Background
Heterogeneity among critically ill patients undergoing invasive mechanical ventilation (IMV) treatment could result in high mortality rates. Currently, there are no well-established indicators to help identify patients with a poor prognosis in advance, which limits physicians’ ability to provide personalized treatment. This study aimed to investigate the association of oxygen saturation index (OSI) trajectory phenotypes with intensive care unit (ICU) mortality and ventilation-free days (VFDs) from a dynamic and longitudinal perspective.
Methods
A group-based trajectory model was used to identify the OSI-trajectory phenotypes. Associations between the OSI-trajectory phenotypes and ICU mortality were analyzed using doubly robust analyses. Then, a predictive model was constructed to distinguish patients with poor prognosis phenotypes.
Results
Four OSI-trajectory phenotypes were identified in 3378 patients: low-level stable, ascending, descending, and high-level stable. Patients with the high-level stable phenotype had the highest mortality and fewest VFDs. The doubly robust estimation, after adjusting for unbalanced covariates in a model using the XGBoost method for generating propensity scores, revealed that both high-level stable and ascending phenotypes were associated with higher mortality rates (odds ratio [OR]: 1.422, 95% confidence interval [CI] 1.246–1.623; OR: 1.097, 95% CI 1.027–1.172, respectively), while the descending phenotype showed similar ICU mortality rates to the low-level stable phenotype (odds ratio [OR] 0.986, 95% confidence interval [CI] 0.940–1.035). The predictive model could help identify patients with ascending or high-level stable phenotypes at an early stage (area under the curve [AUC] in the training dataset: 0.851 [0.827–0.875]; AUC in the validation dataset: 0.743 [0.709–0.777]).
Conclusions
Dynamic OSI-trajectory phenotypes were closely related to the mortality of ICU patients requiring IMV treatment and might be a useful prognostic indicator in critically ill patients.