21. A Tale of Two Surgeries: Predicting the Need for Secondary Revision Procedures after Breast Reconstruction Using Machine Learning

Abstract

PURPOSE: Breast revision surgery after two-stage implant-based reconstruction is an effective way to correct existing defects and improve aesthetic outcomes. For new patients, providing a quantifiable “likelihood of revision surgery to achieve satisfaction” may enable them to make a more informed decision regarding the modality of reconstruction. METHODS: Retrospective chart review was done on two-stage breast reconstruction patients between 2012-2021. The outcome of revision/corrective surgery was determined based on physician note and operative record. Patient characteristics that emerged as significant on both univariate and multivariate analysis were used to construct the machine learning model. Supervised learning models were evaluated using k-fold cross validation (k=3). A neural network model was also evaluated with a 0.8/0.1/0.1 train/validate/test split on the dataset. RESULTS: 185/406 (45.57%) required post-operative revision surgery to achieve more optimal aesthetic outcomes. On multivariate analysis, never smoking (OR 0.61), hypertension (OR 0.40), and textured expander (OR 0.52) corresponded to lower odds of undergoing revision surgery (p<0.05). Higher initial tissue expander volume (OR 1.002), vertical radial incision (OR 4.79), and larger nipple-inframammary fold distance (OR 1.12, p-value 0.03) conferred higher odds of needing additional revisions. Support Vector Machine achieved the highest performing machine learning model, with an accuracy of 61.3, and ROC AUC 0.67+/-0.05. CONCLUSION: Machine learning can help predict the probability of requiring secondary surgeries after implant-based reconstruction. As an example, for patients at high risk of needing further revisions, surgeons can provide more robust counseling on the use of autologous abdominal tissue vs. tissue expander/implant.