Identifying Data-Driven Clinical Subgroups for Cervical Cancer Prevention With Machine Learning: Population-Based,...

Area of Science:

• Oncology
• Genomics
• Public Health

Background:

• Cervical cancer prevention (CCP) remains a significant global health challenge.
• Personalized, data-driven CCP strategies are needed to improve outcomes.
• Tailoring prevention to phenotypic profiles can reduce disease burden.

Purpose of the Study:

• Identify distinct cervical precancer and cancer risk subgroups using machine learning.
• Validate subgroup predictions across independent datasets.
• Propose a computational phenomapping strategy to enhance global CCP efforts.

Main Methods:

• Applied unsupervised machine learning to a deeply phenotyped cohort to identify CCP subgroups.
• Used weighted logistic regression to determine risks of cervical intraepithelial neoplasia (CIN2+ and CIN3+).
• Trained a supervised model for individual classification and validated it on an external cohort.

Main Results:

• Identified 5 distinct CCP subgroups from over 550,000 women.
• Subgroups CCP2-4 exhibited significantly higher risks for CIN2+ and CIN3+ compared to CCP1.
• Validated a triple strategy prioritizing high-risk subgroups (CCP3-4) for colposcopy and scaling HPV screening for CCP1-2.

Conclusions:

• Machine learning and electronic health records can enhance CCP strategies.
• Identifying key determinants of CIN2+/CIN3+ risk and classifying subgroups provides a data-driven foundation for tailored prevention.
• The proposed triple strategy offers a scalable tool to complement existing cervical cancer screening guidelines.