Survival prediction in metastatic breast cancer using artificial intelligence: a scoping review

Accurately predicting survival in metastatic breast cancer (MBC) is essential to support personalized treatment decisions. This scoping review examines the current applications of artificial intelligence (AI) models for survival prediction in MBC and highlights their relevance in improving clinical outcomes.

Of 1787 records screened, 15 studies met inclusion criteria. These studies used supervised learning approaches, including random survival forests (13.3%), Naïve Bayes classifiers (13.3%), and logistic regression models (20.0%), to predict overall survival, progression-free survival, and treatment response. Input data varied widely, incorporating electronic health records, clinical data, imaging, and genomic profiles. Among included studies, 66.7% addressed all three major breast cancer subtypes, 20.0% focused on ER-positive HER2-negative cases, and 13.3% did not specify subtype. Model performance varied, with sensitivities ranging from 42% to 90%, specificities from 53% to 90%, and area under the curve values between 0.70 and 0.85.

AI models show promising potential for improving survival prediction in MBC, offering tools to support more individualized care. However, limitations remain, including inconsistent data quality, suboptimal model performance, and a lack of external validation. Future work should focus on refining models and ensuring clinical applicability through robust validation.