External validation of four breast cancer risk prediction models for personalized screening in a prospective Dutch screening cohort

Le résumé fourni par la source

Tailoring breast cancer screening to individual risk can improve its harm-benefit ratio compared to a ‘one-size-fits-all’ approach. We externally validated four risk prediction models—Gail, BCSC, BOADICEA, and IBIS—using data from the Personalized RIsk-based MAmmography screening (PRISMA) cohort, embedded in the Dutch biennial screening program (inclusion period 2014–2019). We predicted 5-year breast cancer risk for 38,767 participants using questionnaire data (personal, lifestyle, hormonal, family history) and mammogram-derived breast density, measured using Volpara v1.5.5.4. DNA information was not included. Breast cancer diagnoses until October 2023 were ascertained through linkage with the Netherlands Cancer Registry. Performance was evaluated with concordance index (C-index), observed-expected (O/E) ratio, and calibration slope. During a median follow-up of 4.3 years, 609 cancers were diagnosed. Gail performed poorly (C-index 0.56 [95% CI 0.53–0.59], O/E 0.79 [0.71–0.86], slope 0.59 [0.38–0.80]). BCSC, BOADICEA, and IBIS had modest discrimination (C-indices 0.60 [0.57–0.63], 0.60 [0.57–0.63] and 0.61 [0.58–0.63]), with good calibration for BCSC (O/E 0.93 [0.84–1.05], slope 0.99 [0.69–1.23]) and BOADICEA (O/E 0.96 [0.86–1.06], slope 0.78 [0.57–0.95]). IBIS overpredicted risk (O/E 0.71 [0.56–0.70], slope 0.66 [0.47–0.80]). These results suggest that traditional models have limited accuracy, and better-performing models are needed to realize the potential of personalized screening in nationwide programs.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.