers an alternative option for preventing relapses and reducing transmission. In 2018, over 237,000 cases of malaria were reported to the Brazilian health system, of which 91.5% were due to P. vivax. Methods and findings We evaluated the impact of introducing tafenoquine into case management practices on population-level transmission dynamics using a mathematical model of P. vivax transmission. The model was calibrated to reflect the transmission dynamics of P. vivax endemic settings in Brazil in 2018, informed by nationwide malaria case reporting data. Parameters for treatment pathways with chloroquine, primaquine, and tafenoquine with glucose-6-phosphate dehydrogenase deficiency (G6PDd) testing were informed by clinical trial data and the literature. We assumed 71.3% efficacy for primaquine and tafenoquine, a 66.7% adherence rate to the 7-day primaquine regimen, a mean 5.5% G6PDd prevalence, and 8.1% low metaboliser prevalence. The introduction of tafenoquine is predicted to improve effective hypnozoite clearance among P. vivax cases and reduce population-level transmission over time, with heterogeneous levels of impact across different transmission settings. According to the model, while achieving elimination in only few settings in Brazil, tafenoquine rollout in 2021 is estimated to improve the mean effective radical cure rate from 42% (95% uncertainty interval [UI] 41%–44%) to 62% (95% UI 54%–68%) among clinical cases, leading to a predicted 38% (95% UI 7%–99%) reduction in transmission and over 214,000 cumulative averted cases between 2021 and 2025. Higher impact is predicted in settings with low transmission, low pre-existing primaquine adherence, and a high proportion of cases in working-aged males. High-transmission settings with a high proportion of cases in children would benefit from a safe high-efficacy tafenoquine dose for children. Our methodological limitations include not accounting for the role of imported cases from outside the transmission setting, relying on reported clinical cases as a measurement of community-level transmission, and implementing treatment efficacy as a binary condition. Conclusions In our modelling study, we predicted that, provided there is concurrent rollout of G6PDd diagnostics, tafenoquine has the potential to reduce P. vivax transmission by improving effective radical cure through increased adherence and increased protection from new infections. While tafenoquine alone may not be sufficient for P. vivax elimination, its introduction will improve case management, prevent a substantial number of cases, and bring countries closer to achieving malaria elimination goals.


Individual-based model
Previously, we developed a mathematical model of the transmission dynamics of P. vivax in Papua New Guinea (PNG), and the potential impact of varying levels of vector control and case management to treat blood-stage infections. Fig A shows a compartmental representation of the transmission model. The model is implemented in two complementary formats: firstly, as a deterministic compartmental model described by a system of ordinary differential equations, and secondly, as an individual-based stochastic model. Notably, the two model implementations provide identical predictions for large population sizes when stochastic effects average out. The population dynamics of multiple mosquito species are described using compartmental models. Further details can be found in White et al. [1]. Infected individuals can be in one of three compartments depending on whether blood-stage parasitaemia is detectable by PCR (IPCR), light microscopy (ILM) or has high density with accompanying fever (ID). A proportion of individuals that progress to a symptomatic episode of P. vivax will undergo treatment with a blood-stage drug (T) leading to clearance of blood-stage parasitaemia and a period of prophylactic protection (P) before returning to the susceptible state (S). The superscript k denotes the number of batches of relapse causing hypnozoites in the liver. Red arrows denote new blood-stage infections arising from either new mosquito bites or relapses. Each square denotes a compartment and the circles denote the dependence of transition rates between compartments on levels of anti-parasite immunity (Ap) and levels of clinical immunity (Ac).
The model was calibrated to data from epidemiological studies involving almost 60,000 individuals in PNG. The model was further validated to data from systematic reviews of the associations between multiple metrics of malaria transmission [2,3]: P. vivax prevalence by light microscopy (PvPRLM); P. vivax prevalence by PCR (PvPRPCR); the detected incidence of clinical malaria per 1000 people per year; and the entomological inoculation rate (EIR). The model corresponded well to data from these systematic reviews of P. vivax from across the world, providing evidence that a model developed in a Papua New Guinean context, can be adapted to describe the epidemiology of P. vivax in other endemic regions such as Brazil. However, there are some important differences between the epidemiology of malaria in PNG and Brazil that need to be explicitly accounted for.

Model adaptation for Brazilian epidemiology
Malaria transmission epidemiology is extraordinarily complex, with exposure occurring in a wide range of occupational and environmental settings. Despite this complexity, malaria transmission epidemiology can be broadly categorised in two ways as follows: Peri-domestic transmission with most exposure occurring in the household and close to where people live. This type of exposure pre-dominates in much of Africa and PNG. (ii) Occupational exposure where people's work activities bring them into contact with mosquitoes in areas that may be quite far from where they live and sleep.
The relative contribution of these two modes of transmission can be assessed by examining the age and gender distribution of clinical cases of malaria. In an example of peri-domestic transmission, cases are distributed across all ages and between males and females. Notably, a large proportion of cases occur in young children. Settings with occupational exposure are typically characterised by having substantially more cases in men than in women, and more cases in working age adults (16 to 65 years of age).
The individual-based transmission model was adapted to account for these two distinct modes of transmission via the following changes: (i) Stratification into males and females. (ii) Males are further stratified according to whether they are exposed to malaria via their occupation, or are exposed to malaria solely through domestic transmission. (iii) Two separate populations of mosquitoes corresponding to peri-domestic and occupational exposure settings are simulated.
In particular we assume that 37% of men aged 16 to 65 years are at risk of occupational exposure [4]. The additional risk associated with occupational exposure is estimated based on the proportion of reported cases in adult males in a municipality, as recorded in the SIVEP database [5].

Existing case management guidelines with primaquine
In order to model the potential impact of introducing tafenoquine with testing for G6PD deficiency into cases management for P. vivax in Brazil, it is important to first assess the existing case management strategies and our model assumptions.
In Brazil, individuals with symptomatic cases of malaria are provided with free treatment at their local public health centre. Upon microscopic confirmation of P. vivax infection, patients enter the primaquine treatment pathway shown in Fig 1B. Primaquine is not prescribed to infants < 6 months of age. Primaquine is not prescribed to pregnant or lactating women because of the potential risk to their foetus or young infant. We assume all P. vivax positive cases receive chloroquine without any counter-indications. G6PD testing is not required for primaquine eligibility. Current primaquine treatment thus covers a very high proportion of clinical cases. It is assumed that primaquine is not efficacious in individuals who do not adhere to the full 7-day regimen and do not account for any dose-response. Primaquine is not assumed to be efficacious in individuals with a low CYP2D6 metaboliser phenotype. Furthermore, even in patients capable of properly metabolising primaquine, we assume it will not always be efficacious.
The Brazilian endemic states have an average G6PDd prevalence of 5.5% (from municipalityspecific estimates see Section 3.6). If we consider G6PDd cases prescribed primaquine (due to lack of testing) who did not fully adherence to the radical cure regimen due to adverse events, we estimated that on average 40.5% [95% CI 40.3%-41.0%] of cases were effectively treated. Of all P. vivax cases treated annually in Brazil, we estimate that 3.2% or 7000 cases may be at risk of developing haemolysis due to G6PDd.

Introduction of tafenoquine for case management
In order to be included for P. vivax case management, tafenoquine must be accompanied by point of care testing for G6PD deficiency. Patients with microscopically confirmed P. vivax infection are first stratified by age. Only individuals older than 16 years of age are eligible to receive tafenoquine upon initial roll-out. Pregnant or lactating women are excluded from receiving either primaquine or tafenoquine. All patients who are eligible for receiving either primaquine or tafenoquine will have their G6PD activity measured using a SD Biosensor test. Note that this even includes children <16 years of age who are not eligible for tafenoquine. Depending on their age, pregnancy/lactating status and measured G6PD activity, patients will be prescribed one of (i) chloroquine (25 mg/kg total dose over 3 days); (ii) chloroquine plus primaquine (3.5 mg/kg total dose over 7 days); (iii) chloroquine plus tafenoquine (single dose 300 mg).
Notably, tafenoquine will not completely replace primaquine, as there will be many patients not eligible for tafenoquine who will still benefit from primaquine treatment. Fig 1D shows a schematic representation of the new tafenoquine treatment pathway. The proportion of those receiving chloroquine only is expected to increase in part due to the identification of G6PD deficient individuals but also in part due to the misclassification of intermediates (30-70%) by the G6PD quantitative test. Some participants prescribed primaquine will not adhere to the full regimen. It is assumed that low CYP2D6 metabolisation does not affect tafenoquine efficacy in the standard tafenoquine introduction scenarios; however, we assess the differential impact of low CYP2D6 metabolisation on tafenoquine efficacy in Section 4.5.

G6PD diagnostics
Currently in Brazil, G6PD testing is not a requirement for individuals to receive primaquine. Once tafenoquine is introduced, we assume that quantitative G6PD testing will be implemented in all patients eligible for primaquine and tafenoquine (over 6 months old and not pregnant). An important component that should also be considered is the sensitivity and specificity of the SD Biosensor STANDARD G6PD test measuring quantitative concentration of total hemoglobin (g/dL) and G6PD enzymatic activity (U/g Hb) to be deployed in Brazil. The cut-offs for tafenoquine eligibility will requires at least 70% G6PD activity (normal G6PD activity), 30-70% (intermediate G6PD activity) for primaquine, and < 30% (deficient G6PD activity) for only chloroquine [6][7][8].
Pal et al. assessed the sensitivity and specificity of the SD Biosensor G6PD test on fresh venous blood and found a 100% sensitivity and 97% specificity in those with < 30% G6PD activity, and 95.5% sensitivity and 97% specificity in those with < 70% G6PD activity [9]. Using their data on US venous blood samples, we estimated the probability of classifying individuals in the G6PD activity groups in the model by calculating the proportion of correctly classified and misclassified G6PD activity scores (Fig B). Since African Americans were included in the study and similar G6PDd genetic variants are observed between African Americans and Brazilians [10] (see section 2.4), we assume that the SD Biosensor G6PD test will have comparable classification between the two populations. The relationship between genotypic prevalence of G6PD deficiency (as measured by allele frequency in males) and quantitative G6PD activity measured either by a reference assay or the point of care SD Biosensor is shown in Fig C. Fig C: Association between G6PD genotype and phenotype. The G6PD activity model described in Section 2.4 allows generation of estimates of G6PD activity given genotypic prevalence of G6PD deficiency (measured by allele frequency in males). This model provides estimates of the proportions in three G6PD activity categories (< 30%; 30% -70%; and > 70%) as measured by a reference assay. When tested by a SD Biosensor with imperfect accuracy, these estimates can be further adjusted as shown.

Gaussian mixture model fitting
There is a complex relationship between G6PD phenotype as measured by G6PD activity, and G6PD genotype, especially since the gene for G6PD deficiency is x-linked. We have developed a method to estimate the distribution of G6PD activity scores in the population from a large G6PD deficient prevalence survey (Table D, Fig F) for each municipality.
We define q to be the prevalence of G6PD deficiency, based on the proportion of males with a deficient genotype. Note that the x-linkage of the gene for G6PD deficiency means that males are either hemizygous deficient or hemizygous normal. Based on the Hardy-Weinberg principles of genotype frequencies in the population, we can estimate the following values: deficient males q, normal males (1 -q), homozygous deficient females q 2 , heterozygous deficient females 2q(1q), and homozygous normal females (1 -q) 2 Based on this principle, we developed Gaussian mixture models to estimate the distribution of G6PD activity scores for males and females. We assumed a normal distribution of activity scores for each genotype. The Gaussian mixture model for G6PD activity scores in males is a combination of a normal distribution of G6PD deficients q and a normal distribution of G6PD normals (1 -q). For females, the model combines a normal distribution of homozygous deficient females q 2 , a normal distribution of heterozygous deficient females 2q(1 -q), and a normal distribution of homozygous normal females (1 -q) 2 . Each normal distribution is determined by a mean μ and standard deviation σ. As a result, the following seven model parameters were estimated: q, μdef, σdef, μhet, σhet, μnor, σnor.
We collated datasets of G6PD activity scores in both men and women from publicly available data (Table A). For model fitting, four datasets of healthy individuals in Bangladesh, Cambodia, Indonesia, and Israel were standardized so that the mean activity scores of G6PD normal males was set to 10. In order to determine which males were considered G6PD "normal" we either used a visual inspection of the distribution of the data or used the cutoff proposed by the original publication.
We  We compare the distribution of the datasets used to estimate parameters for the G6PD activity model described in Section 2.4 and the model prediction for men and women for each study. G6PDd prevalence model estimates are consistent with those reported by each study (Table L).
While the G6PDd prevalence model estimates are consistent with those reported by each study, G6PD mutations in these populations are considered more severe than the prevalent G6PD A(−) variant (202 G>A mutations) commonly found in South America [12][13][14]. The common variant forms in the datasets: Bangladesh (Orissa, Kalyan-Kerala and Mahidol); Cambodia (Viangchan); Indonesia (Vanua Lava); Israel (Mediterranean) tend to be more severe and as a result, the fitted model may overestimate the G6PD activity in the Brazilian population [15,16].
In particular, the prediction of heterozygous women under the 70% threshold in Brazil may be overestimated in comparison to the data used to fit the model. As a result, more women could in fact be eligible for tafenoquine treatment than predicted by the model. Overall, since this concerns a small proportion of the overall population, this will not have a major impact on our prediction of tafenoquine impact on transmission; however, further optimization of the model will be required using Brazilian data when it becomes available.

Parameter estimates
A systematic literature review was conducted to identify appropriate model parameters estimates for the Brazilian setting. In particular, we identified population factors, treatment regimen and efficacy, G6PD diagnostic performance, and data on low CYP2D6 relevant for Brazil. The summary of the review is shown in Table A.

Demography
Demographic data was obtained from the Brazilian census collated by Instituto Brasileiro de Geografia e Estatística (IBGE). An overview of the Brazilian population is shown in Fig E. The last census was undertaken in 2010, providing age and gender stratified population estimates for every municipality, plus data on a large number of other covariates. Municipal population size and average age was extracted from IGBE reported 2018 estimates.

Relapses
Brazil is typically characterised as having the tropical relapse phenotype of P. vivax (Battle 2014; Lover 2013) [28,29]. However, it is challenging to accurately study the epidemiology of P. vivax relapses in Brazil, because radical cure with primaquine is routinely administered as part of case management. As such, many epidemiological studies reporting recurrent P. vivax episodes may not capture natural relapses patterns, but relapse patterns following some degree of primaquine treatment. Despite the challenges in identifying which recurrences are relapses, and the potential for partially efficacious primaquine treatment to prevent some relapses, there are very good sources of data on recurrences from the SIVEP databases. Based on analysis of data from 26,295 recurrences Daher et al. estimated a median time between recurrences of 69 days [30]. The epidemiology of P. vivax relapses can be described via three parameters as shown in Table B [31].

Entomological parameters
There are a number of species of Anopheles mosquito that contribute to malaria transmission in Brazil, most notably An. darlingi, An. albitarsis, An. nuneztovari, An. aquasalis, and An. braziliensis [33][34][35][36]. An. darlingi is the most widespread vector and makes the largest contributes to transmission. The mosquito component of the model is calibrated using data from An. darlingi in Brazil on mosquito life expectancy, duration of gonotrophic cycle, and the duration of the sporogonic cycle [37]. The assumed parameter values are presented in Table C. Note that since we are not analysing the impact of vector control interventions, model simulations will not be sensitive to the exact choice of entomological parameters.

Primaquine adherence
We conducted a literature review for publications measuring adherence to 7 days of prescribed low dose primaquine (0.5 mg/kg/day) among P. vivax cases in South American settings. We identified three studies in Brazil [19,20,38] one study in Peru [21] and one study in Ecuador [39] published between 2000 and 2015. Two studies including estimates of self-reported adherence to treatment for both P. falciparum and P. vivax infections in Mato Grosso, Brazil of 83.8% and 77.8% were excluded [40,41]. The findings of the included studies are summarized in Table M.
Three studies assessed both pill-count and self-reported questionnaires and estimated adherence to primaquine at day 7 at 62.2%, 86.4%, and 66.7%. Only the last and most recent study by Almeida et al. assessed the sensitivity and specificity of their questionnaire and was able to estimate an accuracy of 97.1% on the Likert scale questionnaire based on responses to pillcount. Therefore for our model parameters, we assume a 66.7% adherence to 7-day primaquine in Brazil based on this study [19].

Low CYP2D6 metaboliser prevalence
In 2013, Bennett et al. showed a significant association between low CYP2D6 metaboliser activity and primaquine treatment failure [23]. Therefore it is important to evaluate the prevalence of CYP2D6 metaboliser activity, in addition to adherence, when assessing primaquine efficacy. We have compiled relevant studies estimating low CYP2D6 prevalence in Table N Brazilian study also found a similar prevalence of AS < 1 at 9.6% among healthy adults [46]. When considering CYP2D6 prevalence among P. vivax cases in the Brazilian population, one estimate found a prevalence 19.5% of AS ≤ 1 [27]. Due to varying levels of uncertainty on the varying definition of low metabolisation, we assumed a low CYP2D6 prevalence of 8.1% for our baseline scenarios and tested the sensitivity of a low (4%) or high (20%) prevalence on results in Section 4.

Impact of CYP2D6 activity score and P. vivax malaria recurrence
When Bennett et al. first shed light on the potential impact of CYP2D6 activity scores and P. vivax malaria recurrence due to low CYP2D6 metaboliser enzyme for high-dose primaquine (30 mg/day for 14 days) to effectively clear hypnozoites in the liver, their results suggested that for AS < 1 individuals, primaquine was not metabolised effectively [23]. Although a significant association was found, the results have been contested to also include risk among those with an AS = 1. Baird [25].
Impaired CYP2D6 alleles occurred in 95% of therapeutic failures following directly supervised, high-dose primaquine of good quality in subjects followed for a year where reinfection was highly unlikely. Those failures occurred at a rate of 15% among several hundred such treatments [42,43].
Concerning the impact of CYP2D6 metabolic activity on tafenoquine efficacy, the two previously mentioned randomized multi-center trials did not detect an association with poor or intermediate CYP2D6 metabolisers and clinical relapse for both low-dose tafenoquine (300 mg/day for 1 day) and high-dose tafenoquine (300 mg/day for 2 days) [24,47]. On the other hand, one study conducted in mice found that tafenoquine metabolism might not be CYP2D6 dependent [48]. While the surveyed data are certainly consistent with the efficacy of tafenoquine not being affected by low CYP2D6 metabolisation, we do not consider this to be a very high standard of evidence due to the potential for confounding. Therefore, we also implement sensitivity analyses where this assumption is varied.
The association between CYP2D6 activity and the efficacy of 8-aminoquinolines against P. vivax hypnozoites is undoubtedly a complex one, with multiple sources of evidence combining to suggest a non-linear dose-response relationship between activity and efficacy. In our main analysis, we make the simplifying assumption that primaquine is not efficacious in individuals with CYP2D6 activity score (AS) < 1, and that primaquine has 71.3% efficacy in individuals with AS ≥ 1. We assume an 8.1% prevalence rate of poor CYP2D6 metabolisers in the Brazilian population. We assume that the efficacy of tafenoquine is not affected by CYP2D6 activity score. Due to the uncertainty of these assumptions, we also consider the impact of varying levels of low CYP2D6 on both primaquine and tafenoquine in extended analyses in Section 4.

G6PD deficiency prevalence in Brazil
A systematic review reported an average 5.2% G6PD deficiency in Brazil of 9 different states from studies conducted in 2016 and before (

Chloroquine resistance and efficacy in clearing blood-stage parasites
Chloroquine resistance has been observed in Brazil. A study assessed resistant parasites to chloroquine mono-therapy in P. vivax positive volunteers in Manaus, Brazil, and found that 10.1% of the subjects were confirmed with therapeutic failure at day 28 (P. vivax chloroquine resistant) [17]. In another study, the standard treatment regimen of 3-day chloroquine with 7-day primaquine found resistant parasites at day 28 in a total of 5.2% study participants [18]. Due to these differing observations in resistance to chloroquine in Brazil, we will assume 89.9% chloroquine treatment efficacy in clearing blood-stage parasites in patients who are not prescribed primaquine. In patients who are prescribed primaquine, we assume 94.8% efficacy of chloroquine treatment against blood-stage parasites. Note that in patients who are prescribed primaquine but do not adhere to the full 7 day regimen, we also assume 94.8% chloroquine efficacy. In patients that are prescribed both chloroquine and tafenoquine, we assume 100% efficacy against blood-stage parasites due to the very long half-life of tafenoquine.
Extended results assessing a reduction in the prophylactic period of chloroquine are shown in Section 4.10.

Primaquine efficacy
In order to estimate efficacy of 7-day low dose primaquine treatment in Brazil (3.5 mg/kg total dose), we referred to Brazilian studies of travellers from endemic to non-endemic regions. One study found that 39.6% of treated cases returning from malaria endemic regions relapsed (Pedro 2012) [22]. If we consider only cases who travelled to the Brazilian Amazon Region (n=38), 14 of them relapsed (36.8%). We assume that out of these relapses, 8.1% failed due to due to low CYP2D6 metabolisers since they were not assessed in the study. Therefore, we assume that lowdose 7-day primaquine efficacy in Brazil is 71.3% (1 -0.368 + 0.081). This value is also consistent with the proportion of patients free from recurrence seen in clinical trials of 69.6% [24]. In an additional scenario, we assume the potential introduction of a higher efficacy primaquine regimen with 90% efficacy similar to those observed in studies prescribing 5 mg/kg total dose over a course of 14 days [49].

Tafenoquine efficacy
Phase 3 clinical trials showed comparable efficacy between primaquine and tafenoquine. In patients administered single dose 300 mg tafenoquine, 62.4% (95% CI; 54.9% -69.0%) remained free from recurrence after 6 months. In patients administered primaquine (15 mg (equivalent to 0.25 mg/kg) daily for 14 days), 69.6% (95% CI; 60.2% -77.1%) remained free from recurrence after 6 months [24]. This estimate is substantially lower than the DETECTIVE trial of 300mg tafenoquine which reported efficacy against recurrences in Brazil of 89.2% (95% CI; 77% -95%) [47]. Based on the phase 3 trial data, we assume that primaquine and tafenoquine have equal efficacy. As the efficacy against recurrence at 6 months does not necessarily equal the efficacy against relapses (due to the possibility of reinfection via mosquitoes), we use the value of 71.3% from the Pedro et al. study [22]. We also simulated a hypothetical scenario where tafenoquine could be safely given to reach 90% efficacy similar to what was observed in the DETECTIVE trial for a 600 mg dose [50].

Weekly primaquine
In Brazil, pregnant or lactating women are not prescribed the standard 7-day primaquine regimen. Instead they are prescribed weekly low dose primaquine (0.5 mg/kg) for 8 weeks. Very few studies have assessed the efficacy of this strategy. Only one study found 86.5% efficacy of 0.75 mg/kg primaquine/week for 8 weeks in Pakistan [51]. Furthermore, we were unable to find any data on adherence to this 8-week regimen. We therefore assumed that weekly low dose primaquine for 8 weeks was not efficacious.

Radical cure estimates
Based on 2018 case reports, we estimated the number of cases of effective radical cure with the current chloroquine and primaquine treatment per municipality without G6PD testing (  If cases in 2018 were treated with tafenoquine and G6PD testing was introduced (S1), we would expect a 59.0% effective radical cure (range [26.6%, 61.9%]) with an estimated 127,100 treated cases (Table E, Fig H). An additional 35,000 cases could be treated (Fig I).

Asymptomatic infections
The proportion of asymptomatic infections increases with higher transmission intensity. As transmission intensity increases, a higher proportion of sub-patent cases are detected by PCR. These lower levels of parasitemia can go undetected when light microscopy (LM) is the main tool used to detect cases for treatment. As higher intensity of P. vivax exposure occurs over time, the population develops levels of immunity that can suppress higher parasitemia all while transmission is still ongoing. These trends explain the differences in the simulation results across different settings such as Itaituba, Pará with an annual parasite incidence (API) or 23 cases per 100 and São Gabriel da Cachoeira, Amazonas, with an API of 267 (Fig J). In lower transmission settings, a higher proportion of cases have clinical disease who are more likely to be detected and treated by a high coverage health care system like that of Brazil (e.g. 18%, Fig J A). As more cases enter the treatment pathway, the effect of radical cure on onwards transmission will be more apparent. As a result, in moderate and high transmission settings, a smaller proportion of cases will be treated out of those who are transmitting P. vivax (Fig J B-C). Even if tafenoquine introduction into case management greatly improves the rate of effectively treated cases, the effect size is smaller because only a small percent of cases (i.e. < 15%) are being treated. We observed that in a low transmission setting like Itaituba, Pará, 18% of cases have clinical disease and tafenoquine introduction (S1) would have a more significant impact on transmission reducing incidence from 23 cases to 8.2 cases per 1000 (65% reduction). However, as transmission intensity increases in settings like peri-urban Manaus, Amazonas less than 15% of cases have clinical symptoms and more than 50% have very low parasitemia. In São Gabriel da Cachoeira, where very few cases may be captured by the healthcare system, tafenoquine introduction would only impact transmission in less than 10% of the population and even less because the majority of cases are in children. However, it is important to note that even such small reductions in transmission would result in an important reduction of clinical cases in high transmission settings there is a lower effective size. In the intervention strategies considered, the number of courses prescribed is impacted by both transmission intensity, case eligibility, and the impact of different intervention strategies on transmission. Notably, when we compare S1 and S3, a higher efficacy tafenoquine scenario would lead to fewer G6PD tests and fewer courses of tafenoquine being prescribed. This is because the higher efficacy regimen is predicted to lower P. vivax transmission more in the community. In some cases, this phenomenon also extends to the situation where children are prescribed tafenoquine (S2). For example, if tafenoquine can be safely prescribed to children >2 years of age, initially more G6PD tests and more courses will be required. However, if expanded case management leads to a reduction in population-level P. vivax transmission over the long term, then fewer courses will need to be prescribed. This is also the case in settings with low preexisting primaquine adherence that would see the most significant impact with tafenoquine introduction and prescribe the least amount of 8-aminoquinolines overall (S6).

Cases averted, treatment courses and G6PD tests in archetype settings
Treatment courses and G6PD tests for archetype settings are provided in Table G.

vivax clinical cases averted 5-years post-tafenoquine introduction in 126 simulated settings (excluding Manaus, including per-urban Manaus) and 298 non-simulated settings with 1 to 99 annual cases reported in 2018.
For simulated setting: assuming incidence and population size calibrated to 2018 data, the total annual clinical cases for each scenario were compared to a baseline scenario with no tafenoquine introduction. We reported the cases averted as the difference in the total cumulative annual cases from 2021 to 2025 in the baseline scenario compared to the tafenoquine introduction scenarios 1-6 over this period. For non-simulated setting: assuming no change in annual reported P. vivax cases across all years and the impact as the median estimated effect size in 2025 per scenario. All values are rounded to the nearest 100.  G6PDd in 2021 in Itaituba, peri-urban Manaus, and São Gabriel da Cachoeira respectively. Incidence represents the average of 100 independent simulations per scenario with a moving-average smoothing of the data. Incidence was calibrated to clinical cases reported in 2018 per 1000 population. (B) (D) (F) The effect size (in %) is reported as the percentage reduction in incidence after 5 years (incidence in 2020 compared to 2025) for S1-4 in Itaituba, peri-urban Manaus, and São Gabriel da Cachoeira respectively. (G) The 5-year post-intervention effect size in all simulate settings. A Loess fitted line and 95% confidence interval bands are shown. For municipalities with an incidence below 25 cases per 1000 population, stochastic fade out in the absence of importation prevents accurate assessment of impact.

Effect size over time for S1
For our main results, we chose to report the effect size of tafenoquine introduction into case management practices 5-years post-introduction. However, the effect size can be assessed at any point. We show how the percent change intromission across the simulated municipalities under Scenario 1 varies over time (Fig L).  Table I). An additional median 19.9% and 13.6% gain in impact of transmission is expected in the second year respectively. Impact is not expected to change in settings with very high incidence over time (> 200 cases per 1000 population). Over time, settings will reach a new equilibrium 10-years post-introduction with only small gain in the last 5 years.

Effect size summary table for all scenarios 1-10
Refer to section 4.8 and 4.9 for a description of scenarios 7-10.
In S1 we would expect a median 37.5% impact on transmission in all simulated municipalities after 5 years as compared to 18.2% in settings with an incidence > 25 where we have more certainty (Table J). Additional impact can be achieved with the introduction of a higher efficacy dose of tafenoquine safely in adults (S2), in children (S3, S4), and improving primaquine efficacy (S7). Several factors may also give tafenoquine an additional advantage such as low pre-existing primaquine adherence (S6) and significant low CYP2D5 metabolisation in only primaquine (S10).

Varying assumption of 8-aminoquinoline efficacy
The efficacy of 8-aminoquinolines varies across populations and regimens. In Brazil, we estimated the efficacy of 7-day 3.5mg/kg total dose of primaquine at 71.3% (Section 3.9). Since the regimen for tafenoquine to be introduced in Brazil is a single 300mg dose that administered in clinical trials and shown to have similar efficacy, we assumed that for our standard tafenoquine scenario (S1), tafenoquine would have equivalent efficacy at 71.3% (Section 3.10).
Based on such data, the main advantage of tafenoquine as compared to primaquine is reducing low adherence. In a hypothetical scenario where a high efficacy tafenoquine dose can be safely administered (S2), we would expect an additional impact on transmission of 14.0% as compared to S1 (Fig M). In addition, in South East Asian settings, a 14-day 5 mg/kg primaquine dosing regimen has been shown to have close to 90% efficacy (Section 3.9). In a hypothetical scenario where a similar primaquine regimen is introduced in Brazil along with a higher tafenoquine regimen (S7), we would expect some additional impact. Scenario 7 could lead to a median impact of 55.5% as compared to S2 of 51.5%. In particular, such a regimen would be beneficial to higher transmission settings that have more cases in children and as a result, a higher proportion of cases prescribed primaquine (Fig M C-F). In lower transmission settings or occupational settings, most cases would already be eligible for tafenoquine and very few cases would be impact by an increase in primaquine efficacy (Fig M A-B). The effect size (in %) is reported as the percentage reduction in incidence after 5 years (incidence in 2020 compared to 2025) for S1-4 in Itaituba, peri-urban Manaus, and São Gabriel da Cachoeira respectively. (G) The 5-year post-intervention effect size in all simulate settings. A Loess fitted line and 95% confidence interval bands are shown. For municipalities with an incidence below 25 cases per 1000 population, stochastic fade out in the absence of importation prevents accurate assessment of impact.

Varying assumption on low CYP2D6 metabolisation
As previously described in Sections 3.5 and 3.6, there is a high level of uncertainty on the impact of low CYP2D6 metabolisation on 8-aminoquinoline activity and efficacy. The first uncertainty due to lack of sufficient data, is the impact of low CYP2D6 metabolisation of tafenoquine. In S1, we assumed an impact only on primaquine; however, in another scenario (S8), assuming that similar mechanisms of primaquine activation also apply to the drug family of 8-aminoquinolines including tafenoquine. We tested the impact of S8 with an assumed 8.1% low CYP2D6 prevalence on transmission across various settings in Brazil (Fig N). In such a scenario, we would expect the impact of tafenoquine on transmission to reduce by 6.9% (from a median of 37.5% to 28.6%) across all municipalities as compared to S1. In settings with a high burden of disease, this could have significant impact. However, in higher transmission settings where a lower proportion of cases receive tafenoquine (i.e. most cases in children), we would not expect a significant difference (Figs N E-F).
A second level of uncertainty also arises in the classification of low CYP2D6 metabolisation and as a result the prevalence estimate of low metabolisers in Brazil. We assumed a prevalence of 8.1% in S1 and tested the potential parameter uncertainty by simulating scenarios of low CYP2D6 prevalence at 4% (S9) and at 20% (S10). We estimated that a prevalence of 4% with effect only in primaquine would reduce the impact of tafenoquine on transmission 2.2% but generally it would have very little impact on incidence as compared to S1. As the prevalence of low CYP2D6 increases to 20% as shown in some studies in Brazil (Section 3.5, 3.6), we would expect a slightly higher impact for tafenoquine introduction (potential additional increase in impact of 10.6%). If low CYP2D6 does not have an impact of tafenoquine, tafenoquine would be more advantageous in S10 than S1 as compared to baseline; however, this relies on a major assumption that CYP2D6 metabolisation is not an important factor for tafenoquine efficacy, which would overall reduce the impact of any tafenoquine rollout scenario (S8). The effect size (in %) is reported as the percentage reduction in incidence after 5 years (incidence in 2020 compared to 2025) for S1-4 in Itaituba, peri-urban Manaus, and São Gabriel da Cachoeira respectively. (G) The 5-year post-intervention effect size in all simulate settings. A Loess fitted line and 95% confidence interval bands are shown. For municipalities with an incidence below 25 cases per 1000 population, stochastic fade out in the absence of importation prevents accurate assessment of impact.

Variation in blood-stage prophylactic duration
In our model, we assumed that the duration of prophylaxis protecting from new infectious bites in the Brazilian population was 28 days with the 3-day regimen of chloroquine. We assumed little to no chloroquine resistance across all modelled settings. We assessed the impact of chloroquine failure before 28 days set to 14 days of post-treatment prophylaxis with chloroquine across all scenarios (Table K). On average, due to the reduction in the post-blood-stage treatment prophylactic period in these scenarios, we observe a 4-5% increase in the impact of tafenoquine on transmission.