Support for the development of this article was provided by Sight Diagnostics
CURIAL, an AI-driven triage tool developed independently by the University of Oxford, leverages routine clinical data – such as vital signs and complete blood count – collected at the point of care to rule out COVID-19 within an hour of patients arriving at a hospital emergency department
In emergency departments (EDs), rapid and accurate COVID-19 screening is critical for effective infection control and ensuring timely – and appropriate – delivery of care for patients. With the pandemic relentlessly exerting an unprecedented burden on the healthcare system worldwide, this call for fast identification of COVID-19 at the point of care is more urgent than ever.
Currently, the nasopharyngeal swab RT-PCR test – the gold standard for SARS-CoV-2 testing – has a few limitations, including a long turnaround time (12–24 hours), limited sensitivity and requiring laboratory infrastructure.1 The rapid lateral flow antigen detection test yields results faster, but has other limitations such as poor sensitivity. For example, in a study in Liverpool, the overall sensitivity of the rapid test was only 40.0%, i.e., the test only detected four in 10 people who tested positive by PCR.2 And due to this risk of false results and imprecisions in the manufacturer’s accuracy claims, the US Food and Drug Administration recalled and withdrew the rapid tests from sale in the US in June 2021.3
Because it is vital to make quick decisions about a patient’s care pathway (admission, treatment approach, discharge, etc.) while keeping the hospital environment safe, there is a system-wide operational and safety impact when effectively performing front-door COVID-19 triage in the ED.
Leveraging artificial intelligence (AI) for faster COVID-19 screening
The CURIAL algorithm – developed by infectious disease and machine learning experts at Oxford University – is a viable alternative to traditional testing that can successfully rule out COVID-19 within the first hour of a patient’s arrival at an emergency department. The CURIAL AI algorithm uses the data from routine CBCs, urea and electrolyte tests, vital signs, liver function tests, C-reactive protein (CRP) tests and blood gas to predict the probability of a patient testing positive for COVID-19.4
Therefore, to combat the transmission of SARS-CoV-2 while providing high-quality care to patients – all within a reasonable timeframe – CURIAL can be an effective, and faster, solution to triage patients for COVID-19 in the ED setting.
CURIAL data and model accuracy
CURIAL developed two models to predict COVID-19 in patients: The ED model (for all patients visiting the ED) and the admissions model (for subsequently admitted patients). The training data, gathered from four teaching hospitals in Oxfordshire, assessed 155,689 adult patients at Oxford University Hospitals between 1 December 2017 and 19 April 2020. When calibrated during training to a sensitivity of 80% – the ED model attained 77.4% sensitivity and 95.7% specificity for identifying COVID-19 patients among all patients presenting to hospital. And correspondingly – with the same calibration – the admissions model achieved 77.4% sensitivity and 94.8% specificity. In terms of negative predictive values (NPVs) – the probability that patients with a negative result truly don’t have COVID-19 – both models achieved high NPVs (>98·5%) across a range of prevalences (≤5%), allowing for a quick and reliable rule-out.
During a real-world evaluation of the CURIAL models over a two-week test period (20 April–6 May 2020) in Oxford University Hospitals’ EDs, CURIAL displayed impressive accuracy. The ED model (3326 patients) achieved 92.3% accuracy (NPV 97.6%), and the admissions model (1715 patients) achieved 92.5% accuracy (NPV 97.7%) in comparison to PCR results.4
Using point of care diagnostics alongside AI
To understand which individual features had the most significant influence on model predictions, CURIAL ran a relative feature importance analysis. For both models, eosinophils and basophils had most significant effects on model performance,4 meaning complete blood counts directly impact the CURIAL algorithm’s success.
As effective AI-powered COVID-19 screening relies on time-sensitive and accurate CBC results, the University of Oxford researchers created a version of CURIAL, called CURIAL-Rapide, that leverages only CBC results and vital signs to screen for COVID-19 in patients. Consequently, CURIAL-Rapide creates a new collaborative pathway where point of care CBC analysers – such as Sight Diagnostics’ OLO haematology analyser – are used in conjunction with CURIAL to potentially further expedite the overall screening turnaround time. And when the analyser can provide rapid results with lab-grade accuracy – OLO produces accurate results in approximately ten minutes – it can play a pivotal role in ensuring a safer hospital environment by triaging COVID-19 patients efficiently. For example, by reconfiguring the care pathway and removing the time and logistical constraint of using conventional lab infrastructure, hospitals can create separate areas (hot-labs) for CBC testing before patients enter EDs to reduce operational strain and staff work time while keeping the hospital safe from COVID-19 transmission.
To substantiate this innovative pathway and the CURIAL-Rapide algorithm, the University of Oxford deployed OLO analysers at the John Radcliffe Hospital in February 2021 to power lab-free screening. The study’s interim evaluation was successfully completed, and results will be published soon.
Having accurate CBC results in minutes, from OLO, would help CURIAL-Rapide make predictions even sooner, potentially reducing care delays and supporting infection control within hospitals. Our goal is to get the right treatment to patients sooner by helping rule out COVID at triage for a majority of patients who don’t have the infection. This project shows that artificial intelligence can work with rapid diagnostics to help us select the best care pathways and minimise risks of spreading the infection in hospitals.5
Andrew Soltan, academic clinician and machine learning researcher at Oxford University
Sight OLO® performs with high accuracy for all CBC parameters
Sight OLO haematology analyser streamlines the typical blood staining workflow while maintaining lab-grade accuracy. Through a quick finger prick and the culmination of cutting-edge innovations in physics, optics, sample preparation, and an AI-based computer vision algorithm, the self-contained quantitative multi-parameter analyser can deliver fast and accurate CBC results within minutes in point of care settings.
During a recent study, the accuracy of OLO was compared with the Sysmex XN-1000 System. Samples – covering a broad clinical range for each tested parameter – from 355 males (52%) and 324 females (48%) aged 3 months to 94 years were analysed. The regression analysis results showed a consonance in correlation coefficient and slope, bias and intercept between OLO and Sysmex XN. Therefore, the study concluded that OLO performs with high accuracy for all CBC parameters,6 thus, making OLO the perfect partner to use alongside an AI COVID-19 screening initiative, such as CURIAL.
References
- Tang Y et al. The laboratory diagnosis of COVID-19 infection: current issues and challenges. J Clin Microbiol 2020;58:1–9.
- Taylor-Phillips S, Dinnes J. Asymptomatic rapid testing for SARS-CoV-2. BMJ 2021;374:n1733.
- US Food and Drug Administration. Stop Using Innova Medical Group SARS-CoV-2 Antigen Rapid Qualitative Test: FDA Safety Communication 2021. www.fda.gov/medical-devices/safety-communications/stop-using-innova-medical-group-sars-cov-2-antigen-rapid-qualitative-test-fda-safety-communication (accessed Oct 2021).
- Soltan AAS et al. Rapid triage for COVID-19 using routine clinical data for patients attending hospital: development and prospective validation of an artificial intelligence screening test. Lancet Digit Health 2021;3(2):e78–87.
- Mageit S. Oxford University Hospital deploys blood analyser as part of COVID screening [internet]. Mobi Health News 2021; March 15. www.mobihealthnews.com/news/emea/oxford-university-hospital-deploys-blood-analyser-part-covid-screening (accessed Oct 2021).
- Bachar N et al. An artificial intelligence-assisted diagnostic platform for rapid near-patient haematology. AJH 2021;96(10):1264–74.
Learn more about CURIAL’s rapid identification of COVID-19 and CBC results’ role in improving ED patient flow in a Sight Diagnostics CURIAL webinar. Register at www.sightdx.com/events/curial-rapide to watch the full webinar.
