Acute respiratory distress syndrome (ARDS) is a life-threatening condition. Accurate judgement of the disease progression is essential for controlling the condition in ARDS patients. We investigated whether changes in the level of serum sRAGE/esRAGE could predict the 28-day mortality of ICU patients with ARDS. A total of 83 ARDS patients in the ICU of the Second Affiliated Hospital of Nantong University from January 2021 to June 2022 were consecutively enrolled in this study. Demographic data, primary diagnosis and comorbidities were obtained. Multiple scoring systems, real-time monitoring systems, and biological indicators were determined within 6 h of admission. The clinical parameters for survival status of the ARDS patients were identified by multivariate logistic regression. Receiver operating characteristic (ROC) curve analysis was employed to verify the accuracy of the prognosis of the related parameters. The admission level of sRAGE was significantly higher in the nonsurvival group than in the survival group (p < 0.05), whereas the serum esRAGE level showed the opposite trend. Multivariate logistic regression analysis showed that sRAGE (AUC 0.673, p < 0.05), esRAGE (AUC 0.704, p < 0.05), and ELWI (extravascular lung water index) (AUC 0.717, p < 0.05) were independent risk factors for the prognosis of ARDS. Model B (ELWI + esRAGE) could not be built as a valid linear regression model (ELWI, p = 0.079 > 0.05). Model C (esRAGE + sRAGE) was proven to have no significance because it had a predictive value similar to that of the serum levels of esRAGE (Z = 0.993, p = 0.351) or sRAGE (Z = 1.116, p = 0.265) alone. Subsequently, Model D (sRAGE + esRAGE + ELWI) showed the best 28-day mortality predictive value with a cut-off value of 0.426 (AUC 0.841; p < 0.001), and Model A (sRAGE + ELWI) had a cut-off value of 0.401 (AUC 0.820; p < 0.001), followed by sRAGE (AUC 0.704, p = 0.004), esRAGE (AUC 0.717, p = 0.002), and ELWI (AUC 0.637, p = 0.028). In addition, there was no statistically significant difference between Model A and Model D (Z = 0.966, p = 0.334). The admission level of sRAGE was higher in the nonsurvival group, while the serum esRAGE level showed the opposite trend. Model A and Model D could be used as reliable combined prediction models for predicting the 28-day mortality of ARDS patients.