Clinical Scorecard: Utilizing SHAP Analysis to Enhance Machine Learning Models for Predicting Negative Outcomes in Breast Cancer Surgical Procedures
At a Glance
Category
Detail
Condition
Breast cancer postoperative adverse outcomes including recurrence and metastasis
Key Mechanisms
Machine learning prediction models using perioperative data; SHAP interpretability to identify key predictive factors
Target Population
Treatment-naïve female patients undergoing initial breast cancer surgery
Care Setting
Perioperative surgical care in tertiary medical centers
Key Highlights
XGBoost algorithm outperformed other machine learning models in predicting adverse postoperative outcomes with AUCs of 0.840 (internal) and 0.780 (external).
SHAP analysis identified systemic immune-inflammation index (SII), prognostic nutritional index (PNI), and age as the top three predictive factors.
The model demonstrated good calibration and clinical net benefit, supporting its utility for early identification of high-risk patients.
Guideline-Based Recommendations
Diagnosis
Utilize perioperative core clinical indicators to assess risk of adverse postoperative outcomes in breast cancer surgery.
Incorporate machine learning models, particularly XGBoost, for risk stratification.
Management
Apply predictive model outputs to guide personalized postoperative surveillance and intervention strategies.
Focus on modifiable factors such as nutritional and inflammatory status preoperatively.
Monitoring & Follow-up
Monitor systemic immune-inflammation index (SII) and prognostic nutritional index (PNI) perioperatively as key indicators.
Use model predictions to tailor follow-up intensity and timing.
Risks
Recognize that patients with elevated SII, low PNI, and advanced age have higher risk of adverse outcomes.
Consider these factors when planning perioperative care to mitigate recurrence and metastasis.
Patient & Prescribing Data
Treatment-naïve female breast cancer patients undergoing initial surgery
Perioperative data-driven machine learning models can identify patients at higher risk for adverse outcomes, enabling targeted interventions.
Clinical Best Practices
Collect comprehensive perioperative clinical data including inflammatory and nutritional indices for model input.
Employ interpretable machine learning models such as XGBoost combined with SHAP analysis to enhance clinical decision-making.
Integrate model predictions into multidisciplinary care planning to improve personalized patient outcomes.
