Impact of surgical approach and survival prediction of malignant phyllode tumor by machine learning

Area of Science:

• Oncology
• Surgical Oncology
• Data Science in Medicine

Background:

• Malignant Phyllode Tumors of the Breast (MPTB) are rare and require careful treatment planning.
• The impact of different surgical approaches on MPTB patient outcomes is not well-established.
• Developing accurate prognostic models is crucial for personalized patient management.

Purpose of the Study:

• To evaluate the effect of surgical approach (mastectomy vs. breast-conserving surgery) on survival in MPTB patients.
• To develop and validate machine learning-based prognostic prediction models for MPTB.
• To compare the efficacy of different machine learning algorithms in predicting MPTB survival.

Main Methods:

• Retrospective analysis of MPTB patients (aged 18-80) from the SEER database (2000-2020).
• Propensity-score matching (PSM) to reduce covariate imbalance.
• Cox proportional hazard regression and Kaplan-Meier analysis for survival assessment.
• Ten machine learning models were trained to predict 5-year breast cancer-specific survival (BCSS) and overall survival (OS).

Main Results:

• Multivariate Cox analysis indicated no significant difference in long-term survival outcomes between mastectomy and breast-conserving surgery (BCS) for MPTB patients (p > 0.05).
• This finding held true even for patients with poor pathologic types.
• The CatBoost model demonstrated superior predictive performance, achieving an AUC of 0.8488 for 5-year BCSS and 0.8512 for 5-year OS.

Conclusions:

• Surgical approach (mastectomy vs. BCS) does not significantly affect long-term survival in MPTB patients.
• BCS is a feasible and preferred option when adequate surgical margins can be achieved.
• The CatBoost model shows promise as a reliable tool for guiding systemic treatment decisions in MPTB management.