<sec><title>BACKGROUND</title>Low body mass index (BMI) is a globally important risk factor for TB progression. Little is known about this association in people living with HIV (PLHIV) and the functional form of the BMI-TB incidence curve.</sec><sec><title>METHODS</title>Secondary analysis of a randomised controlled trial of TB preventive therapy among PLHIV in South Africa, Mozambique, and Ethiopia. Participants received 3 months of weekly high-dose rifapentine-isoniazid given once or twice over a period of 2 years. Multivariable fractional polynomials (MFPs) were used to investigate functional forms of BMI. Time to incident TB was modelled using Cox's proportional hazard regression.</sec><sec><title>RESULTS</title>A total of 76 TB events were documented, giving an overall TB incidence rate of 1.2 per 100 person-years (95%CI 1.0-1.6). Baseline BMI <18.5 kg/m² was associated with a 2.6-fold increased hazard of TB compared with BMI 18.5-24.9 kg/m² (aHR 2.6, 95% CI 1.4-4.8, P < 0.001). BMI ≥30 kg/m² was associated with a lower hazard of TB (aHR 0.5, 95% CI 0.2-1.0). Continuous and categorical BMI showed weak evidence of quadratic dose-response relationships (P = 0.08 and P = 0.09, respectively). MFP analysis was consistent with a decline in TB incidence for increasing BMI to around 25 kg/m², followed by a less steep decline in TB incidence for increasing BMI >25 kg/m².</sec><sec><title>CONCLUSIONS</title>In PLHIV, BMI showed an inverse log-linear association with TB incidence. The MFP approach showed that the relationship is more complex than a simple log-linear association.</sec>.