Repurposing cardiovascular disease prediction models for cancer.

Area of Science:

• Epidemiology
• Preventive Medicine
• Health Informatics

Background:

• Population cancer screening focuses on early detection, not future risk assessment.
• Cardiovascular disease (CVD) risk models are widely implemented but not typically used for cancer risk prediction.

Purpose of the Study:

• To evaluate the utility of established CVD risk prediction models in forecasting 10-year cancer risk.
• To compare the performance of CVD risk models against the QCancer risk model for cancer prediction.

Main Methods:

• Four CVD prediction models (QRISK3, PCE, SCORE2, SCORE2-OP) were assessed.
• Models were recalibrated using the UK Biobank cohort and validated in the Clinical Practice Research Datalink.
• Performance was measured using discrimination (c-statistics) and calibration (slope and intercept), with feature importance analysis.

Main Results:

• CVD models demonstrated moderate discrimination for any cancer (c-statistic 0.63) and comparable performance for specific cancers (e.g., lung, renal).
• Recalibrated CVD models showed excellent calibration in both validation cohorts.
• Age, smoking status, and systolic blood pressure emerged as key predictors of cancer risk.

Conclusions:

• Widely used CVD risk models show comparable performance to QCancer for incident cancer prediction.
• These models can be leveraged to inform cancer prevention strategies and guide risk-stratified surveillance.
• Recalibrated models are accessible via an open-source web application.