Improving surveillance and prediction of emerging and re-emerging infectious diseases

Infectious diseases are emerging at an unprecedent rate in recent years, such as the flu pandemic initialized from Mexico in 2009, the 2014 Ebola epidemic in West Africa, and the 2016-2017 expansion of Zika across Americas. They rarely happened previously and thus lack resources and data to detect and predict their spread. This highlights the challenges in emerging an re-emerging infectious disease surveillance. In the dissertation, I mainly put efforts in developing methods for early detection of such diseases, and assessing predictive power of various models in early phase of an epidemic. In Chapter 2, I developed a two-layer early detection framework which provides early warning of emerging epidemics based on the idea of anomaly detection. The framework could evaluate and identify data sources to achieve the best performance automatically from available data, such as data from the Internet and public health surveillance systems. I demonstrated the framework using historical influenza data in the US, and found that the optimal combination of predictors includes data sources from Google search query and Wikipedia page view. The optimized system is able to detect the onset of seasonal influenza outbreaks an average of 16.4 weeks in advance, and the second wave of the 2009 flu pandemic 5 weeks ahead. In Chapter 3, I extended the framework in Chapter 2 to identify large dengue outbreaks from small ones. The results show that the framework could personalize optimal combinations of predictors for different locations, and an optimal combination for one location might not perform well for other locations. In Chapter 4, I investigated the contribution of different population structures to total epidemic incidence, peak intensity and timing, and also explored the ability of various models with different population structures in predicting epidemic dynamics. The results suggest that heterogeneous contact pattern and direct contacts dominate the evolution of epidemics, and a homogeneous model is not able to provide reliable prediction for an epidemic. In summary, my dissertation not only provides method frameworks for building early detection systems for emerging and re-emerging infectious diseases, but also gives insight to the effects of various models in predicting epidemics.