RESEARCH ARTICLE A Novel Tool Improves Existing Estimates of Recent Tuberculosis Transmission in Settings of Sparse Data Collection

In any setting, a proportion of incident active tuberculosis (TB) reflects recent transmission (“recent transmission proportion”), whereas the remainder represents reactivation. Appro-priately estimating the recent transmission proportion has important implications for local TB control, but existing approaches have known biases, especially where data are incom-plete. We constructed a stochastic individual-based model of a TB epidemic and designed a set of simulations (derivation set) to develop two regression-based tools for estimating the recent transmission proportion from five inputs: underlying TB incidence, sampling cover-age, study duration, clustered proportion of observed cases, and proportion of observed clusters in the sample. We tested these tools on a set of unrelated simulations (validation set), and compared their performance against that of the traditional ‘n-1 ’ approach. In the validation set, the regression tools reduced the absolute estimation bias (difference between estimated and true recent transmission proportion) in the ‘n-1 ’ technique by a median [interquartile range] of 60 % [9%, 82%] and 69 % [30%, 87%]. The bias in the ‘n-1’ model was highly sensitive to underlying levels of study coverage and duration, and sub-stantially underestimated the recent transmission proportion in settings of incomplete data coverage. By contrast, the regression models ’ performance was more consistent across dif-ferent epidemiological settings and study characteristics. We provide one of these regres-sion models as a user-friendly, web-based tool. Novel tools can improve our ability to estimate the recent TB transmission proportion from data that are observable (or estimable) by public health practitioners with limited available molecular data