Artificial intelligence (AI) itself is not a recent technology as the field has been intensively investigated since Alan Turing’s pivotal paper on “Imitation Game” in the 1950s. Since that time, AI has emerged from simple rule-based systems to advanced machine learning and deep learning models, significantly increasing its ability to process and interpret complex data. Key breakthroughs of this evolution include the development of neural networks leading to remarkable advancements in image recognition, task automation, and natural language processing.¹ While being mostly restricted to research and application by engineers, AI (especially the subcategory of generative artificial intelligence) has recently become a mainstream application though integration into user-friendly chatbot interfaces that can even be directly used by laypersons without prior education, and therefore is experiencing explosive adoption worldwide.²

In light of those developments, the expansion of AI into clinical settings is a very recent but logical development. AI is used in multiple ways to increase diagnostic accuracy, especially in radiological and pathological imaging, to improve patient workflows, innovate clinical trial design, and support decision making in therapeutic interventions.³ Because of the vast amount of data, medicine is an ideal place for implementation of such tools. However, requirements regarding safety and fault tolerance in medicine are higher than for other applications. Broad adoption and diffusion of those AI tools require an understanding of the applicability and limitations of AI-based frameworks. Therefore, transparent and standardized reporting of study results is paramount to ensure understanding, trust, and adoption by physicians for clinical use.

Need of Guidelines for Data Reporting in Urology

The main reason for new guidelines is that AI is an intervention fundamentally distinct from previous ones in that it incorporates neither clear human or not-human features. It is located somewhere in between this intersection. AI is an intervention that could potentially act as a human and applications often interfere with human responsibilities and autonomy. Those ethical considerations cumulate in a clear call to action for new guidelines or guideline extensions of existing guidelines.⁴ Data from AI-based diagnostic and therapeutic tools provide new challenges for researchers and clinicians alike as they have to be reliable and transparent, reported to ensure reproducibility of the results as well as enabling urologists to understand and critically evaluate the significance of findings as well as transferability of algorithms to single patients. Selection bias observed in conventional studies also applies for AI-based diagnostic tools and impacts results. For instance, AI-based assessment of prostate magnetic resonance imaging is limited by the used training dataset. Conclusions might only be true for patients similar to the validated cohort. Current studies are, however, highly limited in this reporting, and therefore in the comparability of their results.⁵ In addition to lack of transparencies and biases known from other research areas, AI applications face the challenge of relying partially on black box approaches as it can be difficult to understand why an algorithm comes to a certain conclusion. Therefore, precise description of key items and features to allow for better interpretation of the findings is key.⁶ Minimum reporting guidelines ensuring description of those key features are therefore paramount and have to be created.

Guidelines for Data Reporting

The EQUATOR (Enhancing the Quality and Transparency of Health Research) network, working closely with AI guideline developers, has released new updated guidelines to ensure that AI is reported properly (Figure). Several guidelines for data reporting have already been updated to meet the requirements of AI-based interventions. If no prior guideline was available, new checklists and guidelines were developed:

Figure. Guidelines for reporting of studies on artificial intelligence (AI).^7-11,15,16 GAI indicates generative artificial intelligence; RoB, risk of bias. ^ßOngoing work. ^§Claim 2.0 is an initiative ongoing and will be released in 2024.

CONSORT-AI: The CONSORT (Consolidated Standards of Reporting Trials) AI extension is released to provide minimum reporting criteria for randomized trials including an AI-based intervention. This extension adds 14 new items to the established CONSORT 2010 items that help to increase transparency of the technology.⁷

SPIRIT-AI: The SPIRIT (Standard Protocol Items: Recommendations for Interventional Trials) of 2013 has been extended to standardize reporting for clinical trial protocols for interventions with an AI component. Fifteen new items have been added to the SPIRIT 2013 items.⁸

DECIDE-AI: The DECIDE-AI (Developmental and Exploratory Clinical Investigations of Decision support systems driven by Artificial Intelligence) reporting guideline aims to address the need for early-stage clinical evaluation of AI-based decision support systems to ensure safety. The guideline provides 17 AI-specific and 10 generic reporting items to ensure replicability of the studies.⁹

CLAIM: The Checklist for Artificial Intelligence in Medical Imaging (CLAIM) has been released to improve reporting for medical imaging by providing 42 items.¹⁰ The CLAIM guideline is expected to be released as CLAIM 2.0 in early 2024.

Further guidelines or guideline extensions are currently under development and will soon be released to improve methodology and result reporting in health care:

TRIPOD-AI: The TRIPOD (Transparent Reporting of a multivariable prediction model of Individual Prognosis Or Diagnosis) statement has been further developed by an AI extension.¹¹

STARD-AI: The STARD (Standards for Reporting of Diagnostic Accuracy Study) AI extension is under development based on STARD 2015 to address challenges of AI.¹²

PROBAST-AI: The PROBAST (Prediction model Risk Of Bias Assessment Tool) is currently updated with 2 Delphi consensus rounds already held and reports expected soon.^11,13

PRISMA-AI: The PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines will soon be updated with AI extensions to cover systematic reviews and meta-analysis on the topic of AI.¹⁴

CANGARU: CANGARU (Chat-GPT, Generative Artificial Intelligence, and Natural Large Language Models for Accountable Reporting and Use) guidelines will standardize reporting of generative artificial intelligence applications.¹⁵

Adoption of Guidelines for Urologists

The described guidelines are developed in general for application of AI in medicine. However, their relevance extends specifically to the field of urology, where AI is currently explored in multiple areas and first applications are released into the clinical setting, necessitating their adoption in this specialty. The widespread dissemination of these guidelines is crucial, ensuring that professionals involved in developing, testing, or validating AI applications in urology are well informed and adhere to these standards in their reporting. Additionally, with the increasing volume of publications in this domain, urology journals should consider incorporating these guidelines into their standard submission requirements. Our team, together with AI specialists in urology and editors from leading urological journals, are initiating the development of recommendations. These guidelines will align with the established standards to ensure accurate reporting of methodology and data in AI-based urological studies. A comprehensive understanding of these guidelines might significantly enhance the clinical applicability of AI-based diagnostic and therapeutic interventions in urology and promote those solutions to become the perfect copilot for urologists.