Fig. 1

The OSI-trajectory phenotypes of patients with IMV treatment in ICU. As a finite mixture model, group-based trajectory model (GBTM) could identify the distinct phenotypes that follow particular trajectories. In this work, based on the OSI trajectories, four distinct phenotypes were identified: phenotype 1 (low-level stable), in which the OSI remained consistently stable at low levels; phenotype 2 (ascending), in which the OSI gradually increased over time; phenotype 3 (descending), in which the OSI gradually decreased over time; and phenotype 4 (high-level stable), in which the OSI remained consistently stable at high levels. This trend has been observed in the MIMIC-IV dataset, EICU-CRD dataset, as well as in the combined dataset of the two databases. Phenotype 4 had almost 1.4 times higher odds ratio of mortality compared to the phenotype 1 in the doubly robust estimation when controlling unbalanced covariates. a The OSI-trajectory phenotypes in the combined dataset. b The OSI-trajectory phenotypes in MIMIC-IV dataset. c The OSI-trajectory phenotypes in EICU-CRD dataset. OSI: oxygen saturation index; ICU: intensive care medicine; ARDS: acute respiratory distress syndrome; APS-III score: Acute Physiology Score III; MIMIC-IV: the Medical Information Market for Intensive Care IV; EICU-CRD: the eICU Collaborative Research Database