Machine Learning Approach for the Determination of the Best Cut-Off Points for Ki67 Proliferation Index in Adjuvant and Neo-Adjuvant Therapy Breast Cancer Patients

Abstract

BackgroundThis study aims to evaluate Ki67 cut-off points for differentiating low and high-risk patients based on survival and recurrence and find the best Ki67 cut-off points in breast cancer patients undergoing adjuvant and neoadjuvant therapy using machine learning methods. Patients and MethodsPatients with breast cancer treated at 2 referral hospitals between December 2000 and March 2021 who had invasive breast cancer entered this study. There were 257 patients in the neoadjuvant group and 2139 in the adjuvant group. A decision tree method was used to predict the likelihood of survival and recurrence. The 2-ensemble technique of RUSboost and bagged tree were imposed on the decision tree method to increase the accuracy of the determination. 80 percent of the data was used to train and validate the model, and 20% was used as a test. ResultsIn adjuvant therapy breast cancer patients with Invasive ductal carcinoma (IDC) and Invasive lobular carcinoma (ILC) the cutoff points for survival were 20 and 10, respectively. For luminal A, luminal B, Her2 neu, and triple-negative adjuvant therapy patients’ the cutoff points for survival were 25, 15, 20, and 20, respectively. For neoadjuvant therapy luminal A and luminal B group, survival cutoff points were 25 and 20, respectively. ConclusionDespite variability in measurement and cut-off points, the Ki-67 proliferation index is still helpful in the clinic. Further investigation is needed to determine the best cut-off points for different patients. The sensitivity and specificity of Ki-67 cutoff point prediction models in this study could further prove its significance as a prognostic factor.