SARS-CoV-2 severity prediction in young adults using artificial intelligence

Abstract

Aim: to build a predictive model for severe COVID-19 prediction in young adults using deep learning methods. Materials and methods: data from 906 medical records of patients aged 18 to 44 years with laboratory-confirmed SARS-CoV-2 infection during 2020–2021 period was analyzed. Evaluation of laboratory and instrumental data was carried out using the Mann-Whitney U-test. The level of statistical significance was p≤0,05. The neural network was trained using the Pytorch framework. Results: in patients with mild to moderate SARS-CoV-2 infection, peripheral oxygen saturation, erythrocytes, hemoglobin, total protein, albumin, hematocrit, serum iron, transferrin, and absolute peripheral blood eosinophil and lymphocyte counts were significantly higher than in patients with severe СOVID-19 (p< 0,001). The values of the absolute number of neutrophils, ESR, glucose, ALT, AST, CPK, urea, LDH, ferritin, CRP, fibrinogen, D-dimer, respiration rate, heart rate, blood pressure in the group of patients with mild and moderate severity were statistically significantly lower than in the group of severe patients (p < 0.001). Eleven indicators were identified as predictors of severe COVID-19 (peripheral oxygen level, peripheral blood erythrocyte count, hemoglobin level, absolute eosinophil count, absolute lymphocyte count, absolute neutrophil count, LDH, ferritin, C-reactive protein, D-dimer levels) and their threshold values. A model intended to predict COVID-19 severity in young adults was built. Conclusion. The values of laboratory and instrumental indicators obtained in patients with SARS-CoV-2 infection upon admission significantly differ. Among them eleven indicators were significantly associated with the development of a severe COVID-19. A predictive model based on artificial intelligence method with high accuracy predicts the likelihood of severe SARS-CoV-2 course development in young adults.