Automating social network models for tuberculosis contact investigation

Contact investigation to locate individuals at risk of acquiring tuberculosis can be an effective procedure, yet the resulting data can be unstandardized and difficult for disease investigators to synthesize when searching for transmission patterns. Social networks have been increasingly used to visualize these transmission patterns and to guide further contact investigation. Although off-the-shelf software exists to display networks, use of these tools requires technical knowledge, data manipulation, and database experience. In this dissertation, I hypothesize that software enabling interactive querying and visualization of networks will improve health workers' access to this data, and that contact investigation data are sufficiently standardized to utilize social network models.; The utility of social network models for contact investigation is first shown though case studies in San Francisco in which manually-generated networks are used to describe the social structure underlying two distinct outbreaks in homeless shelter and daycare settings. The nature of household contacts is also explored, showing a possible relationship between housing density and incidence of tuberculosis. Having provisionally demonstrated the utility of these models, I developed the Outbreak Investigator, a user-friendly tool that uses graph layout algorithms to automatically produce interactive social network graphics based on natural language queries. The tool is designed to simplify adaptation to alternate database architectures, and the user interface uses pre-defined queries developed in consultation with infectious disease controllers.; Formal evaluation of the software supports the primary hypothesis: users saved time performing common queries with Outbreak Investigator, and their overall satisfaction was improved over existing software. The standardization of contact investigation data is addressed with a nationwide survey of contact investigation forms, which are statistically analyzed to identify a core set of fields appropriate for social network models. The results points to a path of standardization and greater data consistency. This dissertation contributes to public health informatics by integrating social network modeling tools with under-utilized public health data, enabling widespread evaluation of a promising technique to improve outbreak management in tuberculosis and other diseases.