BREAST CANCER RISK PREDICTION MODELS: A SYSTEMATIC REVIEW

In Vietnam, breast cancer ranks as the leading malignancy among women, underscoring the importance of developing accurate risk prediction tools to facilitate screening and early diagnosis. Objective: To review breast cancer risk prediction models developed worldwide and propose a suitable model for the Vietnamese context. Methods: Models were evaluated based on AUC, sensitivity, specificity, and risk of bias assessed by the PROBAST tool. Meta-analysis using a random-effects approach was conducted to pool AUC estimates across studies. Results: Eighteen models were identified, five of which had undergone external validation, using up to 19 predictors and sample sizes ranging from 222 to 1,455,493 participants. AUC values ranged from 0.46 to 0.80; traditional models (Gail, IBIS, BCSC, Rosner–Colditz) achieved AUCs of 0.55–0.70, while BRCAPRO and AI/deep learning models reached 0.70–0.80. The pooled AUC was 0.65 (95% CI: 0.63–0.68; I² = 97.5%). Traditional models showed lower risk of bias than AI-based models per PROBAST. Conclusion: The Gail model is a feasible initial option for Vietnam given its simplicity, but requires local adaptation and validation.