Anticipating Outbreaks: Predictive Modeling to Improve Infectious Disease Surveillance
MetadataShow full item record
CitationMcGough, Sarah. 2019. Anticipating Outbreaks: Predictive Modeling to Improve Infectious Disease Surveillance. Doctoral dissertation, Harvard University, Graduate School of Arts & Sciences.
AbstractRapid and effective responses to disease outbreaks require the ability to accurately detect and anticipate changing dynamics of an outbreak over time. However, disease surveillance is frequently undermined by extended delays between symptom onset and official case reports, often due to complex and multi-tiered disease reporting and communication systems interacting at national, state, and city levels. Timeliness of reporting and response may be further exacerbated in settings that experience resource constraints.
Digital data streams that are available in real- or near-real-time have the potential to complement or improve traditional disease surveillance by quickly and continuously capturing signals of population health that may be meaningful for disease tracking and forecasting. In addition, digital data are trending towards being made freely and publicly available through public servers and APIs, which remove barriers to data access and open up avenues for predictive modeling independent of resource level. Further, methodologies that focus on data-driven and self-adaptive learning can yield flexible and readily-implementable models for the public health sector.
Focusing on a collection of inputs, including Google search trends, Twitter, news reports, and satellite weather data, and employing statistical and machine learning methodologies to process, synthesize, and analyze these data, I present several applications of disease detection and forecasting models, which are developed as real-time decision support tools. Each project uses, as a case study, a mosquito-borne disease outbreak, which requires anticipation on the scale of weeks or months to effectively interrupt transmission. Across case studies, I describe flexible models functional at both large (e.g. national) and small (e.g. city-level) spatial scales. Over the course of this thesis, I move towards increasingly more generalizable modeling techniques such that learning from input data becomes more autonomous, requires less human input, and can be applied to a wider range of systems (e.g. surveillance bodies, diseases). In all cases, I show how predictions can fill a critical time gap between case onset and case reporting, with the goal of supporting early warnings and outbreak anticipation within public health surveillance systems.
Citable link to this pagehttp://nrs.harvard.edu/urn-3:HUL.InstRepos:42029601
- FAS Theses and Dissertations