A Prediction Model for Clinical Outcomes of COVID-19 Hospitalized Patients: Construction and Accuracy Assessment.

Background: The coronavirus disease 2019 (COVID-19) pandemic spread rapidly with considerable morbidity nationwide since China's liberalization in December 2022. Our work has focused on identifying different predictive factors from the laboratory examination of critically ill patients, and forecasting the unfavorable outcome of critically ill patients with COVID-19 through a combined diagnosis of biological markers.
Methods: We conducted a retrospective study at the Department of First Affiliated Hospital of Wenzhou Medical University, China, from December 24, 2022, to January 10, 2023, where 434 critically ill patients who met the inclusion criteria were involved. Machine analysis was employed to search for the parameters with the highest predictive value to calculate COVID-19 mortality by exploiting 66 typical laboratory results.
Results: Combined diagnosis of serum albumin (ALB), lactate dehydrogenase (LDH), direct bilirubin (Dbil), ferritin, pulse oxygen saturation (SpO2), and neutrophil count (NEUT#) was evaluated, and the result with the highest predictive value (NEUT#) was selected as the predictor for COVID-19 mortality with a sensitivity of 89.2% and a specificity of 77.4%.
Conclusions: The increased levels of LDH, Dbil, ferritin, and NEUT#, along with lowered ALB and SpO2 levels are the most decisive variables for forecasting the mortality for COVID-19 according to our machine-learning-based model. The combined diagnosis could be used to improve further diagnostic performance.