Background: Many children with pulmonary tuberculosis remain undiagnosed and untreated with related high morbidity and mortality. Diagnostic challenges in children include low bacterial burden, challenges around specimen collection, and limited access to diagnostic expertise. Algorithms that guide decisions to initiate tuberculosis treatment at primary healthcare centres in resource-limited settings could help to close the persistent childhood tuberculosis treatment gap. Recent advances in childhood tuberculosis algorithm development have incorporated prediction modelling, but studies conducted to date have been small and localised, with limited generalisability. We assembled individual participant data (IPD) from children being investigated for pulmonary tuberculosis in high-tuberculosis incidence settings, which we leveraged to 1) evaluate the performance of currently used diagnostic algorithms and 2) develop evidence-based algorithms to assist in tuberculosis treatment decision-making for children presenting to primary healthcare settings.

Methods: We collated IPD including clinical, bacteriological, and radiologic information from prospective diagnostic studies in high-tuberculosis incidence settings enrolling children <10 years with presumptive pulmonary tuberculosis. Using this dataset, we first retrospectively evaluated the performance of several existing treatment-decision algorithms. We then developed multivariable prediction models and investigated model generalisability using an internal-external cross-validation framework. A team of experts provided input to adapt the models into scoring systems with pre-determined sensitivity thresholds of 85% to be incorporated into pragmatic treatment-decision algorithms for use in resource-limited, primary healthcare settings.

Findings: Of 4,718 children from 13 studies from 12 countries, 1,811 (38.4%) were classified as having pulmonary tuberculosis; 541 (29.9%) bacteriologically confirmed and 1,270 (70.1%) unconfirmed. Existing treatment-decision algorithms had highly variable diagnostic performance. The scoring system derived from the prediction model that included clinical features and features from chest x-ray had a combined sensitivity of 86% [95% confidence interval (CI): 0.68-0.94] and specificity of 37% [95% CI: 0.15-0.66] against a composite reference standard. The scoring system derived from the model that included only clinical features had a combined sensitivity of 84% [95% confidence interval (CI): 0.66-0.93] and specificity of 30% [95% CI: 0.13-0.56] against a composite reference standard.

Interpretation: We adopted an evidence-based approach to develop pragmatic algorithms to guide tuberculosis treatment decisions in children, irrespective of the resources locally available. This approach will empower health workers in resourcelimited, primary healthcare settings to initiate tuberculosis treatment in children in order to improve access to care and reduce tuberculosis-related mortality. These algorithms have been included in the operational handbook accompanying the latest WHO guidelines on the management of tuberculosis in children and adolescents. Future prospective evaluation of algorithms, including those developed in this work, is necessary to investigate clinical performance.