Dados do Trabalho

Título

THE USE OF AN AI MODEL TO PREDITC POTENCY AND CONTINENCE AFTER ROBOTIC RADICAL PROSTATECTOMY: A PROSTATE CANCER REFERRAL MODEL ON 8,524 PATIENTS.

Resumo

Introduction
AI is an emerging and novel technology rapidly being employed for predictive modelling in a range of different disciplines, including Healthcare. The primary objective of this study was to determine the feasibility of training and internally validating an AI based model to predict likelihood of potency and continence from a relatively high-quality, large data set of patients undergoing nerve sparing robotic radical prostatectomy (RARP) for prostate cancer.

Materials & methods:
All patients from 2008 to 2023 with at least 12 months follow-up undergoing nerve sparing RARP at our institution were included. The primary outcome was the likelihood of continence and potency at 12 months following surgery. Continence was defined as no use protective urine pads and potency was defined as the ability to penetrate and satisfactorily complete intercourse, with or without PDE5-i, in more than half of the attempts. The dataset was initially divided into testing and training sets with an 80:20 random split. Four distinct AI models were assessed, including Andor® Health's proprietary Artificial Neural Network (ANN) machine learning model, XGBoost (XBG), Support Vector Machine (SVM), and K-Nearest Neighbor (KNN). 8,524 patients were included with 6,819 patients allocated to the training set and 1,705 patients allocated to the testing set.

Results:
When utilizing the ANN model, the area under the receiver operating characteristic (ROC) curve for predicting potency at 12 months was 0.74 with a positive predictive values (PPV) of 70.6% and negative predictive value (NPV) of 83.0%. The ROC curve for predicting likelihood of continence at 12 months was 0.68 with a PPV of 89.8% and NPV of 99.8%.

On feature importance analysis the preoperative characteristics that were most significant for potency, listed in order of their importance, were as follows; diabetes, history of coronary artery disease, pre-operative SHIM score, hypertension, and patient age.

For recovery of urinary continence, the most significant characteristics in order of importance were found to be diabetes, Latino ethnicity, hypertension, patients age, and history of coronary artery disease.

Conclusion:
Using an AI model to predict urinary continence and sexual potency recovery is a feasible method however it will require external validation to test its reliability and generalizability.