Leveraging mathematical models of disease dynamics and machine learning to improve development of novel malaria interventions

Abstract

Background Substantial research is underway to develop next-generation interventions that address current malaria control challenges. As there is limited testing in their early development, it is difficult to predefine intervention properties such as efficacy that achieve target health goals, and therefore challenging to prioritize selection of novel candidate interventions. Here, we present a quantitative approach to guide intervention development using mathematical models of malaria dynamics coupled with machine learning. Our analysis identifies requirements of efficacy, coverage, and duration of effect for five novel malaria interventions to achieve targeted reductions in malaria prevalence. Methods A mathematical model of malaria transmission dynamics is used to simulate deployment and predict potential impact of new malaria interventions by considering operational, health-system, population, and disease characteristics. Our method relies on consultation with product development stakeholders to define the putative space of novel intervention specifications. We couple the disease model with machine learning to search this multi-dimensional space and efficiently identify optimal intervention properties that achieve specified health goals. Results We apply our approach to five malaria interventions under development. Aiming for malaria prevalence reduction, we identify and quantify key determinants of intervention impact along with their minimal properties required to achieve the desired health goals. While coverage is generally identified as the largest driver of impact, higher efficacy, longer protection duration or multiple deployments per year are needed to increase prevalence reduction. We show that interventions on multiple parasite or vector targets, as well as combinations the new interventions with drug treatment, lead to significant burden reductions and lower efficacy or duration requirements. Conclusions Our approach uses disease dynamic models and machine learning to support decision-making and resource investment, facilitating development of new malaria interventions. By evaluating the intervention capabilities in relation to the targeted health goal, our analysis allows prioritization of interventions and of their specifications from an early stage in development, and subsequent investments to be channeled cost-effectively towards impact maximization. This study highlights the role of mathematical models to support intervention development. Although we focus on five malaria interventions, the analysis is generalizable to other new malaria interventions. Graphical abstract

Design and selection of drug properties to increase the public health impact of next-generation seasonal malaria chemoprevention: a modelling study

“…Our analysis then applied a mathematical framework for predicting determinants of intervention impact and defining minimum drug profile criteria. This framework has been previously described, 28 and was demonstrated in a proof-of-concept study 27 (as summarised in appendix 1.4). In brief, we used the individual-based malaria transmission model to simulate scenarios over a wide range of input values for intervention coverage and drug initial efficacy and duration of protection.…”

The influence of biological, epidemiological, and treatment factors on the establishment and spread of drug-resistant Plasmodium falciparum

“…Therefore, we trained an HGP ( Binois and Gramacy, 2021 ) on a limited set of OpenMalaria simulations (3500–11,500 simulations). We then used the trained emulator to predict the output of OpenMalaria for a large number of simulations and used these outputs to perform the global sensitivity analysis ( Figure 1C ), adapting a similar approach to Golumbeanu et al, 2022 and Reiker et al, 2021 . Our approach involved: (i) randomly sampling combinations of parameters; (ii) simulating and estimating the rate of spread of the resistant genotype for each parameter combination in OpenMalaria; (iii) training an HGP to learn the relationship between the input (for the different drivers) and output (the rate of spread) with iterative improvements to fitting through adaptive sampling; and (iv) performing a global sensitivity analysis based on the Sobol variance decomposition using the trained emulator ( Kilian et al, 2000 ).…”

“…We trained the HGP on the training dataset using the function mleHetGPfrom the R package ‘hetGP’ ( Binois and Gramacy, 2021 ). We chose to use HGP as it was successfully used in two previous studies that performed global sensitivity analyses of OpenMalaria ( Reiker et al, 2021 ; Golumbeanu et al, 2022 ). In addition, Reiker et al, 2021 tested different emulators and found that HGP provided the best fit with a limited number of simulations (analysis not shown in the published study).…”

Modelling to inform next-generation medical interventions for malaria prevention and treatment

“…Modelling captures these dynamics to link individual and community benefits, allowing clinical trial evidence to be integrated into updated models or public health estimates, and can also support translating clinical trial results into implementation considerations. In early product development stages, population-level modelling can provide quantitative evidence linking an intervention’s minimum key performance criteria, such as efficacy or duration with its projected public health impact and benefit to communities towards meeting health targets, thus contributing to a robust evidence base 8 . While clinical evidence is essential to inform efficacy estimates and eventual registration and funding decisions for products, clinical evidence also informs model parameters, and modelling evidence in turn can improve clinical trial planning and design.…”

“…The collaborative framework generates modelling evidence using dynamic, individual-based malaria transmission models, such as the OpenMalaria model that was developed over a period of 15 years. Detailed models are coupled with additional analytical and statistical approaches to enable rapid and computationally efficient searches of multi-dimensional parameter spaces spanning a wide range of intervention characteristics and settings 8 . Modelling the mechanisms of individual-level factors and population-level transmission dynamics links predictions of public health burden reduction to key intervention characteristics.…”