Jointly Leveraging Mathematical Models and Data to Understand Malaria Transmission and Control

Malaria is a vector-borne disease caused by infection with the Plasmodium species parasite, and it is a major threat to health systems around the globe. To maintain progress towards elimination of malaria, a continued and detailed understanding of malaria epidemiology is needed. In this Dissertation, I present five analyses that showcase how mathematical models can be leveraged alongside routinely collected data to better inform our understanding of malaria transmission and control.

First, to demonstrate how the assumptions that we make about routinely collected surveillance data affect our estimates of malaria transmission, I developed an inference algorithm to infer person-to-person networks of P. falciparum malaria transmission and applied it to near-elimination settings. Using Eswatini as a case study, I demonstrated that the choice of data types and assumptions therein affect the epidemiological conclusions that we reach. Further, through extensive simulation, I revealed that routine surveillance data is insufficient to yield reliable estimates of transmission in near-elimination settings.

Second, I fitted transmission models of P. falciparum and P. vivax to a malaria epidemic in Southeastern Venezuela. Accounting for the clinical definitions applied in the surveillance system, I estimated that many P. vivax relapses were likely misdiagnosed as reinfections, potentially leading to an overestimate of P. vivax transmission from routine surveillance data. Then, using the fitted transmission models, I projected the impact of interventions in the region, should resources be mobilized in time. I predicted that long-lasting insecticidal nets are likely to be sufficient to obtain short-term reductions in incidence for P. falciparum, but mass drug administration with radical cure is needed to control P. vivax.

Third, to demonstrate how mathematical models can further augment our understanding from data, I performed a theoretical analysis that identified biases that are likely to occur in phase-III clinical trials of 8-aminoquinolines. Using an individual-based model, I simulated phase-III clinical trials under different epidemiological settings. I found that features of the clinical trial setting, such as transmission intensity and the rate of P. vivax relapse, interacted with features of the clinical trial design, such as the duration of follow-up, to lead to strong downward biases in efficacy estimates. This indicated that the effect of 8-aminoquinolines against P. vivax hypnozoites may be underestimated. Using the individual-based model, I then demonstrated that the effect of these biases can be reduced by using vector control measures and parasite genotyping.

Fourth, to show how mathematical models can synthesize insights across distinct data sources, I performed a systematic review and meta-analysis of studies comparing the performance of light microscopy and polymerase chain reaction for diagnosing Plasmodium spp. infections. Using a hierarchical Bayesian latent class model, I estimated the sensitivity and specificity of light microscopy for each Plasmodium spp. My analysis revealed that the sensitivity of light microscopy is particularly low for P. knowlesi, P. malariae, and P. ovale, suggesting that the burden of these Plasmodium spp. may be underestimated in a clinical setting.

Fifth, to demonstrate how mathematical modeling can help to identify data needs moving forward, I considered how misdiagnosis may cause us to underestimate the transmission and burden of emerging infectious diseases from surveillance data. Using P. knowlesi as a case study, I performed a theoretical analysis using a branching process framework and realistic estimates of misdiagnosis to show that the burden and transmission of P. knowlesi may be greater than previously thought. These results underscore the importance of evaluating the sensitivity of diagnostics for use in surveillance of infectious diseases.

In conclusion, my research demonstrates how the use of mathematical models alongside routinely collected data can provide greater insights about the transmission and control of malaria than would be obtained using either one alone.