Comparison of Predictive Models for Keloid Recurrence Based on Machine Learning

Area of Science:

• Dermatology
• Medical Informatics
• Biostatistics

Background:

• Keloid recurrence after treatment poses a significant clinical challenge.
• Accurate prediction of keloid recurrence is crucial for optimizing patient management and improving outcomes.

Purpose of the Study:

• To develop and compare three machine learning models for predicting keloid recurrence.
• To identify key factors influencing keloid recurrence.
• To evaluate the predictive performance of logistic regression, decision tree, and random forest models.

Main Methods:

• 301 keloid patients undergoing surgery and radiotherapy were included.
• Models were trained on 70% of data and validated on 30%.
• Performance was assessed using accuracy, sensitivity, specificity, precision, recall, kappa coefficient, and AUC.

Main Results:

• Machine learning models identified KAAS, mean arterial pressure, postoperative complications, and inflammatory cell proportion as key predictors.
• The decision tree model achieved the highest accuracy and precision.
• The logistic regression model demonstrated the best performance in terms of AUC.

Conclusions:

• Three machine learning models for keloid recurrence prediction were successfully established.
• KAAS, blood pressure, postoperative complications, and inflammatory cell proportion are significant factors.
• Logistic regression model offers the most favorable prognostic performance based on AUC.