Real-world evaluation of AI driven COVID-19 triage for emergency admissions: External validation & operational assessment of lab-free and high-throughput screening solutions
Soltan A., Yang J., Pattanshetty R., Novak A., YANG Y., Rohanian O., Beer S., Soltan M., Thickett D., Fairhead R., Collaborative CURIALT., ZHU T., EYRE D., CLIFTON D.
Background: Uncertainty in patients’ COVID-19 status contributes to treatment delays, nosocomial transmission, and operational pressures in hospitals. However, typical turnaround time for laboratory PCR remains 12-24h and lateral flow testing (LFD) has limited sensitivity. We recently demonstrated that AI-driven triage (CURIAL-1.0) provides rapid COVID-19 screening using clinical data routinely available within 1h of arrival to hospital. Here we aimed to improve time-to-result, perform external & prospective validation, and deploy a novel lab-free screening tool in a UK emergency department. Methods: We eliminated weakly-informative predictors to improve generalisability and speed, developing CURIAL-Lab with vital signs and readily available blood tests (FBC, U&E, LFT, CRP) and CURIAL-Rapide with vital signs and FBC alone. Models were validated externally for emergency admissions to University Hospitals Birmingham, Bedfordshire Hospitals & Portsmouth Hospitals University NHS trusts, and prospectively at Oxford University Hospitals, by comparison to confirmatory nucleic acid testing. Next, we compared model performance with LFDs and evaluated a combined pathway triaging patients to COVID-19-suspected clinical areas where either model prediction or LFD results were positive. Lastly, we deployed CURIAL-Rapide alongside an approved point-of-care FBC analyser to provide lab-free COVID-19 screening at the John Radcliffe Hospital (Oxford, UK). Our primary improvement outcome was time-to-result. Findings: 72,223 patients met eligibility criteria across four validating hospital groups. CURIAL-Lab & CURIAL-Rapide performed consistently (AUROC range 0.858-0.881 & 0.836-0.854), achieving highest sensitivity at Portsmouth Hospitals (84.1% [Wilson’s 95% CIs 82.5-85.7] & 83.5% [81.8-85.1]) at specificities of 71.3% (70.9-71.8) and 63.6% (63.1-64.1). When combined with LFDs, models predictions improved triage sensitivity from 56.9% for LFDs alone to 85.6% with CURIAL-Lab (81.6-88.9; AUROC 0.925) and 88.2% with CURIAL-Rapide (84.4-91.1; AUROC 0.919), thereby reducing missed COVID-19 cases by 65% & 72%. In prospective deployment of CURIAL-Rapide, 520 patients were enrolled for point-of-care FBC analysis, of whom 436 received confirmatory PCR testing and 10 (2.3%) tested positive. Median time from arrival to CURIAL-Rapide result was 45:00min (32-64), 16:00min (26.3%) sooner than LFD (61:00min, 37-99; log-rank p<0.0001), and 6:52h (90.2%) sooner than PCR (7:37h, 6:05-15:39; p<0.0001). Classification performance was high, with sensitivity of 87.5% (52.9-97.8), specificity of 85.4% (81.3-88.7) and NPV 99.7% (98.2-99.9). CURIAL-Rapide correctly excluded infection for 58.5% of patients who were triaged by a physician to a ‘COVID-19-suspected’ area but went on to test negative by PCR. Impact: Our findings demonstrate the generalisability, performance and real-world operational benefits of AI driven screening for COVID-19 over standard-care in emergency departments. CURIAL-Rapide provides rapid, lab-free screening when used with near-patient FBC analysis, and can reduce the number of COVID-19- negative patients triaged to COVID-19-suspected areas.