Development of a clinical decision rule for triage of women with palpable breast masses

Abstract 

Background and Objective: Given the large numbers of open breast biopsies performed in women who have benign breast masses, we developed a clinical decision rule (CDR), called BREASTAID, to triage women into open biopsy or follow-up.

Methods: A prospective cohort design was used to obtain data on 452 palpable breast masses evaluated at a referral clinic. Breast cancer was defined as ductal carcinoma in situ or invasive cancer at open biopsy. Separate logistic regression models were developed at three logical stages of the clinical workup. Bayes' theorem was applied in a stepwise fashion to revise model probabilities to generate a final probability of cancer. Receiver operator characteristics curves were generated to determine the optimum cut-point. Results derived from the CDR were compared with actual clinical practice.

Results: A total of 452 masses in 380 women were included. Clinical practice resulted in 180 masses (39.8%) undergoing open biopsy, 41 (22.8%) of which were cancers. Age, history of breast cancer in the mother, mass size, mammography findings, and fine needle aspiration biopsy results were included in the final models. When applied to the derivation dataset, BREASTAID successfully identified 40 of 41 cancer masses (sensitivity 97.6%, 95% confidence interval [CI] 94.1–99.9), and 350 of 411 noncancer masses (specificity 85.2%, 95% CI 81.8–88.5). BREASTAID would have reduced the number of biopsies performed on the 411 benign masses from 139 to 61.

Conclusions: This study demonstrated that a CDR based on routinely collected clinical variables has the potential to accurately triage women with palpable breast masses. Further validation of the rule is required before its clinical use can be considered.

Keywords:  Breast cancer, Clinical decision rule, Predictive models, Logistic regression