REMOTE DIGITAL SURVEILLANCE AND MACHINE LEARNING MODELLING TO PREDICT SURVIVAL FOLLOWING RADICAL CYSTECTOMY FOR BLADDER CANCER—A SECONDARY OUTCOME ANALYSIS OF THE IROC TRIAL

INTRODUCTION AND OBJECTIVE: Wearable devices allow for measurement of physical activity, potentially providing supplementary prognostic insights for predicting survival following radical cystectomy (RC), alongside commonly used conventional clinicopathological factors. The primary objective of this analysis is to assess the additional value of wearable device monitoring to pathological data in predicting survival. METHODS: The iROC randomised trial (NCT03049410) compared peri-operative recovery after intracorporeal robot-assisted RC (iRARC) vs open RC (ORC) for bladder cancer. Step-count data was collected using wrist-worn wearable devices, and maximum steps and average steps per day was calculated at baseline and 12 weeks post-RC. Stamina was measured using the 30 second chair-to-stand test at similar timepoints. Clinicopathological data and survival data was collected, and cross-sectional imaging was used to determine cancer recurrence. RESULTS: Among 338 patients in the iROC trial, 319 patients received RC. Overall survival following RC was 87% (319-41/319) patients and RFS was 82% (319-57/319) over a median follow-up of 33 months. Wearable device data was available for 165 patients for analysis. There was no significant reduction in maximum step-counts at baseline (mean 9388 , SD 4552) when compared with 3-months post-operatively (mean 8792, SD 4055). Using clinicopathological features including age, gender, BMI, pathological T-stage and surgical margin, we demonstrated an AUC of 74% to predict RFS which improved slightly to 76% on adding activity and stamina data. For prediction of OS, using clinicopathological features demonstrated an AUC of 71%, which improved to 81% with the addition of activity and stamina data. Kaplan Meier analysis with patients divided into low and high risk groups by the final model showed an increased PFS (99% vs 59%) and OS (98% vs 53%) in 33 months after surgery. CONCLUSIONS: Mobility data may offer information which could add value to traditional prognostic models for survival. With new wearable devices, activity data can be remotely collected alongside other biometric data such as heart rate which may further improve models to predict RFS and OS. Source of Funding: The Urology Foundation and The Champniss Foundation