Presentation Authors: Andrew Hung*, Jian Chen, Los Angeles, CA, Zequn Liu, Beijing, China, People's Republic of, Jessica Nguyen, Paul Oh, Devin Stewart, Daphne Remulla, Tiffany Chu, Ryan Lee, Kartik Aron, Saum Ghodoussipour, Los Angeles, CA, Sanjay Purushotham, Baltimore, MD, Inderbir Gill, Yan Liu, Los Angeles, CA
Introduction: Automated performance metrics (APMs), validated measures of surgical performance, have not yet been applied to deep learning survival analyses (time-to-event analyses). Our study utilizes APMs and clinicopathological data (CPD) to predict time to urinary continence recovery, an important outcome after robot-assisted radical prostatectomy (RARP). The most essential APMs/CPD for continence prediction were subsequently used to rank surgeons and compare historical patient outcomes.
Methods: APMs and CPD from 100 RARPs, performed on a da Vinci Surgical System, were collected and applied as training data for a deep learning model. APMs (instrument motion tracking and systems events data) were recorded with a systems data recorder (Intuitive Surgical). CP data was collected prospectively. Step 1. Using five-fold stratified cross validation, the data was applied to a deep learning model-based survival analysis (DeepSurv) to predict time to urinary continence (no pads or 1 safety pad) after RARP. Concordance Index (CI) and Mean Absolute Error (MAE) measured prediction performance. Data inputs (APMs and CPD) were ranked based on importance for continence prediction. Step 2. Performance in the top-five ranked features was used to categorize eight surgeons into two groups. The four surgeons with higher rated performances were classified as &[Prime]Group 1/APMs&[Prime], with the remainder in &[Prime]Group 2/APMs&[Prime]. CPD from 547 historical cases (January 2015-August 2016) were compared between the two groups.
Results: Urinary continence was attained in 79/100 patients after a median time of 126 days (16-553 days). Median follow up duration was 18 months (4-24 months). DeepSurv achieved CI of 0.6 and MAE of 85.9. The 547 historical cases showed significant distinctions between the surgeon groups. Group 1/APM had shorter surgical times (230 vs. 244 min., p < 0.001), less anastomotic leaks (1.8 vs. 8.4%, p=0.001), less pelvic drain duration (1 vs. 3 days, p < 0.001), and yielded more lymph nodes (18 vs. 14, p < 0.001). Additionally, Group 1/APM had superior rates of urinary continence recovery at 3 months and 6 months post-op (48.5 vs 37.8%, p=0.023 and 71.4 and 62%, p=0.043, respectively).
Conclusions: APMs and CPD can be applied to deep learning models to predict time to urinary continence. Top ranking features for continence prediction can be used to stratify surgeons into groups with significant difference in historical clinical outcomes, especially continence recovery rate at 3- and 6-months.
Source of Funding: This study was funded in part by an Intuitive Surgical Clinical Grant; Intuitive Surgical provided the systems data recorder.Research reported in this publication was also supported in part by the National Institute Of Biomedical Imaging And Bioengineer