麥基爾大學台灣研究生同學會

Alternative and innovative approaches to TB diagnostics research: going beyond the test accuracy paradigm

Tuberculosis (TB) remains a global health problem. Diagnosis is the critical first step in TB control and several promising diagnostics are being developed, along with new drugs and vaccines. While the implementation of new diagnostic tools would greatly contribute to control of this disease, TB diagnostic research is still focused on measures of test accuracy (i.e. sensitivity and specificity). If we draw parallels between the evaluation of diagnostic tests and therapeutic drugs, TB diagnostics research remains in the early Phase I/II stages of development. In a citation analysis of TB diagnostic studies in PubMed and EMBASE from 2007-2008, almost 84% were evaluation studies. Of these, more than 80% focused on test accuracy, with sensitivity and specificity as the most frequently-reported outcomes.
There are limited data on outcomes such as accuracy of diagnostic algorithms (rather than single tests) and their contributions to the health care system, incremental value of new tests, impact of new tests on clinical decision-making and therapeutic choices, cost-effectiveness in routine programmatic settings, and impact on patient-important outcomes. This poses problems because research on test accuracy, while necessary, is not sufficient for policy and guideline development. Test accuracy data are surrogates for patient-important outcomes and cannot provide high quality evidence for policy making. In fact, the GRADE approach, which is used by the WHO and other agencies, downgrades accuracy studies as "low-quality evidence". Thus, the lack of clinically-relevant studies in TB diagnostics research poses a problem for the implementation of new tools for TB control.
My manuscript-based PhD thesis will focus on alternative and innovative approaches to evaluate TB diagnostics. Three projects are planned, two of which are based on secondary data. The first project is a primary data study at the Montreal Children’s Hospital (MCH). Nearly 30% of the children screened at the MCH have positive results on the tuberculin skin test (TST). Since the TST is an imperfect test, it is unclear whether or not they all require preventive therapy. As of June 2009, the MCH has implemented a new blood-based test called QuantiFERON-TB Gold In-Tube (QFT) for children with specific indications. The QFT has higher specificity than the TST and has the potential to reduce false-positive results. The goal of the study is to assess the impact of the new QFT test on clinical decision-making. The study asks pediatricians to document on a standardized questionnaire how the QFT result changed, if at all, their initial diagnostic and treatment decisions based on the TST. This study is ongoing and has recruited 150 participants as of February 2010.
Test accuracy measures assess the value of diagnostic tests in isolation. In reality, no test is performed alone. The value of a new test must be determined by estimating the added or incremental value of new tests, over and above conventional tests. To address this, I will make use of secondary data from the University of Cape Town, South Africa. I will use multivariable logistic regression models to reflect the diagnostic work-up in clinical practice, where tests are carried out in a sequential pattern in increasing order of complexity. By comparing areas under the Receiver Operating Characteristic (ROC) curves for the various logistic models, this approach can be used to estimate the incremental value of a new test in predicting disease beyond the patient data and conventional tests that are already available.
Secondary data analysis will be done using the TB-NEAT database from Groote Schuur Hospital. It consists of 500 consecutively-recruited adults (many HIV-positive) with suspected TB. The database contains clinical data, standard tests for TB (sputum smears and chest x-rays), and results for QFT and another test TSPOT.TB. Multivariable analysis will be performed to determine the added value of both tests. An extension of this research will involve using the coefficients from the regression models to develop a clinical scoring algorithm to diagnose TB. A similar method will be used on a dataset originating from Red Cross Children’s Hospital. I will re-analyze the data using the multivariable approach and create a scoring rule for the diagnosis of active TB in children.
While new technologies in the TB diagnostics pipeline offer great promise for TB control, limited resources mandate that we carefully evaluate them in clinically-meaningful ways before their implementation into routine practice. This manuscript-based PhD thesis attempts to address the need for clinical impact and incremental value studies that go beyond the test accuracy paradigm.