Developing statistical methods to improve the diagnosis and care of hospitalised patients with COVID-19

This research will develop innovative, and generalisable, statistical methods to assess the relationship between, and utility of, combinations of diagnostic tests for healthcare decision-making in COVID-19.

Diagnostic tests are typically assessed by independently synthesising evidence on measures of their accuracy (such as sensitivity/specificity).

However, multiple test combinations (including near-patient testing, serologic immunoassays, radiological imaging and PCR) may need to be evaluated to make a comprehensive diagnosis in complex diseases, which are difficult to diagnose early and include rapidly evolving near-patient testing.

Novel statistical methods are required to describe interdependencies between tests to deliver optimal diagnosis and healthcare decision-making for patients.

To facilitate this goal, I will address each of the following aims: 1) Identify the most appropriate diagnostic test from electronic health records for early detection of COVID-19 This work package will develop methods for the comparison of individual diagnostic tests to identify the most appropriate single test for early detection of COVID-19. 2) Determine the optimal combination of diagnostic tests for diagnosis and management of COVID-19 This work package will develop methods to model the relationship between diagnostic tests to identify the optimal sequences of tests for COVID-19. 3) Allow for different sources (heterogeneity) of evidence to inform healthcare decision-making in COVID-19 This work package will develop methods to combine individual patient data with summary data to ensure that all available evidence is considered for healthcare decision-making.

This research has the opportunity to influence emerging guidance on the appropriate evaluation of combinations of diagnostic tests, and management of COVID-19.