UTI Surgery Risk Predictions

Predicting the risk of complications after surgical interventions to treat stress urinary incontinence with machine learning models using secondary care admin data of English patients

This project was completed by Liliana Valles Carrera, Data Science Manager, in the Data Science Team, as part of the Data Science MRes at the University of Leeds

The aim was to explore the use of machine learning models to predict the occurrence of complications following a surgical intervention to treat female stress urinary incontinence. Classifier models were trained to predict reoperation, renewal or repair, removal, and other complications that occur one, three, six and nine years after the surgical intervention. Secondary care data from English female patients diagnosed with stress urinary incontinence, who had a first-time insertion procedure after 1 January 2007 and were at least 17 years of age at the time of the intervention was used.

Machine learning models used in this study exhibited similar, yet suboptimal, performance in predicting the outcomes following surgical interventions for the treatment of stress urinary incontinence. AdaBoost produced the best results when predicting any outcome within 9 years with an F1 score of 0.6.

While further improvements are required, this study indicates that utilizing secondary care data contained in Hospital Episode Statistics alone is insufficient for achieving a satisfactory prediction performance. Future research needs to explore additional data sources such as SDIS - a new registry that collects and curates surgical devices data used on interventions to treat stress urinary incontinence and implantable medical devices data- to improve the performance of these models. Understanding the risk of complications from SUI interventions can help manage patients’ expectations for the short, medium and long term.