A prediction model for 30-day mortality in patients with ARDS admitted to the intensive care unit

The development of acute respiratory distress syndrome (ARDS) is intricate and uncertain. Although intensive care has made substantial progress in the past several decades, the in-hospital mortality rate of patients with ARDS remains as high as 40%, which imposes a large burden on hospitals and intensive care. This study aimed to develop and validate a nomogram to predict the 30-day mortality of patients with ARDS admitted in the intensive care unit (ICU).

Data of 4920 patients with ARDS were collected from the MIMIC-IV database, as well as data of 248 patients were collected from the Affiliated Hospital of Southwest Medical University. After processing these data, we performed correlation analysis between various types of variables and plotted a heat map to visualize the significance of these correlations. Subsequently, LASSO regression was used to initially screen for risk factors strongly associated with 30-day death in patients with ARDS, and a prediction model was established by multivariate logistic regression. The predictive efficacy of this model was preliminarily evaluated; internally validated; and compared with that of SOFA, APS-III, OASIS, and SAPS-II scores. In addition, it was externally validated using patient data from the ICU in China.

The AUC of the model was 0.78; The AUC values in the internal validation and external validation sets were 0.805 and 0.742, respectively. Moreover, the model demonstrated better predictive efficacy than this three traditional disease severity scoring systems (OASIS, SAPS-II, APS-III, and SOFA), highlighting its good predictive value.

We established a 30-day mortality risk prediction model for patients with ARDS who were first admitted to ICU by simple and easily accessible clinical data. To enhance real-world utility, we engineered an open-access mobile application that provides instant risk stratification at the bedside. This model synergizes with existing ICU scoring systems by enabling early identification of high-risk patients during initial assessment, thereby guiding timely interventions.