Abstract: Background　Staphylococcus aureus bloodstream infection (SAB) progresses rapidly and is associated with high mortality and severe complications, including acute respiratory distress syndrome (ARDS), acute kidney injury (AKI), disseminated intravascular coagulation (DIC), and shock. Traditional critical illness scoring systems have limited capacity to capture pathogenspecific risk profiles and identify early clinical deterioration. Leveraging large-scale intensive care databases and machine learning algorithms to develop individualized prediction models may facilitate the early identification of high-risk patients and aid in clinical risk stratification and decision-making. Objective　To develop and validate machine learning-based prediction models for 30-day mortality and major complications, including AKI, ARDS, and DIC/shock-related outcomes, in patients with Staphylococcus aureus bloodstream infection using the MIMIC- Ⅳ database. Methods　A total of 2 536 adult patients with blood culture-confirmed Staphylococcus aureus bloodstream infection were retrospectively identified from the MIMIC- Ⅳ database. The cohort was randomly divided into training set and validation set at a ratio of 7:3. Multistage feature selection was performed using the Boruta algorithm combined with least absolute shrinkage and selection operator (LASSO) regression. Three nomogram models with different levels of variable complexity, namely large, medium, and small models, were constructed. Model performance was evaluated using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA). Results　 The 30-day mortality rate was 21.6%. Age, lactate level, white blood cell count, platelet count, and comorbidities were identified as important prognostic factors. In the comparison of algorithms using the same 32-variable feature set, the generalized linear model (GLM) showed the best performance, with an area under the ROC curve (AUC) of 0.816 (95% CI: 0.774 - 0.857) in the validation set. In the three-tier nomogram system, the large and small models yielded AUCs of 0.776 (95% CI: 0.737 - 0.814) and 0.719 (95% CI: 0.677 - 0.762), respectively. For ARDS, 7-day AKI, and DIC/shock-related outcomes, the large model achieved AUCs of 0.789 (95% CI: 0.755 - 0.823), 0.690 (95% CI: 0.649 - 0.731), and 0.792 (95% CI: 0.700 - 0.884), respectively. DCA demonstrated positive clinical net benefits across a broad range of threshold probabilities. Conclusion　This study demonstrates that the constructed hierarchical model effectively predicts mortality and complication risks in patients with Staphylococcus aureus bloodstream infection. It exhibits favorable interpretability and provides a scientific basis for individualized diagnosis and treatment.