Investigate the accuracy and economic value of the upstaging prediction model for ductal carcinoma in situ.

Abstract

e12577 Background: The current study aims to define the “high-risk upstaging model” in ductal carcinoma in situ (DCIS) and investigate its accuracy in order to suggest the most precise application of sentinel lymph node biopsy (SLNB). Methods: PubMed, EMBASE, and Cochrane library databases were searched. Prospective and retrospective cohort studies that evaluated the pathological outcomes of SLNB and the upstaging rate in women with DCIS were included. We analyzed factors that potentially indicate nodal involvement risk and combine the significant ones into “high-risk upstaging model”. Consecutively, we conducted a retrospective study by applying our model and NCCN guideline model into DCIS patients extracted from hospital database. Both the excess treatment rate and incomplete treatment rate of SLNB were analyzed. The diagnostic performance of the model in predicting upstaging risk was evaluated with receiver operating characteristic (ROC) curve. Results: We retrieved 43 prospective and 69 retrospective studies including 44,001 DCIS patients. The upstaging rate was significantly higher in patients with estrogen receptor-negative status, palpable mass, tumor size > 2 cm on imaging, and high-nuclear grade and those who received a preoperative diagnosis through core needle biopsy. We then collected 168 DCIS patients who entered in Shuang-Ho hospital from February 2017 to August 2021. Ninety-six patients who didn’t upstage in final pathology have received extra SLNB and two patients who upstaged in final pathology did not receive SLNB in primary surgery. The areas under the ROC curve of the model developed by multivariable logistic regression for predicting upstaging risk in the NCCN guideline group and “high-risk upstaging model” group were 0.642 and 0.706, the sensitivity at 66.0% and 83.0%, and the specificities at 62.5% and 49.2%, respectively. Conclusions: “High-risk upstaging model” shows higher areas under the ROC curve and sensitivity as well compared with current NCCN guideline. Our prediction model can provide surgeons with a more accurate and specific indications of SLNB in DCIS.