Additive Value of Computed Tomography Severity Scores to Predict Lengths of Stay in Hospital and ICU for Covid-19 Patients: A Machine Learning Study

Mikaeil  Molazadeh; Seyed Salman  Zakariaee; Hossein  Salmanipour; Negar  Naderi

doi:10.18502/jbe.v10i4.18528

Mikaeil Molazadeh Department of Medical Physics, Faculty of Medicine, Tabriz University of Medical Sciences, Tabriz, Iran.
Seyed Salman Zakariaee Department of Medical Physics, Faculty of Paramedical Sciences, Ilam University of Medical Sciences, Ilam, Iran.
Hossein Salmanipour Department of Radiology, Faculty of Medicine, Ilam University of Medical Sciences, Ilam, Iran.
Negar Naderi Department of Midwifery, Faculty of Nursing and Midwifery, Ilam University of Medical Sciences, Ilam, Iran.

DOI: https://doi.org/10.18502/jbe.v10i4.18528

Keywords: Chest CT severity score; COVID-19; CT-SS; Machine learning; Length of stay.

Abstract

Introduction: During the outbreak of COVID-19, most hospitals faced resource shortages due to the greatsurges in the influx of infected COVID-19 patients and demand exceeding capacities. Predicting the lengths ofstay (LOS) of the patients can help to make proper resource-planning decisions. CT-SS accurately determinesthe disease severity and could be considered an appropriate prognostic factor to predict patients’ LOS.In this study, we evaluate the additive value of CT-SS in the prediction of hospital and ICU LOSs of COVID-19patients.

Methods: This single-center study retrospectively reviewed a hospital-based COVID-19 registry database from 6854 cases of suspected COVID-19. Four well-known ML classification models including kNN, MLP, SVM, and C4.5 decision tree algorithms were used to predict hospital and ICU LOSs of COVID-19 patients. The confusion matrix-based performance measures were used to evaluate the classification performances of the ML algorithms.

Results: For predicting hospital LOS, the kNN model with an accuracy of 77.1%, sensitivity of 100.0%, precision of 68.6%, specificity of 54.2%, and AUC of around 99.4% had the best performance among the other three ML techniques. This algorithm with 94.4% sensitivity, 74.6% specificity, 84.5% accuracy, 78.8% precision, 85.9% F-Measure, and an AUC of 95.3% had also the best performance for predicting ICU LOS of the patients.

Conclusion: The performances of the ML predictive models for predicting hospital and ICU LOSs of COVID-19 patients were improved when CT-SS data was integrated into the input dataset.