Axillary staging in the setting of a preoperative diagnosis of ductal cancer in situ (DCIS): results of an international expert panel and a critical guideline performance using frequentist and Bayesian analysis

Abstract

Background/Objective: Sentinel lymph node biopsy (SLNB) is not routine in DCIS. Guidelines suggest SLNB when there is high risk for underlying invasion (large size, high grade, symptomatic lesion) or for detection failure (e.g., after mastectomy). However, guidelines and current practice patterns are inconsistent. Moreover, whilst SLNB is thought to be feasible and accurate after wide local excision (WLE), there is less consensus to support its use after oncoplastic breast-conserving surgery (OPBCS), which can reduce the need for mastectomy (Mx) and is gradually adopted as standard of care. The study aim was to assess if guidelines or individualized assessment result in optimal selection of patients for upfront SLNB. Methods: A panel of 28 international experts (20 surgeons, 8 oncologists, Europe 20, USA 5, Asia/Australia 3) was formed, all blind to the identity of the others. They reviewed anonymized patient cases from the SentiNot study (n=184, m. age 60 years, DCIS m. size 4 cm, Grade 2/3= 36%/64%, mass lesions 13,4%, underlying invasion 24.5%) and answer if they would consider upfront SLNB and why. Consensus and majority were set to >75 and >50%. At the same time, 6 independent raters (4 surgeons, 2 oncologists) reviewed guidelines and assessed the same patient cases per each guideline. Accuracy in relation to underlying invasion was assessed by Receiver Operating Characteristic (ROC) curves and Area Under the Curve (AUC) was reported. Agreement was investigated by kappa statistics and decision-making patterns by logistic multivariate regression and cluster analysis. To allow for flexibility and adaptation to current knowledge, both a frequentist and a Bayesian approach were undertaken. Priors were adjusted after a literature review regarding the factors that are commonly thought to be associated with higher risk for underlying invasion. Results: A total of 44,896 decisions were retrieved and analysed. The panel reached consensus/majority for upfront SLNB in 41.3/61.4%, whereas individual rates ranged from 11 to 100%. Agreement among panelists was low (kappa=0.37). In multivariate regression analysis for the entire panel, type of surgery was the most common determinant, (simple WLE=less, OPBCS=more and Mx=constant for SLNB), followed by symptomatic diagnosis and DCIS size. Most (26) members had a clear decision-making pattern regarding SLND, based mainly on DCIS size and type of surgery. Individual decision-making performed modestly in identifying patients with underlying invasion (AUC range 0,47-0,59), resulting mainly in overtreatment in 44-77% of patients. The panel performed similarly by majority (AUC 0,5) and by consensus (AUC 0,55) but “undertreated” 60-75% of patients with invasion, failing to identify them as 'high-risk.' After the recognition of different decision-making patterns, panelists were divided in subgroups with similar decision-making pattern. Analysis identified subgroups with difference in SLNB rate but not with better AUC. The disagreement among panelists in the same subgroups was significant, not only regarding which patients should undergo SLNB, but also on what factors that recommendation was based on. Eight guidelines with relevant recommendations were identified [USA (ASCO/NCCN), Europe (ESMO), Sweden, Denmark, UK, Netherlands and Italy, retrieval date May 2019]. Agreement among raters for each guideline separately varied (kappa: 0.23-0.9). Interpretation as to whether SLNB should be performed ranged widely (4090%) and with varying concordance (32-88%). No guideline demonstrated accuracy (AUC range 0.45-0.55). Overtreatment risk was high (50-90%), whereas 10-50% of patients with invasion were not identified as “high- risk.” Agreement across guidelines was low (kappa=0.24), meaning that different patients had similar risk to be treated inaccurately, regardless of which guideline was examined. Conclusions: Individualized decision-making and guideline interpretation may be highly subjective and with low accuracy in terms of prediction of invasive disease, resulting in almost random risk for over- or undertreatment of the axilla in patients with DCIS. This suggests that current views and guidelines should be challenged. More accurate preoperative workup and novel techniques to allow for delayed SLNB may be of value in this setting.