Predicting ICU Length of Stay in COVID-19 Patients Using a Multivariable Model Incorporating Clinical, Laboratory and Imaging Features

Objective: To predict ICU Length of Stay (LOS) using a multivariable model incorporating clinical, and laboratory and imaging features in hospitalized COVID-19 patients, thereby stratifying patients and allocating resources accordingly. Methods: In this retrospective cohort study, 139 hospitalized patients (aged between 3 to 99) with rRT-PCR confirmed COVID-19 pneumonia requiring intensive care, who had been discharged or deceased, were enrolled. Demographic, clinical, and laboratory findings of eligible patients were all extracted from electronic medical records and, if needed, through phone calls. Semi-quantitative CT Severity Score (CTSS) was calculated and assigned to each encoded patient independently and blindly. We used cox regression model to investigate the prognostic role of semi-quantitative CTSS, clinical and laboratory features to anticipate ICU-LOS. Results: 139 patients with rRT-PCR confirmed COVID-19 pneumonia (including 60 females and 79 males) with a mean age of 58.52 ± 20.58 (ranging from 3 to 99) were included. CTSS was not predictive of ICU-LOS. Additionally, CTSS of more than 11 was predictor of mortality (sensitivity, 60.3%; specificity, 58%; AUC, 0.605; 95% confidence interval, 0.508 - 0.702; P-value, 0.034), and CTSS of above 10 was predictor of oxygen therapy dependency (sensitivity, 70.2%; specificity, 68%; AUC, 699/0; 95% confidence interval, 0.580 - 0.818; P-value, 0.002). CTSS was not significantly associated with respiratory rate and on-admission dyspnea, while it was inversely related to air-room SpO2 on the first day of admission (P < 0.0001, r = -0.341). Conclusion: CTSS is capable of anticipating mortality rate and the chance of undergoing supportive oxygen therapy during ICU hospitalization, while it does not predict ICU-LOS, rate of mechanical ventilation, or corticosteroid therapy.