Bayesian evidence synthesis to estimate subnational TB incidence: An application in Brazil

• Main Authors: Melanie H. Chitwood, Daniele M. Pelissari, Gabriela Drummond Marques da Silva, Patricia Bartholomay, Marli Souza Rocha, Mauro Sanchez, Denise Arakaki-Sanchez, Philippe Glaziou, Ted Cohen, Marcia C. Castro, Nicolas A. Menzies
• Format: Article
• Language: English
• Published: Elsevier 2021-06-01
• Series: Epidemics
• Subjects: Tuberculosis; Bayesian; Subnational; Estimation; Incidence
• Online Access: http://www.sciencedirect.com/science/article/pii/S1755436521000062

Background: Evidence on local disease burden and the completeness of case detection represent important information for TB control programs. We present a new method for estimating subnational TB incidence and the fraction of individuals with incident TB who are diagnosed and treated in Brazil. Methods: We compiled data on TB notifications and TB-related mortality in Brazil and specified an analytic model approximating incidence as the number of individuals exiting untreated active disease (sum of treatment initiation, death before treatment, and self-cure). We employed a Bayesian inference approach to synthesize data and adjust for known sources of bias. We estimated TB incidence and the fraction of cases treated, for each Brazilian state and the Federal District over 2008–2017. Findings: For 2017, TB incidence was estimated as 41.5 (95 % interval: 40.7, 42.5) per 100 000 nationally, and ranged from 11.7–88.3 per 100 000 across states. The fraction of cases treated was estimated as 91.9 % (89.6 %, 93.7 %) nationally and ranged 86.0 %–94.8 % across states, with an estimated 6.9 (5.3, 9.2) thousand cases going untreated in 2017. Over 2008–2017, incidence declined at an average annual rate of 1.4 % (1.1 %, 1.9 %) nationally, and −1.1%–4.2 % across states. Over this period there was a 0.5 % (0.2 %, 0.9 %) average annual increase in the fraction of incident TB cases treated. Interpretation: Time-series estimates of TB burden and the fraction of cases treated can be derived from routinely-collected data and used to understand variation in TB outcomes and trends.