Geographical prioritization of tuberculosis case-finding activities

Tuberculosis (TB) causes more deaths globally than all other infectious diseases. Because people with TB may remain undiagnosed for longer periods of time, it is important to provide early diagnosis to limit the spread of disease. Therefore TB program planners are tasked with identifying geographic areas where active case-finding (ACF) may have the largest public health impact using the fewest resources, i.e. by diagnosing more people with TB while screening fewer.
This thesis explores the challenges of preparing suitable datasets for geographic TB risk profiling, and of geoprioritization for ACF. We described the practice of accounting for spatial autocorrelation in epidemiological studies of geographic patterns in infectious diseases and described the challenges of data digitization in TB ACF settings. We also evaluated the use of telecom mobility data to reflect geographic access to TB services. We developed and evaluated the performance of geographic risk models for ACF efficiency in Pakistan and Viet Nam. In Pakistan we compared the performance of a Bayesian machine-learning model with a negative binomial spatial lag regression, and we assessed changes in the machine learning model’s performance across model updates. In Viet Nam we executed a principal component analysis to rank TB risk at the commune level.
Efficiently digitizing ACF data and collecting spatial information can greatly improve readiness to support subnational TB decision-making. We concluded that there are multiple valid geographic risk profiling methods for TB, and that these can support geoprioritization of ACF in higher risk and greater need areas for improved case-finding efficiency. Where machine learning models are used for this purpose, it is essential to validate these outputs using statistical and human monitored approaches.