Modeling the effect of distance from a hydro-electric dam on malaria incidence based on frailty and mixed Poisson regression models
Yehenew G. Kifle · Ghent University Academic Bibliography (Ghent University) · 2013
The research presented in this thesis is motivated by the question whether hydro-electric dams have an impact on malaria incidence. The specific problem that arises in the malaria data is the confounding of the clustering, village, with the covariate of interest, distance from the dam. The effect of this confounding problem is investigated in the multivariate survival data. In Chapter 1, the burden and epidemiology of malaria is put into context for Ethiopia. A review of different modeling approaches for clustered survival data is given. In Chapter 2, frailty and mixed effects Poisson regression models are applied to investigate whether household distance from a dam has an influence on malaria incidence risk and/or on mosquito abundance, thereby assessing the dynamics of malaria and its vectors in the southwestern part of Ethiopia as a function of season and distance from the dam. In Chapter 3, the performance of both the marginal and conditional Poisson regression models and hazard models are compared for modeling the effect of the distance from the dam on malaria incidence. In Chapter 4, different types of parametric proportional hazards model, i.e., the marginal model, the fixed effects model, the stratified model and the frailty model are presented and compared in the context of modeling the effect of household distance from a dam on time to malaria. Although, the frailty model is often considered to be the standard model for clustered survival data, in the case of the malaria data, parameter estimates from this model are a weighed combination of the within and between village estimate of the distance effect on time to malaria. Such a weighed combination, however, makes only sense if the same relationship holds between and within clusters. This assumption, however, is questionable for the malaria dataset that is considered here. Therefore, in such situation, where there is confounding between the clustering structure (village) and the covariate of interest (distance from the dam), we recommend to use either marginal time to event models with robust standard errors or a frailty model with two orthogonal covariates, one referring to the covariate effect between villages, and another referring to the covariate effect within villages.