Integrated Clinicogenomic Risk Modeling for Metachronous Second Primary Cancers.

Area of Science:

• Oncology
• Genetics
• Bioinformatics

Background:

• Cancer survival rates are improving, leading to an increased incidence of subsequent primary malignancies.
• Existing screening protocols may not fully address the risk of multiple primary cancers (MPC).

Purpose of the Study:

• To develop and validate a classifier for identifying multiple primary cancers (MPC) phenotypes at scale.
• To build predictive models integrating clinicogenomic factors for early detection of second primary cancers in survivors.

Main Methods:

• A programmatic classifier was developed and validated on 81,175 cancer patients to identify first-second cancer pairs.
• Machine-learning models were constructed using germline variants, polygenic risk scores, treatment data, and demographics.
• Model performance was evaluated using a 15-year time-dependent AUC.

Main Results:

• Identified 56 first-second cancer pairs, with 22 exceeding expected incidence rates.
• Found persistent elevated risk for MPC even after accounting for known factors.
• Machine-learning models accurately predicted second ovarian and pancreatic cancers in survivors (AUC 0.70).

Conclusions:

• This study presents the first pan-cancer integration of clinicogenomic factors for predicting second primary malignancies.
• The developed framework enables individualized risk estimation for enhanced surveillance and cancer prevention in survivors.
• Predictive models can facilitate cost-effective, selective surveillance strategies for the growing cancer survivor population.