AI-PEDURO - Artificial intelligence in pediatric urology: Protocol for a living scoping review and online repository
Adree Khondker,Jethro C C Kwong,Mandy Rickard,Lauren Erdman,Andrew T Gabrielson,David-Dan Nguyen,Jin Kyu Kim,Tariq Abbas,Nicolas Fernandez,Katherine Fischer,Lisette A 't Hoen,Daniel T Keefe,Caleb P Nelson,Bernarda Viteri,Hsin-Hsiao Scott Wang,John Weaver,Priyank Yadav,Armando J Lorenzo
DOI: https://doi.org/10.1016/j.jpurol.2024.10.003
2024-10-05
Abstract:Background: Artificial intelligence (AI) and machine learning (ML) methods are increasingly being applied in pediatric urology across a growing number of settings, with more extensive databases and wider interest for use in clinical practice. More than 30 ML models have been published in the pediatric urology literature, but many lack items required by contemporary reporting frameworks to be high quality. For example, most studies lack multi-institution validation, validation over time, and validation within the clinical environment, resulting in a large discrepancy between the number of models developed versus the number of models deployed in a clinical setting, a phenomenon known as the AI chasm. Furthermore, pediatric urology is a unique subspecialty of urology with low frequency conditions and complex phenotypes where clinical management can rely on a lower quality of evidence. Objective: To establish the AI in PEDiatric UROlogy (AI-PEDURO) collaborative, which will carry out a living scoping review and create an online repository (www.aipeduro.com) for models in the field and facilitate an evidence synthesis of AI models in pediatric urology. Methods and analysis: The scoping review will follow PRISMA-ScR guidelines. We will include ML models identified through standardized search methods of four databases, hand-search papers, and user-submitted models. Retrieved records will be included if they involve ML algorithms for prediction, classification, or risk stratification for pediatric urology conditions. The results will be tabulated and assessed for trends within the literature. Based on data availability, models will be divided into clinical disease sections (e.g. hydronephrosis, hypospadias, vesicoureteral reflux). A risk assessment will be performed using the APPRAISE-AI tool. The retrieved model cards (brief summary model characteristics in table form) will be uploaded to the online repository for open access to clinicians, patients, and data scientists, and will be linked to the Digital Object Identifier (DOI) for each article. Discussion and conclusion: We hope this living scoping review and online repository will offer a valuable reference for pediatric urologists to assess disease-specific ML models' scope, validity, and credibility to encourage opportunities for collaboration, external validation, clinical testing, and responsible deployment. In addition, the repository may aid in identifying areas in need of further research.