Computer-aided reading of tuberculosis chest radiography: moving the research agenda forward to inform policy

Tuberculosis is a major source of global mortality caused by infection, partly because of a tremendous ongoing burden of undiagnosed disease. Improved diagnostic technology may play an increasingly crucial part in global efforts to end tuberculosis, but the ability of diagnostic tests to curb tuberculosis transmission is dependent on multiple factors, including the time taken by a patient to seek health care, the patient's symptoms, and the patterns of transmission before diagnosis. Novel diagnostic assays for tuberculosis have conventionally been evaluated on the basis of characteristics such as sensitivity and specificity, using assumptions that probably overestimate the impact of diagnostic tests on transmission. We argue for a shift in focus to the evaluation of such tests' incremental value, defining outcomes that reflect each test's purpose (for example, transmissions averted) and comparing systems with the test against those without, in terms of those outcomes. Incremental value can also be measured in units of outcome per incremental unit of resource (for example, money or human capacity). Using a novel, simplified model of tuberculosis transmission that addresses some of the limitations of earlier tuberculosis diagnostic models, we demonstrate that the incremental value of any novel test depends not just on its accuracy, but also on elements such as patient behaviour, tuberculosis natural history and health systems. By integrating these factors into a single unified framework, we advance an approach to the evaluation of new diagnostic tests for tuberculosis that considers the incremental value at the population level and demonstrates how additional data could inform more-effective implementation of tuberculosis diagnostic tests under various conditions.

Background Chest radiography to diagnose and screen for pulmonary tuberculosis has limitations, especially due to inter-reader variability. Automating the interpretation has the potential to overcome this drawback and to deliver objective and reproducible results. The CAD4TB software is a computer-aided detection system that has shown promising preliminary findings. Evaluation studies in different settings are needed to assess diagnostic accuracy and practicability of use. Methods CAD4TB was evaluated on chest radiographs of patients with symptoms suggestive of pulmonary tuberculosis enrolled in two cohort studies in Tanzania. All patients were characterized by sputum smear microscopy and culture including subsequent antigen or molecular confirmation of Mycobacterium tuberculosis (M.tb) to determine the reference standard. Chest radiographs were read by the software and two human readers, one expert reader and one clinical officer. The sensitivity and specificity of CAD4TB was depicted using receiver operating characteristic (ROC) curves, the area under the curve calculated and the performance of the software compared to the results of human readers. Results Of 861 study participants, 194 (23%) were culture-positive for M.tb. The area under the ROC curve of CAD4TB for the detection of culture-positive pulmonary tuberculosis was 0.84 (95% CI 0.80–0.88). CAD4TB was significantly more accurate for the discrimination of smear-positive cases against non TB patients than for smear-negative cases (p-value<0.01). It differentiated better between TB cases and non TB patients among HIV-negative compared to HIV-positive individuals (p<0.01). CAD4TB significantly outperformed the clinical officer, but did not reach the accuracy of the expert reader (p = 0.02), for a tuberculosis specific reading threshold. Conclusion CAD4TB accurately distinguished between the chest radiographs of culture-positive TB cases and controls. Further studies on cost-effectiveness, operational and ethical aspects should determine its place in diagnostic and screening algorithms.

Background To inform the choice of an appropriate screening and diagnostic algorithm for tuberculosis (TB) screening initiatives in different epidemiological settings, we compare algorithms composed of currently available methods. Methods Of twelve algorithms composed of screening for symptoms (prolonged cough or any TB symptom) and/or chest radiography abnormalities, and either sputum-smear microscopy (SSM) or Xpert MTB/RIF (XP) as confirmatory test we model algorithm outcomes and summarize the yield, number needed to screen (NNS) and positive predictive value (PPV) for different levels of TB prevalence. Results Screening for prolonged cough has low yield, 22% if confirmatory testing is by SSM and 32% if XP, and a high NNS, exceeding 1000 if TB prevalence is ≤0.5%. Due to low specificity the PPV of screening for any TB symptom followed by SSM is less than 50%, even if TB prevalence is 2%. CXR screening for TB abnormalities followed by XP has the highest case detection (87%) and lowest NNS, but is resource intensive. CXR as a second screen for symptom screen positives improves efficiency. Conclusions The ideal algorithm does not exist. The choice will be setting specific, for which this study provides guidance. Generally an algorithm composed of CXR screening followed by confirmatory testing with XP can achieve the lowest NNS and highest PPV, and is the least amenable to setting-specific variation. However resource requirements for tests and equipment may be prohibitive in some settings and a reason to opt for symptom screening and SSM. To better inform disease control programs we need empirical data to confirm the modeled yield, cost-effectiveness studies, transmission models and a better screening test. Electronic supplementary material The online version of this article (doi:10.1186/1471-2334-14-532) contains supplementary material, which is available to authorized users.