Body Mass Index Trajectories and Association With Tuberculosis Risk in a Cohort of Household Contacts in Southern Africa

Background: Studies have demonstrated an inverse log-linear relationship between body mass index (BMI) and tuberculosis incidence. However, a person's BMI is dynamic, and longitudinal changes may be more informative than cross-sectional assessments. We evaluate the association between cross-sectional and changing BMI and risk of tuberculosis and describe longitudinal trajectories in a high-risk cohort.

Methods: ERASE-TB was a prospective longitudinal cohort study of household contacts ≥10 years in Southern Africa (Zimbabwe, Tanzania, and Mozambique), with 6-monthly follow-up up to 24 months. Associations between BMI and tuberculosis were investigated based on baseline (including hemoglobin) and changing BMI, using logistic, Poisson, and Cox models. Prevalent tuberculosis was defined as diagnosis during <30 days after recruitment. Growth mixture modelling was used to model longitudinal latent trajectories.

Results: Of 2107 recruited household contacts (621 [29.5%] adolescents and 1310 [62.2%] female), 520 (24.7%) were underweight. There were 21 and 41 people diagnosed with prevalent and incident tuberculosis, of whom 5/21 (23.8%) and 12/41 (29.3%) were underweight. Being underweight and anemic (adjusted hazard ratio: 3.77; 95% confidence interval: 1.50–9.51) and >10% negative change in BMI during follow-up (adjusted incidence rate ratio: 2.27; 95% confidence interval: 0.22–22.9) were associated with increased risk of incident tuberculosis. The association between continuous BMI-for-age Z-scores were nonlinear, with increased risk of tuberculosis with lower BMI. Four latent groups were defined in the growth mixture modelling: increasing, decreasing, and low/high stable BMI.

Conclusions: Declining BMI, regardless of absolute value, is a strong predictor of tuberculosis among household contacts. Longitudinal measurements should be considered in active case finding among tuberculosis-affected households.