Abstract
Background: The COVID-19 pandemic has been considered as the great threat to global public health. We aimed to clarify the risk factors associated with the development of acute respiratory distress syndrome (ARDS) and progression from ARDS to death and construct a risk prediction model.Methods: In this single-centered, retrospective, and observational study, 796 COVID-19 patients developed ARDS and 735 COVID-19 patients without ARDS were matched by propensity score at an approximate ratio of 1:1 based on age, sex and comorbidities. Demographic data, symptoms, radiological findings, laboratory examinations, and clinical outcomes were compared between with or without ARDS. Univariable and multivariable logistic regression models were applied to explore the risk factors for development of ARDS and progression from ARDS to death and establish a comprehensive risk model. Results: Higher SOFA, qSOFA, APACHE II and SIRS scores, elevated inflammatory cytokines, dysregulated multi-organ damage biomarkers, decreased immune cell subsets were associated with higher proportion of death (34.17% vs 1.22%; P<0.001) and increased risk odds of death (OR=57.216, 95%CI=28.373-115.378; P<0.001) in COVID-19 patients with ARDS. In addition to previous reported risk factors related to ARDS development and death, such as neutrophils, IL-6, D-Dimer, leukocytes and platelet, we identified elevated TNF-α (OR=1.146, 95%CI=1.100-1.194; P<0.001), CK-MB (OR=1.350, 95%CI=1.180-1.545; P<0.001), declined ALB (OR=0.834, 95%CI=0.799-0.872; P<0.001), CD8+ T cells (OR=0.983, 95%CI=0.976-0.990; P<0.001) and CD3-CD19+ B cells (OR=0.992, 95%CI=0.988-0.997; P=0.003) as novel risk factors. Most importantly, the predictive accuracy of the combined model integrating four score systems and these risk factors demonstrated highest among all models for the development of ARDS (AUC= 0.904) and the progression from ARDS to death (AUC= 0.959).Conclusion: COVID-19 patients with ARDS were more likely to develop into death. The potential risk factors and the comprehensive prediction model could be helpful to identify patients developed ARDS with poor prognosis at an early stage, which might help physicians to formulate a timely therapeutic strategy.
Semi-supervised deep learning from time series clinical data for acute respiratory distress syndrome prediction: model development and validation study (Preprint)