UPJ INSIGHT Predicting Surgical Experience After Robotic Radical Prostatectomy Simulation Using Machine Learning

By: Nathan Schuler, MS, Johns Hopkins University, Baltimore, Maryland; Lauren Shepard, MS, Johns Hopkins University, Baltimore, Maryland; Aaron Saxton, MD, University of Rochester Medical Center, New York; Jillian Russo, Case Western Reserve University, Cleveland, Ohio; Daniel Johnston, BS, University of Rochester School of Medicine and Dentistry, New York; Patrick Saba, MS, SUNY Upstate Medical University Norton College of Medicine, Syracuse, New York; Tyler Holler, MPH, University of Rochester Medical Center, New York; Andrea Smith, MS, Intuitive Surgical, Inc, Peachtree Corners, Georgia; Sue Kulason, PhD, Intuitive Surgical, Inc, Peachtree Corners, Georgia; Andrew Yee, PhD, Intuitive Surgical, Inc, Peachtree Corners, Georgia; Ahmed Ghazi, MD, FEBU, MHPE, Johns Hopkins University, Baltimore, Maryland | Posted on: 25 Oct 2023

Schuler N, Shepard L, Saxton A, et al. Predicting surgical experience after robotic nerve-sparing radical prostatectomy simulation using a machine learning–based multimodal analysis of objective performance metrics. Urol Pract. 2023;10(5):447-455.

Study Need and Importance

Nerve-sparing radical prostatectomy represents the current standard of care for localized prostate cancer, prioritizing oncologic outcomes while secondarily seeking to limit injury to the surrounding neurovascular bundle. Current video-based evaluation standards require expert review, are time-consuming to perform, and are subjective to reviewer bias. Encompassing 14.7% of all new cancer diagnoses in the United States in 2023, improving assessment and training of this procedure for prostate cancer management has potential for substantial benefit to patients. Machine learning has recently been employed to objectively assess surgical skills in several surgical tasks, offering promising alternatives to the current standard.

What We Found

We combined robotic kinematic data from the da Vinci console, surgical gesture (cut, dissect, clip, retract) data collected from video review, and model-integrated force sensor data from within our validated hydrogel nerve-sparing robot-assisted radical prostatectomy simulation platform. Using supervised classification algorithms, we were able to achieve receiver operating characteristic area under curve scores of 0.96 and maximum accuracy of 86% in predicting completion of a published learning curve of 250 cases for nerve sparing during the procedure.


This study featured a limited sample size (n=35) and did not include patient postoperative outcome data from participants.

Interpretation for Patient Care

We have identified a series of surgical dissection actions and explainable kinematic-based objective metrics that are common to high-volume surgeons (see Figure), and have demonstrated the ability of these metrics to predict procedure-specific experience with the highest accuracy of any similar published works. Surgeons looking to improve their own techniques can use these metrics as a guide for structured, objective feedback and self-evaluation. Further discussion and evaluation on factors contributing towards these metrics offer a clear pathway towards shortening the learning curve and optimizing patient outcomes sooner within the surgeon career timeline.

Figure. Variable permutation importance for best performing supervised classification algorithm. LR indicates logistic regression; OPI, objective performance indicators; RFE, recursive feature elimination.