Risk prediction models for malignancy upgrade in high-risk breast lesions: a qualitative systematic review

Atypical breast lesions are high-risk findings found on some core needle biopsies that may represent concurrent malignancy. Clinical management remains uncertain due to wide variability in reported upgrade rates and an incomplete understanding of contributing risk factors. Risk prediction models have been developed to estimate likelihood of malignant upgrade (from atypia to malignancy), but these models are highly variable in performance and predictor selection. This systematic review evaluates existing models predicting upgrade to malignancy in high-risk breast lesions, focusing on clinical applicability.

A qualitative systematic review was conducted following PRISMA guidelines. Searches in MEDLINE, Embase, and Scopus identified studies that developed risk prediction models estimating breast malignancy upgrade after atypia diagnosis. Studies analyzing multiple risk factors and providing quantitative risk estimates were included. Extracted data included study characteristics, statistical methods, key predictors, and model performance. Prediction model Risk of Bias Assessment Tool (PROBAST) was used for quality assessment.

Of the 3202 studies screened, 17 met inclusion criteria. Sample sizes ranged from 20 to 525, with reported upgrade rates from 14.9 to 67.3%. Common predictors of upgrade included lesion size, histology, and radiologic-pathologic concordance. Discriminatory performance varied (AUROC 0.514-0.909), and calibration was rarely assessed, limiting reliability. Most studies lacked external validation and exhibited a high risk of bias.

Current risk prediction models for malignant upgrade for high-risk lesions demonstrate significant variability and limitations in widespread use. While they may supplement clinical judgment, further external validation and improved calibration are needed before they can reliably guide management.