Setting: British Columbia (BC) has a low incidence of tuberculosis (TB), with the burden of endogenously acquired disease concentrated among vulnerable populations, including the homeless. In May 2008, a TB outbreak began in a BC homeless shelter, with a single index case seeding multiple secondary cases within the shelter.
Objective: To use nightly shelter records to quantify the risk of latent tuberculous infection (LTBI) among shelter clients as a function of their sleeping distance from and duration of exposure to the index case.
Design: Distance and duration of exposure were visualised and assessed using logistic regression with LTBI status as outcome. We used a novel machine learning approach to establish exposure thresholds that optimally separated infected and non-infected individuals.
Results: Of 161 exposed shelter clients, 58 had a recorded outcome of infected (n = 39) or non-infected (n = 19). Only duration of exposure to the index was associated with increased odds of infection (OR 1.26); stays of ⩾ 5 nights put shelter clients at higher odds of infection (OR 4.97).
Conclusion: The unique data set and analytical approach suggested that, in a shelter environment, long-term clients are at highest risk of LTBI and should be prioritised for screening during an outbreak investigation.