Comparison of conventional clinical models and a multigene assay to assess DCIS risk in a prospective single center study.

Abstract

e12578 Background: Ductal carcinoma in situ (DCIS) is a preinvasive breast cancer typically excised and treated with adjuvant therapy. While there is consensus that this results in overtreatment, there is little agreement on who may avoid radiation or endocrine therapy. Two freely available prediction models, Van Nuys Prognostic Index (VNPI) and Memorial Sloan Kettering Nomogram (MSK-N), are commonly used to identify low-risk DCIS based on standard clinicopathological features. Oncotype DCIS is a newer commercially available tissue-based multigene assay also used to assess risk, but its use is limited due to cost and unclear value over VNPI and MSK-N. We sought to compare these tests’ agreement in determining DCIS ipsilateral breast recurrence (IBR) risk and potential to de-escalate therapy. Methods: In this subanalysis from a prospective single center clinical trial, we analyzed 38 patients with newly diagnosed pure DCIS confirmed at excision. Each risk assessment tool presents risk in a different manner. Oncotype DCIS provides multiple assessments: a raw score (0-100), a DCIS Score Category (low, intermediate, high), and a Refined Score incorporating clinical features (10-year IBR risk). The VNPI calculates a raw score (4-12) and assigns a risk category (low, intermediate, high). MSK-N calculates a 10-year IBR risk. The tests were dichotomized as low vs. not-low risk categories using these thresholds: DCIS Score Category = low, DCIS Refined Score ≤ 10% IBR risk, VNPI risk category = low, MSK-N ≤ 10% IBR risk. Agreement of the 4 models (DCIS Score Category, DCIS Refined Score, VNPI, MSK-N) were compared using Spearman’s rank correlation for continuous data; percent agreement and Cohen’s kappa (k) were used to compare categorized data. Results: There was poor agreement of continuous risk assessments across the 4 models, with only VNPI and DCIS Refined Score showing significant but moderate correlation (r = 0.53, p = 0.001; others ranged r = 0-0.27, p > 0.1). The number of cases identified as low-risk for each assay was DCIS Score Category = 14 (37%), DCIS Refined Score = 3 (8%), VNPI = 1 (3%), and MSK-N = 0. Percent agreement of low vs. not-low risk categorizations between DCIS Refined Score, VNPI, and MSK-N was 92-97%, but each assay classified 3 or fewer cases as low-risk and kappa was not significant (k = 0-0.48, p > .1). DCIS Score Category demonstrated 63-71% agreement with the other 3 assays (k = 0-0.26). Only 1 case was classified as low-risk DCIS on multiple tests. Conclusions: DCIS risk models have poor agreement for determining IBR risk. DCIS Score Category initially identified many low-risk lesions; however, inclusion of clinical features to create a Refined Score decreased this substantially, providing very few additional low-risk cases for de-escalation over VNPI or MSK-N. Additional studies are needed to determine precise IBR rates when these models are used for adjuvant therapy decision-making. Clinical trial information: NCT03495011 .