Evaluation of Four Stool Processing Methods Combined With Xpert MTB/RIF Ultra for Diagnosis of Intrathoracic Paediatric TB

This is a diagnostic study evaluating the diagnostic accuracy of the Ultra assay in stools with a two-stage sequential design starting as a cohort of children with presumptive TB enriched in a second stage with Ultra positive TB cases on respiratory sample. It is both an ancillary to the TB-Speed HIV (C18-27) and the TB-Speed SAM (C18-28) studies and a study enrolling children from routine not enrolled in those two studies.

This design was chosen to be able to evaluate the sensitivity and specificity of the Ultra assay in a smaller sample size that is usually required by a "classical" prospective cohort design and avoiding the bias of overestimation of the sensitivity classically associated with the case-control design. In order to quickly generate data on appropriate stool processing method, and to contribute to the planned WHO recommendations for stool Ultra testing (expected 2nd semester 2020), a two-stage sequential design will be used. Indeed, knowing that on average only 10-15% of children with presumptive TB in a community-based setting will be confirmed, in order to reach the sample size of confirmed cases for the evaluation of sensitivity, 7 to 10 times more children with presumptive TB would need to be enrolled in a prospective design. On the other hand, the number of children with presumptive TB not confirmed with TB for the estimation of the specificity would be reached much earlier. In addition, based on the previous study results, it is known that the specificity of Xpert MTB/RIF assay in stool is high (99% CI:98-99), which would result in a relatively small sample size to evaluate the specificity of the Ultra in stools.

During the first stage, the investigators will offer to join all consecutive presumptive TB cases presenting at study sites to estimate specificity with the expected precision and calculate a preliminary sensitivity estimate. During the second stage, the investigators will keep enrolling only those from TB Speed studies and routine care who are Xpert positive on respiratory samples in order to estimate sensitivity with the expected precision.

This two-stage sequential design first estimating specificity then sensitivity has been described by Wruck et al. as an efficient way of validating diagnostic tests when the prevalence of the disease is low. It would not be feasible to consecutively enrol all children with presumptive TB to describe an expected sensitivity of 60% with 10% precision as this would require over 900 patients, of which, approximately 800 would be culture negative. In the two-stage process described by Wruck, only reference standard positive samples from the original population are selected in stage 2. The investigators adapted this design to the TB context as culture results will only be available after enrollment (and if the child is positive, only after the child has started treatment), hence selecting only those who are Ultra positive on respiratory samples for the second cohort as a way of enriching the study population with a subpopulation that has a higher TB prevalence probability, before their true disease status is confirmed. Other comparable diagnostic studies have either used greater resources to include larger samples sizes or have resorted to reporting imprecise estimates of sensitivity. To our knowledge, this is a relatively unique approach to study design for accuracy studies, with few published examples.

With such design, there should be no bias on the evaluation of the specificity similarly to a classical prospective design because this evaluation will be done among consecutively enrolled children with presumptive TB only. The sensitivity estimates may not be generalizable to all culture confirmed TB children due to the sampling approach. Xpert positive children will be more likely to have higher biological loads, causing a possible inflation of the sensitivity. However, the results will provide valuable information on variations of sensitivities of the different stool processing methods within this population.