Reducing Benign Breast Biopsies in Women Aged 35 Years or Younger Using a Clinician–AI–Age Triage Strategy

Background/Objectives: Symptomatic breast services face increasing diagnostic demand, while younger women with benign-appearing ultrasound lesions frequently undergo biopsy despite a low prevalence of malignancy. This study evaluated whether a rule-based clinical decision algorithm combining patient age, clinician ultrasound assessment and artificial intelligence (AI) decision support could identify a subgroup in whom biopsy might be avoided without excluding malignancy from the biopsy pathway. Methods: A retrospective analysis was performed using prospectively collected service-evaluation data from 240 patients (254 lesions) attending a large NHS breast unit between May and December 2025. Clinician ultrasound assessment, KOIOS DS Breast AI output and histopathology were recorded. Lesions in women aged ≤ 35 years were eligible for biopsy avoidance when clinician assessment was U2 or U3 and the KOIOS AI assessment was BI-RADS 2. Histopathology was the reference standard. Secondary analyses compared clinician and AI diagnostic performance. An illustrative five-year budget impact analysis was undertaken from the NHS provider perspective. Results: Among 151 lesions in women aged ≤ 35 years, 146 were benign and five were malignant. The algorithm would have classified 87 lesions for biopsy avoidance (57.6%; 95% CI, 49.3–65.6%). Of these, 67 (77.0%) had a clinician U3 classification. All 87 were benign on histopathology (100%; 95% CI, 95.8–100%), while all five malignant lesions remained in the biopsy pathway. Clinician assessment showed higher specificity than AI (99.1% vs. 75.7%), whereas AI showed higher sensitivity (96.9% vs. 81.3%). Avoidance of 87 biopsies was associated with approximately £85,000 in gross cost offsets during the evaluation period in the illustrative economic model. Conclusions: The AI-supported clinical decision algorithm could substantially reduce benign biopsy activity in younger women while maintaining cancer detection in this cohort. The small number of malignancies, single-centre design and biopsy population mean that prospective multicentre validation is required before routine clinical implementation.