Using Regression Models to Predict Surgery Duration for Robotic-Assisted Total Laparoscopic Hysterectomy in Endometrial Cancer

Abstract

Study Objective To develop regression models using preoperative variables that can accurately predict the operating room time of robotic surgery for endometrial cancer. Design A retrospective review using preoperative variables including age, BMI, ASA score, surgery start time, preoperative diagnosis, number of previous abdominal surgeries, parity, and uterine volume were used to develop regression models. These regression models were compared to institutional models that used preoperative estimations of operating room times based on historical averages. Setting An urban, academic hospital. Patients or Participants 348 patients undergoing robotic-assisted total laparoscopic hysterectomy with staging for endometrial cancer between 2016-2020 by six surgeons. Interventions N/A Measurements and Main Results The mean operating room time was 234 minutes (range: 130 – 484). The mean preoperative estimated operating room time was 223 minutes (range: 120 – 390). Linear, ridge, lasso, and elastic net models had relatively similar performance measures, and all models outperformed preoperative estimated operating room duration. Ultimately, the elastic net model performed the best. Elastic net had the smallest median root mean square error (RMSE) 57.97 (CI of 54.44-61.72) and greatest median R-squared 0.22 (CI 0.18-0.26). Surgery start time and preoperative diagnosis were found to be most influential variables in predicting operating room time followed by BMI, uterine volume, and number of prior surgeries. Age, ASA, parity and surgery type were not important in predicting operating room time. Conclusion Regression modeling more accurately predicted operating room time than the institutional standard that used historical averages for robotic-assisted hysterectomy with staging for endometrial cancers. This study is novel in its application to the field of robotic surgery in gynecologic oncology as it builds off of previous work from other disciplines demonstrating improved OR time predictions with regression models. Further studies are needed with expanded databases and more robust variable sets to refine the models ability to account for variance.