Incremental Diagnostic Value of Serum Tumor Markers Beyond Multimodal Ultrasound for BI‐RADS Category 4 Breast Lesions
Abstract
Objectives To determine the incremental diagnostic value and potential clinical utility of serum tumor markers beyond multimodal ultrasound for differentiating benign from malignant Breast Imaging Reporting and Data System category 4 breast lesions. Methods This retrospective diagnostic modeling study included 307 patients with 1 pathologically verified BI‐RADS category 4 lesion each. Three nested logistic regression models were evaluated: BI‐RADS subcategory alone, multimodal ultrasound comprising clinical and ultrasound variables, and multimodal ultrasound plus normalized serum CA15‐3, CEA, CA125, and CYFRA21‐1 ratios. Discrimination, calibration, Brier score, integrated discrimination improvement, likelihood‐ratio testing, bootstrap internal validation, and decision‐curve analysis were assessed. Prespecified analyses evaluated lesions 20 mm or smaller and patients with all 4 markers within reference limits. Results Of 307 lesions, 184 (59.9%) were malignant. The area under the receiver operating characteristic curve increased from 0.875 (95% CI, 0.837–0.913) for BI‐RADS alone to 0.920 (95% CI, 0.888–0.952) for multimodal ultrasound and 0.953 (95% CI, 0.931–0.976) after addition of serum markers. Relative to multimodal ultrasound, the combined model improved the AUC by 0.033 (95% CI, 0.013–0.055; p = .002), reduced the Brier score by 0.025 (95% CI, 0.011–0.040; p < .001), and yielded an integrated discrimination improvement of 0.103 (95% CI, 0.069–0.138; p < .001). Bootstrap‐corrected AUCs were 0.908 and 0.940, respectively. In lesions 20 mm or smaller, AUC increased from 0.934 to 0.970 ( p = .003). In the all‐marker‐negative subgroup, the corresponding increase from 0.912 to 0.930 was not statistically significant ( p = .229), although cohort‐derived combined‐model tertiles showed malignancy rates of 4.0, 24.5, and 89.8% ( p for trend <.001). Conclusions Serum tumor markers provided statistically significant incremental information beyond multimodal ultrasound in this BI‐RADS category 4 cohort, with a particularly clear improvement among small lesions. The marker‐negative risk‐stratification findings are exploratory. External validation and prospective assessment of clinical pathways are required before implementation.