Predicción de la sobrevida global y libre de recaída en pacientes con cáncer de mama triple negativo a través de agrupaciones basadas en el aprendizaje automático sobre datos clínicos

Abstract

Introduction

Triple-negative breast cancer (TNBC) accounts for 15%–20% of all breast cancers and is characterised by the absence of hormone and HER2 receptors, limiting therapeutic options to cytotoxic chemotherapy. This subtype is highly aggressive, with frequent relapses and a poor prognosis. The integration of clinical and molecular data using machine learning algorithms offers an opportunity to improve the prediction of clinical outcomes and personalise the management of this disease.

Objective

To develop predictive models to estimate overall survival (OS) and relapse-free survival (RFS) in patients with TNBC.

Methods

A retrospective cohort study was conducted with 4,808 patients diagnosed between 2012 and 2024. Patients were grouped using the k-prototypes algorithm for mixed data. Demographic and clinical variables, biomarkers, and clinical outcomes were analysed. Multivariate analyses and Cox proportional hazards models were used to assess the association between clusters and OS and RFS outcomes.

Results

Four clusters of patients were identified. The highest risk group (cluster 3) had the highest mortality (42.3%; HR: 1.94; 95% CI: 1.63–2.30) and relapse (54.25%; HR: 1.68; 95% CI: 1.45–1.95), whereas cluster 0 had the best outcomes (mortality of 22.51%). Factors such as mutations in PIK3CA (HR: 1.535; p = 0.001) and TP53 (HR: 1.180; p = 0.023) were associated with adverse outcomes.

Discussion

The results support the utility of cluster-based stratification to predict outcomes and guide personalised interventions in TNBC.