In this prospective cohort study, researchers used electronic healthcare records (EHR) of 155,394 adults in Oxfordshire, UK, to develop a rapid triage system to screen hospital patients for COVID-19. Data used from EHRs included vital signs, blood gas testing and presentation blood tests. Screening results were compared to polymerase chain reaction (PCR) test results for SARS-CoV-2. Machine learning was used to develop both linear and non-linear models, and test sets were generated to simulate the progression of a pandemic by varying the prevalence of COVID-19 in the test samples. A model generated for screening all patients visiting the hospital achieved 77.4% sensitivity and 95.7% specificity (emergency department model), while a different model evaluating those actually admitted into the hospital achieved 77.4% sensitivity and 94.8% specificity (admissions model). In a 2-week validation study, the emergency department model achieved 92.3% accuracy and the admissions model demonstrated 92.5% accuracy. The authors stress that the clinical data used in the model can be obtained within one hour of arrival at the hospital via routine tests that are conducted within usual care to effectively triage patients while awaiting PCR test results which can take up to 72 hours to develop. The proposed triage tool using artificial intelligence can be easily implemented in hospitals in middle- and high-income countries.