Development of a Cancer Prevention Vulnerability Score to Support Physician Assistant–Led Risk Stratification in Primary Care

Background: Physician Assistants (PAs) are central to cancer prevention in primary care, yet scalable tools to identify patients with clustered, modifiable prevention vulnerabilities remain limited. Objective: To develop and temporally evaluate a nationally representative, survey-weighted Cancer Prevention Vulnerability Score (CPVS) and examine its discrimination, calibration, and transport performance across survey cycles. Methods: Using HINTS 6 data (n=6,252), we defined high prevention vulnerability as the presence of at least two of three components: current smoking, insufficient physical activity, and lack of knowledge that alcohol increases cancer risk. CPVS was developed using survey-weighted gradient boosting and internally validated with cross-validation. Model performance was evaluated using area under the receiver operating characteristic curve (AUC), Brier score, and calibration. Temporal validation was conducted in HINTS 7 (n=7,278) using frozen transport and structured updating strategies. Results: Internal validation demonstrated moderate discrimination (surveyweighted AUC = 0.678) with acceptable overall accuracy (weighted Brier 0.20) and appropriate calibration. Under temporal transport, performance attenuated with the frozen model but improved after updating, with reclassification analyses indicating greater upward risk movement among vulnerable individuals. Conclusions: The CPVS offers a representative, weighted framework for identifying cancer risk clusters in primary care. Its strong validity and performance support its integration into PA-led preventive workflows.