A machine learning COVID-19 mass screening based on symptoms and a simple olfactory test

Abstract

Abstract The early detection of symptoms and rapid testing are the basis of an efficient screening strategy to control COVID-19 transmission. The olfactory dysfunction is one of the most prevalent symptom and in many cases is the first symptom. This study aims to develop a machine learning COVID-19 predictive tool based on symptoms and a simple olfactory test, which consists of identifying the smell of an aromatized hydroalcoholic gel. A multi-centre population-based prospective study was carried out in the city of Reus (Catalonia, Spain). The study included consecutive patients undergoing a reverse transcriptase polymerase chain reaction test for presenting symptoms suggestive of COVID-19 or for being close contacts of a confirmed COVID-19 case. A total of 519 patients were included, 386 (74.4%) had at least one symptom and 133 (25.6%) were asymptomatic. A classification tree model including sex, age, relevant symptoms and the olfactory test results obtained a sensitivity of 0.97 (95% CI 0.91–0.99), a specificity of 0.39 (95% CI 0.34–0.44) and an AUC of 0.87 (95% CI 0.83–0.92). This shows that this machine learning predictive model is a promising mass screening for COVID-19.