Photo Credit: Freepik

The following is a summary of “Evaluating key predictors of breast cancer through survival: a comparison of AFT frailty models with LASSO, ridge, and elastic net regularization,” published in the April 2025 issue of BMC Cancer by Bosson-Amedenu et al. 

Frailty models have become essential tools in survival analysis, offering a mechanism to account for unobserved heterogeneity among individuals. However, identifying the most robust model for accurate survival prediction, particularly in high-dimensional datasets, remains a persistent challenge. This study systematically evaluates the performance of multiple Accelerated Failure Time, frailty models and investigates the impact of regularization techniques, including LASSO, Ridge, and Elastic Net, on predictive accuracy and model selection.

The analysis involved both simulated datasets and real-world data from patients with breast cancer. Seven AFT frailty models were compared: Weibull, Log-logistic, Gamma, Gompertz, Log-normal, Generalized Gamma, and Extreme Value Frailty AFT. Model performance was assessed using Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), Mean Absolute Error (MAE), and Mean Squared Error (MSE) across three different sample sizes (25%, 50%, and 75%). Regularization approaches were incorporated to optimize model parsimony and mitigate overfitting in high-dimensional settings.

Among the models examined, the Extreme Value Frailty AFT model consistently demonstrated superior performance, yielding the lowest AIC, BIC, MAE, and MSE values across all sample sizes. For instance, AIC values for this model ranged from 100.41 at 25% sample size to 384.58 at 75%, outperforming the next best model—Log-logistic—by a substantial margin. LASSO regularization further enhanced the model’s interpretability by eliminating non-informative covariates such as Age, PR status, and Hospitalization, while preserving essential prognostic indicators, including Competing Risks, Metastasis, Disease Stage, and Lymph Node involvement.

Key clinical variables retained after LASSO regularization had strong prognostic implications. Patients without metastasis experienced a 2.63-fold increase in expected survival compared to those with metastasis. Additionally, early-stage diagnosis and fewer lymph node involvements were associated with 26% and 16% longer survival times, respectively. Other influential predictors included recurrence status (19% improvement), HER2 negativity (20% increase), absence of the Triple Negative subtype (15% gain), and lower tumor grade (11% increase). Kaplan–Meier survival curves revealed sharp declines in survival among patients with metastasis, higher tumor grades, HER2-positive tumors, and Triple Negative subtypes, emphasizing the need for early detection and subtype-specific treatments.

The study highlights the utility of the Extreme Value Frailty AFT model, especially when paired with LASSO regularization, for robust and interpretable survival prediction in breast cancer. These findings reinforce the value of advanced statistical modeling in refining risk stratification, guiding clinical decision-making, and supporting the development of targeted therapeutic strategies.

Source: bmccancer.biomedcentral.com/articles/10.1186/s12885-025-14040-z