A novel machine learning-derived decision tree including uPA/PAI-1 for breast cancer care

Area of Science:

• Oncology
• Biomarkers
• Clinical Decision-Making

Background:

• uPA and PAI-1 are established biomarkers for chemotherapy benefit in HER2-negative, ER-positive breast cancer.
• Current French clinical practices utilize a standard decision tree for chemotherapy indication.
• The integration of uPA/PAI-1 into treatment algorithms requires further investigation for routine clinical use.

Purpose of the Study:

• To evaluate the real-world application of uPA/PAI-1 in clinical settings.
• To develop a novel therapeutic decision tree incorporating uPA/PAI-1 for chemotherapy decisions.
• To assess the concordance between existing and uPA/PAI-1-integrated treatment recommendations.

Main Methods:

• Comparative analysis of chemotherapy indications from a standard decision tree versus actual prescriptions.
• Machine learning techniques applied to analyze concordant and non-concordant prescription data.
• Retrospective observation of clinical routine use of uPA/PAI-1 in medical board decisions.

Main Results:

• A 71% concordance rate was observed between the standard decision tree and actual chemotherapy prescriptions.
• Discrepancies were attributed to contraindications, high tumor grade, and uPA/PAI-1 levels.
• uPA/PAI-1 influenced 17% of final decisions, often avoiding chemotherapy (66% of influenced cases).
• Elevated uPA/PAI-1 alone was insufficient for chemotherapy indication; it was considered with additional markers like Ki 67, vascular invasion, or ER-H score.

Conclusions:

• In routine practice, uPA/PAI-1 are not used as sole indicators for chemotherapy.
• uPA/PAI-1 demonstrate added value when combined with other routinely used biomarkers for therapeutic guidance.
• The study supports the refinement of chemotherapy decision-making by integrating uPA/PAI-1 with existing clinical and pathological data.