Our PhD student with the Population Health Sciences Institute and School of Mathematics, Statistics and Physics at Newcastle University has compared statistical methods that may help with the evaluation of new diagnostic tests.

It is important to measure the accuracy of diagnostic tests outside of the highly controlled research environment. This is crucial as tests can give rise to false positive and false negative results which have consequences for the patient.

The “diagnostic accuracy” of a new test is usually measured through benchmarking it against a highly reliable reference standard. The reference standard is often assumed to be a “gold standard”, that is, “error free”. However, no test is perfect and ignoring this imperfection can result in either over or underestimating the accuracy of a new test.

“Correction methods” have been developed by Staquet et al. and Brenner to help improve these estimations when the reference standard is imperfect. However, to our knowledge, no study has compared the statistical properties of these methods.

We are part of a team that analysed three clinical datasets to understand the impact of using each method to inform clinical decision-making. This work involved strong collaboration across Newcastle University with funding from a Newcastle University Research Excellence Award, the Population Health Sciences Institute within the Faculty of Medical Sciences, the School of Mathematics, Statistics and the NIHR Newcastle In Vitro Diagnostics Co-operative. The project was part of Chinyereugo Umemneku Chikere’s successfully completed PhD, which was supervised by Dr Joy Allen, Prof Luke Vale and Dr Kevin Wilson.

The Staquet et al. correction method outperformed the Brenner method in most scenarios we explored. However, in certain situations, such as where the prevalence of the target condition is very high or low, other statistical methods should be considered.

Read the full article in BMC Medical Research Methodology by following this link.