Improving Multi-Tumor Biomarker Health Check-up Tests with Machine Learning Algorithms.

Area of Science:

• Oncology
• Biostatistics
• Medical Informatics

Background:

• Tumor markers are widely used in global health check-ups, particularly in East Asia, for cancer screening.
• Machine learning (ML) algorithms offer potential to improve the diagnostic accuracy and clinical utility of these screening tests.
• Early cancer diagnosis through enhanced screening can lead to substantial improvements in patient outcomes.

Purpose of the Study:

• To develop and validate machine learning (ML)-based algorithms for improving cancer screening using tumor markers.
• To assess the performance of ML algorithms in early cancer detection and determining the organ of origin.
• To evaluate the clinical utility of ML-derived algorithms in guiding patient follow-up.

Main Methods:

• Developed and validated gender-specific ML algorithms (logistic regression, random forest) using a large real-world dataset (RWD) from asymptomatic individuals.
• Utilized tumor marker values, age, and gender from 27,938 individuals, including 342 cancer cases, for internal and external validation.
• Developed a secondary algorithm to predict the organ of origin for at-risk individuals.

Main Results:

• Gender-specific ML algorithms demonstrated improved performance over single-marker tests, with high areas under the receiver operating characteristic curve (AUROC) in both internal and external validation.
• The algorithms effectively stratified individuals into risk groups, significantly reducing the time to cancer diagnosis (TTD) for higher-risk categories.
• The organ of origin algorithm showed promising sensitivity and specificity in predicting affected organ systems for both men and women.

Conclusions:

• ML-derived algorithms, trained and validated on RWD, significantly enhance tumor marker-based screening for multiple early-stage cancers.
• These algorithms can effectively suggest the tissue of origin, aiding in more targeted diagnostic efforts.
• The developed ML tools provide valuable guidance for patient follow-up, optimizing the management of screen-detected cancers.