Deciphering Accelerated Biological Ageing in Breast Cancer Survivors and Cancer-Free Women Using Explainable Machine Learning

Abstract Purpose Breast cancer survivors are at increased risk of accelerated biological ageing, yet the interconnected variables driving this process remain poorly mapped. This study aimed to decipher the multidimensional factors associated with accelerated biological ageing using explainable Machine Learning (ML) architectures. Methods As part of the Cancer Survivors’ Trajectories of Ageing Research (C*STAR) study, a cross-sectional analysis was conducted among 262 breast cancer survivors and 262 cancer-free controls. Ageing phenotypes, behavioural lifestyle factors, sociodemographic and clinical characteristics were assessed, utilising absolute telomere length as the primary ageing biomarker. An XGBoost model integrated with the Synthetic Minority Over-sampling Technique for Regression with Gaussian Noise (SMOGN) was optimised using Particle Swarm Optimisation (PSO) to map non-linear associations, while SHapley Additive exPlanations (SHAP) evaluated feature importance. Results The XGBoost-SMOGN pipeline substantially reduced the mean absolute error for breast cancer survivors (0.0070 vs. 0.1785) and controls (0.0074 vs. 0.1592) compared to native models. As synthetic sampling was applied globally, these metrics reflect internal feature mapping rather than generalisable predictive accuracy. SHAP analysis deciphered that accelerated ageing (shortened telomeres) in survivors was primarily associated with impaired functional status, excessive sleep, high stress, radiotherapy, and high processed meat consumption. Paradoxically, optimal cognitive status and shorter sitting durations were also linked to accelerated ageing in survivors. Decelerated ageing in survivors correlated with low sleep disturbance and specific preserved cognitive domains. For controls, accelerated ageing was linked to poor sleep efficiency, a family history of diabetes, and high confectionery/condiment intake. Conversely, decelerated ageing in controls was strongly associated with high physical fitness, Malay ethnicity, increased height, and nutrient-dense diets. Conclusions Explainable ML provides a robust exploratory framework for deciphering complex, interrelated variables associated with telomere attrition. Deciphering the complex biological, clinical, and behavioural profiles associated with accelerated biological ageing via machine learning can help identify highly vulnerable survivors, paving the way for targeted, personalised survivorship care to preserve telomere health.