Novel Phenotypes of Acute Respiratory Failure and Differential Response to Awake Prone Positioning: A Multi‐Cohort Study
Nan Shi, Ruiqiang Zheng, Xufeng Chen, Huiying Zhao, Jun Jin, Changsong Wang, Shulin Xiang, Man Huang, Hongsheng Zhao, Yi Wang, Ruixuan Yu, Qin Sun, Hui Chen, Jianfeng Xie, Songqiao Liu, Yi Yang, Ling Liu, Haibo Qiu,ABSTRACT
The heterogeneity of acute respiratory failure (ARF) in non‐intubated patients remains poorly defined. This study aimed to identify novel phenotypes in non‐intubated ARF patients and evaluate their prognostic and predictive ability. We analyzed data from three cohorts: a training cohort from the Chinese Database in Intensive Care, validation cohort A from a randomized controlled trial, and validation cohort B from a multicenter dataset. Latent class analysis and k ‐means clustering were applied independently across cohorts to identify phenotypes. A two‐class model demonstrated the optimal fit in the training cohort ( n = 706). Compared to Phenotype 1, Phenotype 2 exhibited worse oxygenation, higher lactate concentrations, and more severe coagulation dysfunction, along with significantly higher intubation (56.7 vs. 28.7%, p < 0.001) and mortality rates within 28 days (22.0 vs. 9.7%, p < 0.001). Similar phenotypes were verified in validation cohorts A ( n = 409) and B ( n = 609). Notably, Phenotype 1 patients benefited most from prolonged awake prone positioning (APP), with a significant interaction observed ( p = 0.035; adjusted p = 0.048). Two parsimonious models were further developed for phenotype assignment prediction. The two novel phenotypes with both prognostic and predictive ability may help tailor APP to individual non‐intubated ARF patients.