To investigate the clinical value of a machine learning model constructed using perioperative data for predicting adverse postoperative outcomes in patients undergoing breast cancer surgery, emphasizing the identification of key decision factors through SHAP interpretability analysis.
Key Findings:
The XGBoost model showed optimal performance with an AUC of 0.840 in the internal validation set and 0.780 in the external validation set, with decision curve analysis indicating the highest clinical net benefit.
The model demonstrated high specificity (0.881) and an F1 score (0.514) compared to other models.
Calibration analysis indicated good agreement between predicted probabilities and actual incidence rates.
The top three predictive factors identified were systemic immune-inflammation index (SII), prognostic nutritional index (PNI), and age.
Interpretation:
The XGBoost model effectively predicts adverse postoperative outcomes in breast cancer surgery, outperforming traditional models and other machine learning approaches, with preoperative SII as the most critical factor, highlighting its potential for improving clinical decision-making.
Limitations:
Retrospective study design may introduce bias.
Findings may not be generalizable beyond the studied populations.
Potential impact of missing data or unmeasured confounders.
Conclusion:
The XGBoost model constructed using perioperative data can effectively predict adverse postoperative outcomes in breast cancer patients, providing a foundation for personalized treatment strategies.
