Comment www. thelancet. com/respiratory Vol 7 March 2019 203 detection software has the potential to standardise recognition of radiographic abnormalities and reduce cost and personnel requirements relative to standard chest x-ray, 9 ongoing investigation is strongly warranted. However, identification of tuberculosis screening tests meeting WHO-recommended performance and operational characteristics is urgently needed. 7 Relative to the tuberculosis confirmatory test pipeline, the number of novel tuberculosis screening tests in development or under evaluation is low (table). C-reactive protein (CRP; 8 mg/L cutoff point) has shown promise in terms of accuracy (sensitivity 90% and specificity 70%, in reference to culture), cost (≤ $2 per test), and point-ofcare implementation. 10 However, prospective evaluation has been restricted to patients with advanced HIV, and studies suggest that CRP will probably not meet these same standards if applied to HIV subgroups with lower tuberculosis risk. 10 These results are an important reminder that the diagnostic accuracy of any test depends on key characteristics of the intended population: sensitivity depends on the spectrum of clinical tuberculosis severity in the population whereas specificity depends on the prevalence of conditions that can cause false-positive test results. Therefore, it might be unrealistic to expect that a single test or test cutoff point will be appropriate for all populations and settings; different strategies will be needed to screen different populations for tuberculosis. Although novel approaches to tuberculosis biomarker discovery have identified tools (eg, RNA11 and protein12 signatures, Mtb antigen peptides13) that are potentially more sensitive and specific than CRP, these tests are in the proof-of-concept stage in which evaluation has been largely restricted to patients with presumptive tuberculosis (not in the context of tuberculosis screening) and substantial challenges remain in translating these novel approaches to affordable tests. In summary, the shortage of adequate tuberculosis screening tests represents a major obstacle to detecting the so-called missing millions. To develop more effective tuberculosis screening strategies, symptom screening must be recognised as being insufficiently sensitive in populations who account for the majority of tuberculosis cases worldwide. Next, investment into the development of screening tests meeting desired performance and operational characteristics must be substantially increased, as should evaluation of promising tools in well characterised tuberculosis screening cohorts that include populations not currently targeted for systematic screening. Lastly, development and implementation of tuberculosis screening tests that might only benefit specific key populations must not be discouraged. Modelling studies suggest that identification of effective tuberculosis screening tests could substantially reduce tuberculosis incidence and mortality, 14 and the potential clinical and public health effects of such tests argue strongly for increased attention as important components of a comprehensive strategy to end tuberculosis globally.