Application of CHAID analysis in the assessment of breast tumor severity: diagnostic utility by age

Introduction: Breast cancer is a frequent cause of mortality. Mammography and models such as CHAID may improve early diagnosis by structuring interactive clinical and radiological variables. Objective: To identify and rank mammographic features associated with tumor severity according to age group using CHAID analysis. Methods: A cross-sectional observational study was conducted using a secondary clinic-radiological database comprising 961 records, of which 956 women were included in the final analysis. The variables analyzed included tumor severity, age, nodule shape, margins, and density. A decision tree model based on chi-squared automatic interaction detection (CHAID) was applied, incorporating cross-validation and estimating sensitivity, specificity, and accuracy. Results: The CHAID model showed higher specificity in women aged <50 years (89.5%) and higher sensitivity in those aged ≥50 years (77.8%). Regarding structural patterns, in the <50-year group, the most discriminative variable was nodule shape, with irregular shape strongly associated with malignancy (76.3%), followed by non-circumscribed margins and iso-density. In women aged ≥50 years, the tree initially split according to margin morphology, particularly spiculated margins, which were associated with a high probability of malignancy (83.3%), followed by irregular shape and low density. Conclusions: The CHAID model demonstrated age-specific diagnostic utility, with greater accuracy in identifying benign lesions in women younger than 50 years and improved ability to detect malignancy in women aged 50 years or older.