Risk Prediction Based on Clinicopathologic Features in Korean Melanoma Patients by Machine Learning.

Area of Science:

• Oncology
• Bioinformatics
• Medical Machine Learning

Background:

• Malignant melanoma prognosis varies by ethnicity, influenced by histologic subtypes.
• Accurate risk stratification for melanoma is vital but challenging in low-incidence populations like Korea.
• Clinicopathologic features at diagnosis are crucial for predicting patient outcomes.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting 2-year disease-free survival (DFS) in Korean melanoma patients.
• To identify key clinicopathologic predictors of melanoma survival using data available at diagnosis.
• To improve risk stratification for malignant melanoma in a low-incidence ethnic group.

Main Methods:

• Retrospective analysis of 1,657 Korean melanoma patients (2006-2023).
• Training and evaluation of various machine learning algorithms (Decision Tree, Random Forest, Boosting, XGBoost, DNN) using 12 clinicopathologic variables.
• Performance assessment via accuracy, precision, recall, F1 score, and Area Under the Curve (AUC).

Main Results:

• The XGBoost model achieved the highest F1 score of 0.761 in predicting 2-year DFS.
• Primary tumor site, mitotic rate, and body mass index (BMI) were identified as the most significant predictors of survival.
• Kaplan-Meier curves and Cox analysis validated the clinical relevance of these identified features.

Conclusions:

• The XGBoost machine learning model demonstrates superior performance in predicting 2-year DFS for Korean melanoma patients.
• Primary cancer site, mitotic rate, and BMI are critical factors influencing melanoma survival in this population.
• These findings can aid in more accurate risk stratification and personalized treatment strategies for melanoma.