Seminar in applied mathematics and statistics
SPEAKER: Kathryn Colborn (Department of Biostatistics and Informatics, Colorado School of Public Health, University of Colorado).
TITLE: Spatial and Spatio-temporal models for malaria research
ABSTRACT: I will discuss statistical methodologies applied in two malaria projects in Sub-Saharan Africa with which I have been involved over the past two years. The first project was funded by a Grand Challenges Explorations Grant through the Bill and Melinda Gates Foundation, for which I was the PI. In this study, we developed a malaria early warning system (MEWS) for Mozambique. Our outcome of interest was weekly case reports of children under 5 years of age from 142 districts over five years. We developed a spatio-temporal Poisson mixed model based on explanatory weather variables in order to map exceedance probabilities (EPs), defined as the predictive probability that the relative risk exceeds a predefined threshold. This work was published in Scientific Reports in 2018. The second project is the design of the second phase of interventions for a malaria clinical trial in Uganda. In phase one, three communities were assigned to one of three interventions: 1) mass drug administration of antimalarials plus indoor residual spraying of households (IRS); 2) IRS only; 3) standard of care. In phase two, we desired to randomize the 55 villages within arms 1 and 2 to cross-over to a second set of interventions that presumably will “hold-down” prevalence after the more intense interventions. To randomize the villages to one of two new arms, we present a novel framework for spatially explicit covariate-based constrained randomization (CCR) of clusters that first assigns villages to clusters, both spatially and by covariate distances. We then use CCR to assign the clusters to treatment arms in order to achieve balance with respect to cluster-level confounders between treatment arms. This methodology is important when interventions need to be applied to clusters, and contamination of the intervention is possible geographically (e.g., mosquito and human travel).