Association between frailty and breast cancer incidence and construction of a breast cancer prediction model based on machine learning
Related Experiment Video
Updated: Jul 21, 2026
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
Nani Li1, Jian Liu1, Minjing You1
1Department of Breast Medical Oncology, Clinical Oncology School of Fujian Medical University, Fujian Cancer Hospital, Fuzhou, Fujian, China.
Objective:
The relationship between frailty and breast cancer (BC) has been scarcely investigated. This study aims to explore the association between frailty and BC and to develop a predictive model for BC risk based on the frailty index (FI).
Methods:
This study utilized data from subjects with recorded frailty assessments and BC diagnosis collected by the National Health and Nutrition Examination Survey between 2003 and 2020. FI was assessed by 49 frailty indicators. Frailty was defined as FI > 0.20. First, logistic regression was used to analyze the link between frailty and BC. Subsequently, the dataset was divided into training and validation sets in a 7:3 ratio. Based on the training set, LASSO regression and 10-fold cross-validation were used to screen for the variables with the greatest predictive value. Then, the study built and evaluated eight machine learning (ML) models. Finally, SHapley Additive explanations was applied to interpret the best-performing model.
Results:
After data were excluded, a total of 4,473 patients were ultimately included. Logistic analysis revealed a positive correlation between FI and BC prevalence (p < 0.0001). Stratified analysis demonstrated that the correlation between the two variables was not influenced by other covariates. Five key feature variables were identified via LASSO regression and 10-fold cross-validation. Eight ML methods were employed to construct BC prediction models. After all the models were evaluated, the neural network demonstrated robust and reliable predictive performance and emerged as the optimal prediction model. Further the neural network model was elucidated via Shap, and the results demonstrated that the predictive variables ranked by importance from highest to lowest were age, FI, BMI, race, and education level.
Conclusion:
FI is positively correlated with the prevalence of BC. BC prediction model based on the attenuation index that this study developed and validated has high sensitivity, aids in early BC identification in frail individuals, and suggests that future studies refine its applicability.
