Machine Learning Models Predict Breast Cancer Survival in Brazil

Researchers in Brazil have developed and internally validated machine learning models and prognostic nomograms to predict overall survival and breast cancer-specific survival using data from a population-based cancer registry. The study, published in BMC Medical Informatics and Decision Making, utilized a retrospective cohort from Mato Grosso, where structural disparities and limited access to specialized care pose additional challenges for accurate prognostic assessment. By leveraging data linked to mortality records, the team sought to create scalable tools for risk stratification in real-world public health settings.

The study design involved a retrospective population-based cohort study using data from the Population-Based Cancer Registry (RCBP) for the period between 2001 and 2018. This registry data was deterministically linked to the Mortality Information System (SIM) records from 2001 to 2022 to establish survival outcomes. The researchers developed five distinct survival prediction models: four machine learning algorithms and one conventional statistical model. The machine learning models included Random Survival Forest (RSF), Gradient Boosting Survival Analysis (GBS), Survival Support Vector Machine (SSVM), and ComponentWise Gradient Boosting (CWGB). The conventional model used for comparison was the Cox proportional hazards (CPH) model.

To ensure robust evaluation, the dataset was split into training and test sets using an 80:20 stratified ratio. Model performance was rigorously assessed using several metrics, including the concordance index (C-index), time-dependent area under the curve (AUCt), and the integrated Brier score (IBS). Bootstrap confidence intervals were calculated to provide uncertainty estimates for these performance metrics. The highest-performing model was further interpreted using SHAP values and permutation-based feature importance to understand which variables contributed most significantly to the predictions. Additionally, Cox-based nomograms were constructed and internally validated using 1,000 bootstrap resamples to evaluate their clinical utility through risk stratification and decision curve analysis.

The results indicated that the Gradient Boosting Survival Analysis (GBS) model demonstrated the highest overall performance for both overall survival and breast cancer-specific survival. For overall survival, the GBS model achieved a C-index of 0.781, with a 95% confidence interval ranging from 0.776 to 0.786. The time-dependent AUC was 0.843, and the integrated Brier score was 0.129. For breast cancer-specific survival, the GBS model showed a C-index of 0.790, with a 95% confidence interval of 0.784 to 0.796. The AUCt for this outcome was 0.852, and the IBS was 0.120. These metrics suggest strong discrimination and good calibration for the GBS model compared to the other algorithms tested.

In comparison, the conventional Cox proportional hazards model showed comparable discrimination but weaker calibration overall. The Cox-based nomograms maintained stable discrimination at 1-, 3-, 5-, and 10-year time points. There was good agreement between the predicted and observed outcomes for these nomograms. Both the machine learning models and the Cox-derived nomograms achieved significant risk stratification, with log-rank P-values less than 0.001.

The diagnostic method, tumor morphology, marital status, education level, and age were highlighted as the most important variables influencing the models’ outputs.

The study concludes that machine learning models, particularly Gradient Boosting Survival Analysis, and Cox-based nomograms demonstrate individualized prognostic capability for overall and breast cancer-specific survival. These approaches may support effective risk stratification, facilitating the identification of patients at higher mortality risk. They also provide a scalable framework for risk-adapted management in real-world public health settings.

Subject of Research: Machine learning models and nomograms for predicting breast cancer survival in a Brazilian population-based cohort

Article Title: Predicting overall and breast cancer-specific survival: machine learning models and prognostic nomograms from a population-based registry

Article References: Xavier, S. P., Victor, A., Gotine, A. R. E. M., das Neves, M. A. B., Cardoso, L. B., de Albuquerque Maia, F. H., Galvão, N. D., & da Silva, A. M. C. (2026). Predicting overall and breast cancer-specific survival: machine learning models and prognostic nomograms from a population-based registry. BMC Medical Informatics and Decision Making. https://doi.org/10.1186/s12911-026-03931-1

Image Credits: AI Generated

DOI: 10.1186/s12911-026-03931-1

Keywords: breast cancer, machine learning, survival prediction, nomograms, population-based registry, Brazil, prognosis, Gradient Boosting, Cox model, risk stratification, Predicting, overall