152. Breast Cancer Related Lymphedema: Analysis Of Predictors Using Machine Learning

PURPOSE: Breast cancer related lymphedema is a devastating condition that negatively affects quality of life. We sought to identify risk factors related to lymphedema development and factors that portended early lymphedema development. METHODS: Patients with breast cancer that underwent sentinel lymph node biopsy (SLNB) or axillary lymph node dissection (ALND) at our institution between 2007-2021 were identified and sociodemographic and clinical information was extracted. Logistic regression identified risk factors related to lymphedema. Time to event analysis and cox-regression analysis identified factors related to earlier development of lymphedema. Additionally, machine learning prediction models and sensitivity analysis were employed. RESULTS: We identified 1,223 patients, of which 161 (13.2%) developed upper extremity lymphedema within 1.8 (mean, SD=2.5) years post-operatively. Patients with SLNB had significantly lower odds for lymphedema (vs. ALND, OR=0.29 [0.14-0.57]). Patients between 40-49 years of age, and 50-59 (vs. <40 years, OR=2.14 [1.00-4.60]; OR=2.42, [1.13-5.16] respectively), African American patients (vs. Caucasian, OR=1.86 [1.12-3.09]), patients with stage II, III, and IV disease (vs. stage 0, OR=3.75 [1.36-10.33]; OR=6.62 [2.14-20.51]; OR=9.36 [2.94-29.81]), and patients with Medicaid (vs. private insurance, OR=3.56 [1.73-7.28]) had higher rates of lymphedema. Cox-regression analysis showed that African American (HR=1.71 [1.08-2.70]), higher BMI (HR=1.03 [1.00-1.06]), higher stage (stage II, HR=2.22 [1.05-7.09]; stage III, HR=5.26 [1.86-14.88]; stage IV, HR=6.13 [2.12-17.75]), and Medicaid patients (HR=2.15 [1.12-3.80]) had higher hazards for lymphedema. Patients with SLNB had lower hazards for lymphedema (HR=0.43 [0.87-2.11]). Seven machine learning prediction models were created with an average accuracy range of 0.875-0.877. The sensitivity analysis identified age, stage of disease, and BMI as stronger predictors for development of lymphedema. CONCLUSION: Lymphedema development has identifiable risk factors. These factors can reliably be used to predict the risk of lymphedema development and enable clinicians to educate patients better and formulate treatment plans accordingly.