Diagnostic triage and treatment stratification for respiratory virus infections

Background: Acute respiratory tract infections (ARIs) are the commonest syndromic infectious disease presentation to UK emergency departments. Up to 50% of these are viral in aetiology. Despite improvements in molecular testing, there remains substantial underdiagnosis of viral ARIs due to a lack of consensus over the breadth of testing and resource constraints. Treatment of viral ARIs also remains suboptimal, with controlled trials repeatedly failing to show substantial benefits of antiviral treatment in influenza and COVID-19. Novel approaches are required to accurate triage patients most likely to have viral ARI and stratify those mostly likely to benefit from antiviral treatment. Blood RNA biomarkers are being translated to near-patient diagnostics and show promise to facilitate early detection of respiratory viral infections. Further, we have shown that blood MX-1 transcript levels correlate with live virus replication and could guide stratification for antiviral therapy. Aim: To improve the diagnosis of acute respiratory viral infections and stratify patients most likely to benefit from antiviral treatment using blood RNA biomarkers. Objectives: Develop and validate multivariable statistical and machine learning models incorporating RNA biomarkers to predict respiratory viral infection among adults presenting to UK emergency departments. Test the generalisability of the best performing multivariable model from Objective 1 in immunocompromised haematology-oncology patients in the UK. Test the hypothesis that the treatment effect of oseltamivir on influenza viral load is limited to individuals with high baseline MX-1 transcript and protein levels. Methods: I will develop and validate multivariable regression and machine learning models to predict personalised risk of viral infection in adults with suspected ARI (n=500) from the Bioresource in Adult Infectious Disease (BioAID) cohort of emergency department attendees. I will include: (a) routinely collected clinical and laboratory data; (b) contemporaneous prevalence rates for positive virological testing; and (c) the best performing host-response blood RNA biomarker of viral infection identified from a systematic review of candidates. I will evaluate the diagnostic accuracy of the models developed in (1) using blood RNA sequencing and routine electronic health record data, in a prospective cohort of patients with active haematological malignancy on myelosuppressive treatment presenting with undifferentiated fever or ARI to emergency departments at BioAID participating sites (n=100). I will recruit adult participants from the intervention (neuraminidase inhibitor) and control (placebo) arms of the multicentre REMAP-CAP and RECOVERY trials for influenza at participating BioAID sites (n=120). I will measure MX-1 transcript levels (by NanoString quantification) and protein levels (by FebriDx point-of-care tests) to evaluate change in nasopharyngeal PCR viral load at day 3 by trial arm and MX-1 status. Impact: This work will produce a diagnostic model that could act as a triage test for viral ARI, improving diagnostic yield and rationalising confirmatory molecular testing. It will also provide proof-of-concept evidence of a host biomarker-stratified antiviral treatment approach that could then be evaluated in a definitive trial. I will develop skills in advanced epidemiological analysis, prediction modelling and machine learning to further my development as a future research leader in data science for infectious diseases.