Comparative evaluation of methods that adjust for reporting biases in participatory surveillance systems

Abstract

Over the past decade the widespread proliferation of mobile devices and wearable technology has significantly changed the landscape of epidemiological data gathering and evolved into a field known as Digital Epidemiology. One source of active digital data collection is online participatory syndromic surveillance systems. These systems actively engage the general public in reporting health-related information and provide timely information about disease trends within the community. This dissertation comprehensively addresses how researchers can effectively use this type of data to answer questions about Influenza-like Illness (ILI) disease burden in the general population. We assess the representativeness and reporting habits of volunteers for these systems and use this information to develop statistically rigorous methods that adjust for potential biases. Specifically, we evaluate how different missing data methods, such as complete case and multiple imputation models, affect estimates of ILI disease burden using both simulated data as well as data from the Australian system, Flutracking.net. We then extend these methods to data from the American system, Flu Near You, which has different patterns. Finally, we provide examples of how this data has been used to answer questions about ILI in the general community and promote better understanding of disease surveillance and data literacy among volunteers.