Clinical Scorecard: Predictive Model Utilizing Machine Learning for Complications Following Breast Reconstruction After Mastectomy

At a Glance

Key Highlights

• PMBR improves mental, psychosocial, and sexual well-being but carries risk of postoperative complications important for surgical decision-making.
• Machine learning models integrating structured EHR data and unstructured clinical notes can generate individualized risk predictions for major complications.
• The study developed and validated a prognostic ML algorithm using data from two large academic centers, including over 4000 patients with detailed manual EHR review of a random sample.

Guideline-Based Recommendations

Diagnosis

• Use comprehensive clinical data including demographics, treatment history, and surgical characteristics to assess risk preoperatively.
• Incorporate both structured EHR data and unstructured clinical notes for accurate complication attribution.

Management

• Apply individualized risk estimates from ML predictive models to inform surgical planning and patient counseling.
• Consider patient-specific factors such as reconstruction type, timing, and comorbidities in decision-making.

Monitoring & Follow-up

• Ensure at least 12 months of postoperative follow-up to capture major complications including unplanned reoperations and rehospitalizations.
• Use standardized protocols for data abstraction and complication tracking to maintain data quality.

Risks

• Recognize that patients from minoritized racial and ethnic groups and lower income backgrounds may have less access to reliable risk information.
• Acknowledge the complexity of predicting complications and the need for multifactorial models to improve accuracy.

Patient & Prescribing Data

Female patients undergoing therapeutic mastectomy with subsequent breast reconstruction at academic centers

ML models provide personalized risk estimates for major complications, supporting informed surgical decisions and potentially improving patient outcomes.

Clinical Best Practices

• Integrate machine learning predictive tools into preoperative evaluation to enhance individualized risk assessment.
• Use a combination of structured and unstructured clinical data for comprehensive risk modeling.
• Maintain rigorous data collection and follow-up protocols to ensure accurate complication identification and model validation.

References

• TRIPOD+AI guidelines
• Prior ML studies on PMBR complications
• Institutional Review Boards of University of Michigan and Stanford University