Descriptive statistics and key epidemiological differences between trialsA summary (study profile) of the relevant trials included in the WWARN systematic review will be presented to highlight potential selection bias. A summary of the included studies will be presented, including (but not restricted to) treatment tested, inclusion and exclusion criteria, follow up duration, study populations, parasitaemia sampling scheme, treatment of recurrent asymptomatic parasitaemia or not, treatment regimen for recurrent parasitaemia and description of location by country, transmission site(s), regional relapse periodicity and chloroquine resistance. Tests of statistical significance will not be undertaken for baseline characteristics; rather the clinical importance of any differences in the baseline distributions will be noted. We will focus our attention on known biological or epidemiological differences between study sites, using external information such as estimated entomological inoculation rate (EIR), and known vivax relapse peridoicity for the included study sites.Baseline characteristics of patientsA summary of relevant baseline patient characteristics will be presented including age, gender, treatment given, treatment supervision for primaquine arms, G6PD status, baseline asexual parasitaemia, gametocytaemia, temperature >37.5C, prior antimalarial use, prior malaria history. Summary statistics will be broken down by gender and age category. The distribution of continuous variables (e.g. mg/kg total drug dose for each dosing group) will be described using the mean and standard deviation if the data are normally distributed, geometric mean and 95% reference range if the data are normally distributed following a log transformation, or the median and interquartile range if the data are non-normally distributed.Estimating parasite clearance after tafenoquine monotherapy in symptomatic P. vivaxThe first stage of model fitting will be a model of parasite clearance (expressed as a half-life: time to reduce parasitaemia by one half). This will be a Bayesian linear mixed effects (hierarchical) model with the dependent (response) variable the log-parasitaemia, and the independent variables the time since start of treatment and the tafenoquine blood stage concentration (on the log scale), allowing for inter-individual variation in baseline parasitaemia and clearance rate, and study random effects for both the intercept and slope. We will formally test whether increases in peak blood tafenoquine concentrations are predictive of faster parasite clearance, with the a-priori hypothesis that all measured blood concentrations in first days following drug administration are above the concentration providing 99% of Emax. The main output of this analysis will be a distribution of plausible Emax values (Bayesian posterior distribution).Estimating the MIC of tafenoquineWe have previously developed a model framework to estimate minimum inhibitory concentrations of slowly eliminated antimalarial drugs from time to recurrent infection (applicable to falciparum and vivax malaria) [6]. This was fitted to clinical trial data from Southeast Asia to estimate the concentration-effect curve of chloroquine in against blood stages of P. vivax. The model framework uses approximate Bayesian computation to infer the unknown parameters from the asexual stage concentration-response curve. This necessitates a pharmacometric model of the drug of interest and requires individual drug measurements in the terminal phase of elimination (e.g. day 7 and later for chloroquine and tafenoquine) to estimate concentration-time curves during the critical period when parasite growth is delayed.Tafenoquine MIC estimation has an additional complexity: all participants in the phase 2 & 3 field trials received tafenoquine and chloroquine in combination. Chloroquine is also slowly eliminated and the terminal elimination half-life is longer than for tafenoquine (1 month versus 3 weeks). However, chloroquine has bi-phasic elimination and our previous modelling suggests that the MIC is reached on average 20 days after the standard malaria regimen dosing. It is likely, but remains to be demonstrated, that tafenoquine reaches its MIC two weeks later (the MIC of tafenoquine has been suggested to be approximately 40 ng/mL [4] which is reached on average 35 days after a single 300 mg dose). If this is the case, then it will be possible to differentiate the blood-stage effects of the two drugs using data from chloroquine alone versus chloroquine plus tafenoquine. We note that the chloroquine MIC may be lower in Latin America relative to Southeast Asia due to different resistance patterns. The analysis will be as follows:• We will fit a chloroquine population pharmacokinetic model to all trial participants (all will have been treated with chloroquine at the standard mg/kg dose); fit a tafenoquine population model to the subset of participants treated with tafenoquine (varying mg/kg doses). Output a set of predicted time-concentration curves for both drugs in all individuals.• Use approximate Bayesian computation (ABC) under a mechanistic model PK-PD model of parasite growth in presence of active drug to estimate the MIC and slope of the dose-response curve, allowing for site specific random effects. The model has to make assumptions about timing of merozoite maturation, which in previous work was derived from recurrence data after rapidly eliminated drugs (7 days of artemisinin monotherapy).Mechanistic model of methaemoglobin production8-aminoquinoline drugs increase the production of methaemoglobin. An exploratory analysis in 585 patients with acute vivax malaria who were given high-dose primaquine with either chloroquine or dihydroartemisinin-piperaquine suggested that increased levels of methaemoglobin are associated with a lower hazard ratio for the time to first recurrence. Loss of function CYP2D6 polymorphisms known to be associated with reduced radical curative efficacy were also associated with lower levels of methaemoglobin. This suggests that methaemoglobin production measures oxidative activity which is also responsible for antimalarial activity, and therefore correlates with therapeutic responses to 8-aminoquinolines. We want to test this hypothesis by pooling all large studies that have serial measurements of methaemoglobin. We will estimate individual areas under the methaemoglobin curve by fitting a simple two-parameter model of methaemoglobin production as a function of 8-aminoquinoline mg/kg dose.Probabilistic classification of recurrent infectionWe have previously developed a Bayesian model framework for the probabilistic classification of recurrent P. vivax episodes using time-to-recurrence and paired parasite genetic data (initial and recurrent episodes) [7]. The genetic component of the model uses the set of expected relatedness values (as given by expected identity by descent) between all combinations of haploid genomes across infections within a single host to calculate the evidence that the recurrent infection is caused by relapsing hypnozoites versus reinfection versus recrudescence. The published model is currently applicable to microsatellite data (or any small set of polyallelic markers - computational complexity is a function of the number of markers and the complexity of infections), but can be extended to deal with richer genetic data such as whole genome sequencing data. The time-to-event component of the model is a parametric mixture model of `competing risks' (i.e. competing reasons for recurrent infection), whereby the underlying hazard functions are given explicit parametric forms. In particular, relapse is parameterised as a mixture of rapid (or periodic, Weibull distribution) and slow (sporadic, exponential distribution) time to event functions. We will adapt the model framework for the data at hand. Important modifications include:1. The parametric choices and prior distributions for the time-to-reinfection and the time-to-relapse will be derived by forward simulation using the fitted PK-PD blood stage models from the previous steps in the analysis. This means that we can use information directly regarding the blood stage activity of tafenoquine.2. Stratification by site (and country) is needed, both for the genetic model (allele distributions) and the time to event model.Predictors of 8-aminoquinoline efficacyOnce we have a probabilistic classification for each observed recurrence, e.g. Prob(relapse), it is then possible to estimate treatment effects directly from randomised comparisons. For example, Prob(relapse | tafenoquine) - Prob(relapse | placebo). We will fit a hierarchical model to allow for differences in treatment effects across sites and clinical studies, as we expect there is treatment effect modification (for example, the latent hypnozoite load is likely to be proportional to transmission intensity). The main treatment effect of interest is the relative risk reduction: [Prob(relapse | tafenoquine) - Prob(relapse | placebo)] / Prob(relapse | placebo).In addition, we wish to re-explore the relevance and utility of methaemoglobin increases as a pharmacodynamic surrogate of 8-aminoquinoline antimalarial activity. This will require a two-stage approach. First, fitting a simple 1-compartment dynamic model to methaemoglobin concentrations to derive production and clearance estimates. Second, using the estimated production rates as predictors of failure (where failure is the probability of relapse as in [7]).Characterising optimal dosingUnder a simple pharmacodynamic model of parasite growth, assuming that latent hypnozoite burdens are correlated with transmission intensity, we can forward simulate predicted efficacy of different dose-regimens using either primaquine or tafenoquine. These results can then be fed into a formal risk-benefit analysis comparing tafenoquine and primaquine for the radical cure of P. vivax malaria, updating existing models [8].1. Lacerda MVG, Llanos-Cuentas A, Krudsood S, Lon C, Saunders DL, Mohammed R, et al. Single-Dose Tafenoquine to Prevent Relapse of Plasmodium vivax Malaria. New England Journal of Medicine. 2019;380: 215-228. doi:10.1056/NEJMoa17107752. Llanos-Cuentas A, Lacerda MVG, Hien TT, VĂ©lez ID, Namaik-larp C, Chu CS, et al. Tafenoquine versus Primaquine to Prevent Relapse of Plasmodium vivax Malaria. New England Journal of Medicine. 2019;380: 229-241. doi:10.1056/NEJMoa18025373. Fukuda MM, Krudsood S, Mohamed K, Green JA, Warrasak S, Noedl H, et al. A randomized, double-blind, active-control trial to evaluate the efficacy and safety of a three day course of tafenoquine monotherapy for the treatment of Plasmodium vivax malaria. PLoS One. 2017;12: e0187376. doi:10.1371/journal.pone.01873764. Dow G, Smith B. The blood schizonticidal activity of tafenoquine makes an essential contribution to its prophylactic efficacy in nonimmune subjects at the intended dose (200 mg). Malaria Journal. 2017;16: 209. doi:10.1186/s12936-017-1862-45. CDC - Malaria - New antimalarial tafenoquine. 24 Feb 2020 [cited 12 Jan 2021]. Available:

https://www.cdc.gov/malaria/new_info/2020/tafenoquine_2020.html&;#13; 6. Watson J, Chu CS, Tarning J, White NJ. Characterizing Blood-Stage Antimalarial Drug MIC Values In Vivo Using Reinfection Patterns. Antimicrobial Agents and Chemotherapy. 2018;62. doi:10.1128/AAC.02476-177. Taylor AR, Watson JA, Chu CS, Puaprasert K, Duanguppama J, Day NPJ, et al. Resolving the cause of recurrent Plasmodium vivax malaria probabilistically. Nature Communications. 2019;10: 5595. doi:10.1038/s41467-019-13412-x8. Nekkab N, Lana R, Lacerda MVG, Obadia T, Siquiera A, Monteiro WM, et al. Tafenoquine for Plasmodium vivax control and elimination: a modelling case study of Brazil. PLOS Medicine (in press). 2021.