Abstract:
Background: Locoregional recurrence (LRR) in breast cancer remains a significant concern in breast cancer survivors, impacting their prognosis and guiding subsequent treatment. Accurate risk stratification and prediction tools and models are substantial for personalised care and surveillance planning. This study aims to systematically review and evaluate the currently available evidence on risk assessment tools and prediction models for assessing LRR in breast cancer patients.
Methods: A comprehensive literature search was conducted across four electronic databases: PubMed, Medline (Ovis), Scopus, and Web of Science, identifying 2,702 records. Following the PRISMA 2020 guidelines and TRIPOD+AI checklist criteria for model evaluation, 12 studies were included to constitute the analysis in this study. Studies were screened by two independent reviewers for methodological rigor, model development strategies, validation approaches, and clinical applicability. A third reviewer opinion was sought to reach a consensus.
Results: The included models varied in design, population characteristics, predictors used, and outcome definitions. Most incorporated clinicopathological features such as patient variables, such as tumour size, nodal status and hormone receptor status, and treatment variables. While several models demonstrated moderate-to-good predictive performance (AUCs ranging from 0.70 to 0.85), external validation was limited, and calibration was infrequently reported. Few models were deployed to clinical practice or assessed for impact on decision-making.
Conclusion: Despite numerous attempts at validating LRR risk prediction in breast cancer patients, the majority of models are not well externally validated and implemented in the clinical practice. Future studies are warranted to focus on enhancing predictive performance, transparency of reporting, and external validation of current models in representative cohorts to facilitate personalised risk-based follow-up strategies.
