Grant: Operating Grant

For many diseases (including many infectious diseases and mental illnesses) lack of a perfect gold-standard test makes disease diagnosis and evaluation of new diagnostic tests challenging. We propose to develop statistical methods needed for the development of a diagnostic prediction model for childhood tuberculosis (TB), a disease for which there is no perfect gold-standard. TB presents a huge public health burden globally, and is also a concern among certain groups in Canada such as recent immigrants and First Nations. Accurate diagnosis is a key part of the public health strategy to control TB. Diagnosis of childhood TB must rely on multiple imperfect measures, some of which are specific for TB diagnosis (like smear microscopy or the GeneXpert test) and others which are non-specific (like assessment of signs and symptoms or chest x-ray). The risk of TB is also affected by other variables, such as whether the child is malnourished or has HIV. Statistical models called latent class models can aid in developing a systematic diagnostic rule for TB based on the imperfect tests and patient characteristics. We propose to develop such models to address the following objectives: i) Estimation of the accuracy of the imperfect measures of childhood TB, ii) Development of a diagnostic prediction rule, and iii) Determination of the added value of new, expensive tests over routinely available measures. The main contribution of the project is that it will aid in improving diagnosis of TB in children. This project will bring together researchers in Canada, South Africa and The Netherlands. The methods and statistical software we propose will also find application in areas besides tuberculosis where there is no gold-standard test.