Artificial Intelligence Screens for COVID-19 26% Faster Than Lateral Flow Tests

This article is based on research findings that are yet to be peer-reviewed. Results are therefore regarded as preliminary and should be interpreted as such. Find out about the role of the peer review process in research here. For further information, please contact the cited source.

As society transitions to "living with COVID-19", having access to both efficient and accurate screening tools is integral. In hospital environments where dozens of patients and staff cross paths each day, we must be able to quickly distinguish between patients that are presenting with COVID-19 and those that are not to ensure safety and infection control. 

CURIAL-Rapide is an artificial intelligence (AI)-based screening test that has been developed by researchers at the University of Oxford. It is an updated form of the CURIAL-1.0 system that was originally tested and released in 2020, and utilizes data from clinical tests performed at the patient's bedside to screen for COVID-19.

In a new preprint, the research team behind CURIAL-Rapide share their data from a real-world study that aimed to determine the logistical and safety performance of CURIAL-Rapide vs standard-care practisSes. Technology Networks spoke with Dr. Andrew Soltan, NIHR academic clinical fellow at the John Radcliffe Hospital and first author of the study, to learn more about the AI test and why it is faster than PCR or lateral flow tests. 

Molly Campbell (MC): Why do we need other methods – in addition to lateral flow tests – to screen for COVID-19, particularly in clinical settings?

Andrew Soltan (AS): Although turnaround times for the gold-standard test – the PCR – have come down during the pandemic, these typically still require 12-24 hours to return in larger UK hospitals, and much longer in more remote areas. This can delay care, prevent patients from being transferred to wards and ultimately increase strain on the hospital.

Lateral flow tests have been adopted in emergency departments to try and solve this problem, but this has been fraught with problems because of poor performance. An evaluation at Oxford University Hospitals, our group of hospitals in Oxfordshire, showed that lateral flow tests gave negative results for almost half (43%) of all COVID-19 patients being admitted to our hospitals.

As patients in hospital are amongst the most vulnerable groups, the poor performance of lateral flow tests potentially places patients at risk – for example, if a patient with COVID-19 was unintentionally transferred to a ward of uninfected patients.

MC: The CURIAL-Rapide AI screening test was developed using routine healthcare data from electronic health records for 115,394 patients and 72,310 admissions. Can you please expand on this?

AS: We developed CURIAL by teaching it to discriminate patients with COVID-19 from patients who did not have COVID-19, but instead a broad spectrum of alternative conditions. To do this, we needed to make sure that the data used to teach the AI was accurately labeled, but problems with testing during the first wave meant this was not straightforward.

We addressed this by training the AI using patients who were certain to have COVID-19 that had tested positive by PCR, and patients certain to not have COVID-19, by looking back to before the pandemic and using a broad cohort of over one hundred thousand "pre-pandemic" patients. This allowed us to have breadth of presentations and confidence in labelling. With CURIAL-Rapide, we built on the initial model by optimizing the input tests required, cutting the time needed for a result.

Next, we wanted to test (or validate) whether the AI performed well enough. We did this by collaborating with three other NHS trusts and looking at how CURIAL-Rapide performed in their emergency departments. We also tested CURIAL-Rapide for all patients coming to our hospital during the second wave in the UK. In every evaluation, we found that the AI performed consistently and reliably. When we compared the AI to lateral flow tests, we found that it was much more effective at identifying patients being admitted with COVID-19.

MC: Can you describe the structure of the three-month evaluation study? Were there any limitations?

AS: We deployed CURIAL-Rapide at our hospital’s emergency department (the John Radcliffe Hospital in Oxford) in February 2021 and assessed how the test performed and how long it took for results to come back.

We measured how long it took for AI results, lateral flow test results and PCR results to come back for our patients, and found that CURIAL-Rapide results were available 26% sooner than lateral flow tests.

In addition to this, we evaluated performance of the AI for all patients admitted to three other NHS trusts (University Hospitals Birmingham NHS Foundation Trust, Bedfordshrie Hospitals NHS Foundation Trust and Portsmouth University Hospitals NHS Trust), and found that the AI was consistent across the trust.

Our study was UK-based and we therefore cannot extrapolate our evaluation to other countries. We also do not yet know how CURIAL would fit in as a decision-support aid – i.e., how clinicians would interact with the test, and whether it would help clinicians make better decisions overall. A randomized controlled trial would give the very best evidence, however the study presented here gives an accurate report of how CURIAL-Rapide performed in the real-world and is therefore still significant.

MC: How does the cost of an AI-informed screening test compare with tests that rely on biological markers?

AS: CURIAL uses clinical data that is already collected for virtually all patients being admitted to hospital – and therefore requires no extra tests to be performed. This means that AI-based screening can be performed at essentially no cost, and with no waste.

MC: What are the next steps for CURIAL-Rapide?

AS: We are applying for further funding to roll out CURIAL-Rapide and CURIAL-Lab across the NHS. We would be particularly interested to hear from doctors and healthcare workers abroad who might be interested in using CURIAL-Rapide in their practice.

Dr. Andrew Soltan was speaking to Molly Campbell, Science Writer for Technology Networks. 

Reference: Soltan AAS, Yang J, Pattanshetty R, et al. Real-world evaluation of AI-driven COVID-19 triage for emergency admissions: External validation & operational assessment of lab-free and high-throughput screening solutions. medRxiv. 2021. doi: 10.1101/2021.08.24.21262376.