Estimating the changing prevalence of molecular markers of artemisinin partial resistance in <i>Plasmodium falciparum</i> malaria in Sub-Saharan Africa

Abstract Background Artemisinin-based combination therapies (ACTs) are the most widely used treatment for Plasmodium falciparum malaria. Kelch 13 mutations associated with artemisinin partial resistance (ART-R) have emerged in Sub-Saharan Africa (SSA) and are now reported in an increasing number of countries. ACT treatment failure rates are at risk of unprecedented increase. To summarise existing surveillance data and guide future surveillance, we produce modelled estimates of the spatiotemporal distributions of Kelch 13 and partner drug marker prevalence in SSA. Methods We develop and validate spatiotemporal statistical models, fitted within a Bayesian framework, given molecular surveillance data. We estimate the prevalence of Kelch 13 mutations that are validated or candidate markers of ART-R and the prevalence of the mutations Pfcrt -K76T, Pfmdr1 -N84Y, Pfmdr1 -Y186F, and Pfmdr1 -D1246Y, associated with selection by pressure from the ACT partner drugs amodiaquine and lumefantrine. Findings Our models reflect all existing clusters of ART-R-associated Kelch 13 mutations. We estimate the prevalence of these Kelch 13 mutations to be greater than 10% in 23% of the area of endemic malaria transmission in SSA in 2026. We also estimate that 5.8% of malaria cases in 2026 will be affected by a validated or a candidate ART-R marker. Our estimates of the prevalence of Pfcrt -K76T and other partner drug markers reflect sustained pressure from artemether-lumefantrine: we estimate the median prevalence of Pfcrt -76T across SSA to be 19% in 2026. Interpretation Our models allow readers to visualise variation in observed mutation prevalences and to extrapolate prevalence to regions in space and time that are not represented in surveillance data. To monitor the changing distribution of antimalarial resistance markers within the constraints of the current global health funding climate it is critical that validated, statistical frameworks are incorporated into decision-making workflows to make the best use of molecular surveillance data. Funding This research was funded by the European Union under the Global Health EDCTP3 Joint Undertaking (grant agreement 101103076) and the Australian National Health and Medical Research Council (APP2019093). Research in context Evidence before this study Ongoing systematic reviews of molecular surveillance of antimalarial resistance markers collate evidence of changing prevalences of Kelch 13 mutations associated with artemisinin partial resistance. However, there is a high degree of sampling bias in this data, and there are regions where limited surveillance has been carried out. We searched PubMed with the search terms: (((spatial OR spatiotemporal) AND (artemisinin OR Kelch)) AND (Africa)) AND (model* OR map OR mapping) which returned 30 results. We identified one recent pre-print describing spatiotemporal models of molecular markers of ART-R and partner drug resistance, however these models were not formally validated and model uncertainty may have been under-estimated. Added value of this study We use spatiotemporal statistical models to estimate resistance marker prevalence in regions where there has been no molecular surveillance. Our models’ predictions are contextualised by estimates of model uncertainty, and we validate our modelling framework through posterior predictive checks and by evaluating its predictive performance on held-out data. Implications of all available evidence Kelch 13 mutation prevalences are rising in all existing clusters where mutations have been identified, including in southern Africa. We estimate elevated prevalences in regions that neighbour existing clusters that are not well-represented in our surveillance dataset.