Fig. 3

The early predictive model to identify the OSI-trajectory phenotypes related to the different prognoses. a Six distinctive characteristics of patients with worse prognosis (the high-level stable and ascending phenotypes) and better prognosis (the descending and low-level stable phenotypes) were selected using LASSO regression analysis. b, c MIMIC-IV and EICU-CRD database was used as training and validation group, respectively. The ROC curve, calibration curve and DCA analysis demonstrated that the predictive model exhibited strong ability of discrimination, calibration and clinical utilization. d The nomogram was plotted to better visualize our predictive model. OSI: oxygen saturation index; LASSO: The Least Absolute Shrinkage and Selection Operator; MIMIC-IV: the Medical Information Market for Intensive Care IV; EICU-CRD: the eICU Collaborative Research Database; ROC: receiver operating characteristic; DCA: decision curve analysis; BMI: body mass index; NMBAs: neuromuscular blockades; APS-III score: acute physiology score III