Screening

Further evidence is needed about the performance of CAD software stratified according to the characteristics of the individual being evaluated (e.g. by smear status, HIV status, age cohort, history of TB, smoking status, sex) to allow for better setting-specific and patient-specific calibration of CAD programmes.

More research on users’ perspectives is needed about CAD technologies in TB screening and triage, including their perceived acceptability to patients, providers and other stakeholders.

The GDG considered data on using mWRDs for screening children and adolescent outpatients accessing health care. They felt that the data, which included 2 studies with 787 participants and had results demonstrating substantial heterogeneity, represented insufficient evidence to establish an accurate and reliable estimate of the diagnostic accuracy of mWRDs, and, thus, the GDG decided not to issue a recommendation for or against their use as a screening tool for children and adolescents.

TB screening tools are designed to distinguish people with a higher probability of having TB disease from those with a low probability and can be assumed to be free of TB disease. They are not intended to provide a definitive diagnosis. In general, they need to be able to be implemented easily and relay results rapidly in order to be informative in a screening context. Screening tests need to be followed by a diagnostic test, offered as part of a comprehensive clinical evaluation, to confirm or rule out TB disease in individuals who screen positive.

Well-designed clinical trials are needed to strengthen the evidence about the accuracy, effectiveness (including the impact on patient-important outcomes such as mortality), feasibility and cost implications of using the W4SS, CRP, CXR and mWRD to screen for TB across all HIV subpopulations in settings with low, medium and high burdens of HIV and TB with and without high ART coverage.

Well-designed trials and rigorous quasi-experimental studies in different settings are required to investigate the effects of population-wide systematic screening for TB on individual-level outcomes (diagnostic delay, treatment outcomes, costs to patients, social consequences) and populationlevel outcomes (TB prevalence, incidence, transmission), as well as to guide implementation choices (including the method of delivery, screening algorithms, the duration of screening intervals and frequency of screening, and the mode of delivering the intervention).

When CAD is implemented, local calibration is advised to customize the score thresholds to the requirements of the programme (see the operational handbook for more details about how this can be done) (7). After this initial calibration, ongoing monitoring and analysis of CAD performance should be carried out to assess the correlation with human readers’ interpretations and with bacteriological confirmation, and with the proportion of images read as abnormal and requiring further investigation.

A successful screening programme may lead to a diminishing yield, at least if the risk group is a fixed population. Over time, changes in the background burden of TB, as well as changes in the profile of TB patients in the community (e.g. a trend towards fewer patients with symptomatic TB) can lead to a reduction in the yield from screening, an increase in the NNS, a reduction in cost–effectiveness and a change in the ratio of benefits to harms.