The study developed a nomogram combining initial SOFA1 score, dynamic change in ΔSOFA 3–1, and age to predict 28-day mortality in ICU patients. Overall 28-day mortality was 18.9% and increased with higher SOFA1 scores (stratified into 4–7, 8–11, ≥12) and ΔSOFA 3–1. The nomogram achieved high discrimination with a C-index of 0.852 for SOFA1 4–7 and 0.845 for SOFA1 8–11, as well as good calibration. The XGBoost model with the same predictors had an AUC of 0.833 in the internal training cohort, 0.863 in the internal testing cohort, and 0.671 in the external MIMIC-IV validation cohort. SHAP analysis showed ΔSOFA 3–1 as the most influential predictor. Dynamic changes of ΔSOFA 3–1 especially in patients with intermediate SOFA1 (4–11) improve prognostic accuracy in combination with age. The nomogram serves as an intuitive tool for risk stratification, XGBoost demonstrates the potential of machine learning, but external validation emphasizes the need for multicenter studies.