Surveillance data from large and small spatial scales play an essential role in public health decision making and the scientific investigation of infectious disease. However, such data are subject to missing observations, delays in reporting, and observational biases that can lead to increased uncertainty and incorrect conclusions. Existing uncertainty and bias are amplified when we attempt to predict future disease incidence. This project aims to develop and extend statistical and modeling methodologies to correct for biases in surveillance data, impute missing data, predict the course of epidemics, and appropriately characterize uncertainty at relevant spatial scales. An integrated Bayesian framework and Markov chain Monte Carlo methods will be used to combine data from multiple sources at different spatial scales to better understand macro- and micro-scale dengue dynamics in Thailand. Dengue is a mosquito-borne virus which circulates in over 100 countries, is reemerging throughout much of the western hemisphere, and is responsible for an estimated 50 million infections and 19,000 deaths worldwide each year. Specifically, this projects aims to: (1) develop an integrated statistical framework for predicting missing surveillance data by using information from correlated locations and non-traditional data (e.g., search queries); (2) develop methods that exploit mechanistic models of disease transmission and spatial smoothing techniques to bridge the gap between data collected at differing spatial scales while appropriately quantifying uncertainty; and (3) create improved methods for the analysis of spatio- temporal point pattern data that: appropriately account for biases in data collection, elucidate patterns of spatio-temporal dependence at the appropriate scale and define the spatial scale of disease transmission. Methods will be tested and validated using simulated data sets, over four decades of province level dengue surveillance data collected by the Thai Ministry of Public Health, district level surveillance level for over 20 years from several Thai provinces, and point pattern data on the exact location where dengue cases presenting at select hospitals reside. This project will result in the creation and dissemination of novel methods for dealing with bias, missing data and uncertainty in regional, local and point pattern statistics. These methods will aid the appropriate interpretation of macro-levels statistics at the local level and create new tools for using point pattern data to answer scientific questions about disease spread. In addition, fulfillment of these aims will increase our understanding of dengue transmission in Thailand at a variety of spatial scales, improve the Thai dengue surveillance system by providing an integrated approach for predicting future incidence and imputing missing data, and aid in modeling and responding to emerging dengue epidemics elsewhere in the world. The resulting methods will be disseminated in peer reviewed publications and R packages (including source code and validation data sets) freely available on the Comprehensive R network; allowing public health practitioners and researchers from a variety of disciplines to utilize and build upon this work.
Surveillance data from large and small spatial scales play an essential role in public health and the scientific research, but these data are subject to missing observations, delays in reporting, and observational biases. The proposed study aims to develop and extend statistical and modeling methodologies to correct for biases in surveillance data, impute missing data, predict the course of epidemics, and appropriately characterize the uncertainty in estimates and predictions at relevant spatial scales. Methods will be tested and validated using Thai dengue surveillance data, but should be applicable to a wide variety of diseases and contexts.
Showing the most recent 10 out of 27 publications
