Date of Award

January 2013

Document Type

Open Access Thesis

Degree Name

Master of Public Health (MPH)

Department

School of Public Health

First Advisor

Maria Diuk-Wasser

Abstract

Background: Avian Influenza (H5N1) has become entrenched in Egypt since its emergence in 2006. Control measures have failed and surveillance systems remain inadequate. A relatively new method for regression called Random Forests is presented here with the goal of providing accurate and timely predictions of the weekly number of outbreaks in each of the Egyptian governorates.

Methods: Predictions were generated from Random Forests models using outbreak data from the FAO EMPRES-i database, and local weather data from Weather Underground. This data was lagged by one and two weeks in order to make prospective predictions with the current week's data in the future. Model performance was assessed using a variety of methods.

Results: The percent of the variance in observed outbreaks explained by the model in each of the governorates varied greatly, ranging between 20 and 60 percent in governorates with high and medium outbreak activity. The models typically predicted poorly in governorates with low activity. Linear regression of the observed outbreaks on the predicted values provided evidence that while outbreaks were consistently underpredicted across all governorates, predictions in some models tracked observed outbreaks quite accurately.

Discussion: The varying levels of model performance in each of the governorates raises many questions about why this is. While we cannot deduce these reasons from the models themselves, public health officials can use the lessons learned here as a guide to focus future research to better understand what is occurring. Predictive models can be used to evaluate local surveillance systems, and find additional covariates for the model to determine the spatio-temporal risk of avian influenza. As a result of better surveillance data and more complete models, control and prevention measures may be more effectively put in place where and when they are needed most.

Comments

This is an Open Access Thesis.

Share

COinS