1. Title: Evaluating diagnostic accuracy of tests and decision rules in the absence of a perfect reference test: Application to extra pulmonary tuberculosis

Funding: $229,500 from Canadian Institutes of Health Research (CIHR), 2018-2021.

Read more:

Tuberculosis (TB) is the world’s leading infectious disease killer with an estimated 9.6 million new cases of TB in 2014. Though public health efforts typically focus on pulmonary TB, extra pulmonary TB (EPTB) accounts for an unknown 15%-40% of the burden and is associated with important morbidity and mortality.

A key challenge to combating EPTB is the lack of a perfect diagnostic test, presenting a hurdle for disease diagnosis in individual patients, and for estimation of disease burden and evaluation of new tests at the population level. To date most studies estimating disease burden or diagnostic accuracy have relied on naïve methods resulting in biased estimates. Our objective is to address this by developing suitable Bayesian latent class models. This work is joint with Karen Steingart, Mikashmi Kohli, Samuel Schumacher, Emily MacLean, Madhukar Pai and others.

2. Title: Bayesian Methods for Latent Class Models

Funding: $80,000 from Natural Sciences and Engineering Research Council (NSERC), 2019-2024

Read more:

For many diseases, e.g. pneumonia, there is no perfect diagnostic test. This complicates estimation of the disease prevalence or the evaluation of a new diagnostic test. Latent class models (LCMs) provide a realistic way to quantify the uncertainty in these problems, yet they are not widely applied as several methodological concerns remain, e.g. modeling of conditional dependence, model comparison, choice of prior distributions.

This project explores possible solutions for these problems in the context of a single study and in the context of a meta-analysis. Another area of research is planning the sample size required to fit a latent class model. Throughout a Bayesian approach will be used for estimation and inference. This approach is particularly valuable as latent class models can be non-identifiable even when many tests are observed, and the sample size is large.

3. Title: Evaluation of Ultrasensitive Toxin Detection and Molecular Assays for the Diagnosis of Clostridium Difficile Infection (CDI) and Asymptomatic Colonization

Funding: $299,988 from Canadian Institutes of Health Research (CIHR), 2018-2021.

Read more:

Clostridium difficile (CD) is the most important cause of healthcare-associated infectious diarrhea in industrialized countries and is associated with increased morbidity and even mortality. Hence, accurate and rapid diagnosis of CDI are crucial for prompt appropriate treatment and prevention of transmission.

Diagnosis of CDI is supported by laboratory results. Currently available diagnostic tests for CD have variable performance characteristics and considerable debate persists regarding the optimal method of detection. Recently, a novel assay was developed using single molecular array (Simoa) technology. This proposal is focused on evaluating the accuracy of the Simoa test and its added value to routinely used tests. As none of the observed tests is considered perfect, the data will be analyzed using latent class (LC) analysis, a statistical modeling technique that can be applied when there is no adequate reference standard. The nominated principal investigator on this project is Dr. Vivian Loo.


Bayesian inference for HSROC diagnostic meta-analysis model

We pursue our series of blog articles illustrating rjags programs for diagnostic meta-analysis with an article describing how to carry out Bayesian inference for the...

Read more

Bayesian inference for the Bivariate model for diagnostic meta-analysis


Read more

Bayesian Sample Size Calculations for Difference in Proportions

Want to use a Bayesian approach to plan your study for comparing proportions? The R package SampleSizeProportions implements sample size calculations based on...

Read more

Learning rjags using simple statistical models

Here are some examples of fitting simple statistical models in rjags. They are useful for beginners learning Bayesian inference using the rjags library. Notice in...

Read more