Artificial intelligence (AI) is the theory and application of computer programs that learn rules and relationships from static or regularly updating data. In health care, applications of AI tools and techniques range from analyzing radiologic, pathologic, or endoscopic images to clustering patients based on clinical trajectory or identifying those at risk of poor outcomes.^1,2 Providers most frequently interact with AI-based tools and techniques through clinical decision support tools that are intended to augment clinicians.^2,3

In urology, the applications of AI-based decision support tools remain limited.⁴ Limited use is due in part to several implementation barriers: oversight and regulation, interpretability of results, clinical staff education, patient engagement, performance assessment, and accountability.³ Development of AI-based decision support tools that are reliable, able to function efficiently, remain up to date, and identify and correct potential failures is crucial to implementation.³ AI-based decision support tools have potential for both harm and benefit. Like new devices and medications, these tools should be evaluated prospectively, and when their use is intended to change clinical practices, should undergo clinical trials.³ Within urology, the majority of AI-based tools have not been tested for clinical use via prospective studies, though many tools have been tested in randomized controlled trials in other fields.⁵

The following are examples of urology-pertinent applications of AI tools that have been tested in clinical randomized controlled trials and compared to conventional methods:

• Machine learning–based treatment planning for prostate brachytherapy (evaluated by humans prior to implementation) was compared to human treatment planning alone.⁶ The authors found that both techniques were similar in terms of dosimetry, but the AI algorithm had significant reduction in staff and planning time.⁶
• Perioperative outcome risk assessment for perioperative death and acute kidney injury, randomizing clinicians to predict these outcomes either with or without access to machine learning predictions (ongoing study).⁷
• Accuracy of predicting surgical case duration comparing machine learning model to standard of care (electronic health record average, surgeon, or scheduler’s estimate). The study found that machine learning prediction models lead to reduced schedule duration errors.⁸
• Randomizing men with prostate cancer to an interactive, patient-facing, web-based decision aid platform with real-time decision analysis (often using conjoint analysis) vs standard brochure.⁹ The study found that men who undergo the interactive decision aid were more certain about their treatment decision and reported decreased levels of decisional conflict. This application has also been studied in prospective cohort of men with benign prostatic hyperplasia.¹⁰

The field of AI-based decision support is rapidly growing. However, in order to implement these tools into routine clinical practice, the algorithms and tools—combined—must be reliable. Prospective studies compared to standard of care are required. To date, there have been few, prospectively studied, AI decision support tools applicable to urologic care, and there are few planned trials on urological applications of AI-augmented decision support within clinicaltrials.gov.