Machine Learning Approaches for Early Prostate Cancer Prediction Based on Healthcare Utilization Patterns

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning

Background:

• Prostate cancer diagnosis often occurs at later stages, limiting treatment effectiveness.
• Healthcare utilization data offers insights into patient health trends.
• Early detection of prostate cancer is crucial for improving patient outcomes.

Purpose of the Study:

• To develop a machine learning model for early prostate cancer prediction.
• To identify patterns in healthcare utilization preceding prostate cancer diagnosis.
• To evaluate the efficacy of supervised machine learning techniques for this prediction task.

Main Methods:

• Retrospective analysis of healthcare utilization patterns for 2916 prostate cancer patients.
• Examination of medical activity frequency and changes in the three years prior to diagnosis.
• Application and comparison of various supervised machine learning algorithms, including XGBoost.

Main Results:

• The XGBoost model, analyzing data 1-2 years before diagnosis, achieved the highest prediction accuracy.
• Achieved a high F1 score of 0.9 and an Area Under the Curve (AUC) score of 0.73.
• Identified significant changes in healthcare utilization patterns preceding prostate cancer diagnosis.

Conclusions:

• Machine learning applied to healthcare utilization data shows promise for early prostate cancer detection.
• The developed model can aid in identifying at-risk individuals for timely intervention.
• Further research is warranted to validate and implement these findings in clinical practice.