The intersection between medicine and artificial intelligence (AI) has arrived. While the exploration of AI applications in health care is not new, it has received mainstream attention with the recent emergence of generative AI (GAI) tools such as OpenAI’s ChatGPT. These chatbots are built on large language models (LLMs), which utilize natural language processing to generate text-based responses to user queries. Although GAI will not replace physicians, physicians adopting GAI have the potential to provide higher-quality care. This article focuses on GAI tools that can potentially bridge the gap between patients and urologists. While mostly yet to be validated for safe and responsible use in medical settings, potential applications of GAI tools include assisting with diagnosis and treatment, simplification of medical information, patient scheduling and organization, and discharge/follow-up materials (Figure). In 2024 and beyond, AI will complement all aspects of urologic care.

Figure. Generative artificial intelligence (AI)–based framework for empowering patient experience.

Patients routinely use the internet to seek urologic health care information,¹ and they will likely adopt AI-powered chatbots for this purpose. In light of this, studies have begun assessing the quality of outputs generated by ChatGPT in response to patient inquiries related to urologic conditions. In a recent edition of The Journal of Urology^®, Davis et al assessed ChatGPT output quality in response to benign and oncologic urologic patient inquiries.² Generally, outputs were rated highly. However, shortcomings such as responses lacking key critical information and missing emergency diagnoses led to the conclusion that it is not yet ready for widespread public use for this purpose. On the other hand, AI chatbots have been shown useful in answering typical “curbside consult” questions with accurate and helpful medical information.³ Chatbots provide patients with 24/7 access to information, answer questions about urological conditions, provide resources, and may potentially help patients find the appropriate urologist for their unique condition (ie, andrologist vs oncologist), all of which can help patient understanding of their urologic condition and streamline care. Current shortcomings include the known bias (potential biases in the trained datasets will be propagated in chatbot responses) and hallucinations (inaccurate information put forward as fact). Users should be aware of these limitations and ultimately consult with their doctor.

Another potential application is the summarization and simplification of medical literature for the public. Upon receiving a medical diagnosis, patients are often overwhelmed with complex medical jargon and forced to choose a treatment plan without a complete understanding of the risks and benefits of each option.⁴ To this end, studies have begun assessing how well LLMs can summarize and simplify medical literature. For example, a recent study demonstrated ChatGPT’s ability to produce accurate layperson summaries of urologic study abstracts at a substantially more understandable reading level.⁵ Given the international push to improve health literacy and empower patient understanding of their medical conditions, this study proposed a fast, automatic, and innovative AI application to support these efforts.

Studies have also begun assessing GAI’s ability to assist in disease diagnosis and treatment decisions.⁶ If trained properly, there is potential to form complex and comprehensive differential diagnoses and recommend evidence-based treatment decisions, complementing well-known resources like UpToDate. Of course, a well-documented limitation of emerging LLMs is that they are only as sophisticated as the content trained on (ie, garbage in, garbage out). In response, companies like Google have developed specialized LLMs like MedPALM 2 that are trained specifically on credible and up-to-date medical literature—unlike ChatGPT, which collates information from the internet not accounting for credibility of sources. As these models become more accurate and less prone to artificial hallucination, their usefulness in clinical settings will continue to increase.

Until recently, the lack of AI integration with and access to electronic medical record information has been a limitation, but this is already changing. With integration into the electronic medical record,⁷ AI can be utilized to generate individualized follow-up reminders to patients, summarizing their clinical course and emphasizing the significance of adhering to their unique disease management plan. Furthermore, studies show that incorporating AI can bolster medication adherence by automatically generating a tailored medication schedule that minimizes drug interactions, aids with pill count, and provides clear explanations for the relevance of each medication to the patient.⁸ Moreover, AI has the potential to streamline appointment scheduling, reducing conflicts and ensuring adherence to long-term surveillance protocols, ultimately improving the continuity of care.⁹ As AI becomes increasingly integrated into health care, it is critical that patient information and HIPAA compliance remain at the forefront with strict data encryption, storage, and transmission policies. Patients interacting with AI should be informed on how their data will be collected and used, and explicit consent should be obtained.

AI will not only be used to help patients but will also help urologists in both the clinic and hospital settings. Writing outpatient notes can be time-consuming for urologists, but as speech-to-text AI applications advance, there is promise that AI systems will be able to transcribe patient-urologist conversations accurately and succinctly. Less time documenting means more time allotted to providing direct patient care. In the hospital setting, discharge summaries are often plagued by incompleteness and minor inaccuracies in the hospital course. However, an AI model with access to inpatient information could generate a comprehensive and precise discharge summary with details about the procedure indication, perioperative course, and follow-up plan.¹⁰ Additionally, discharge summaries could be modified into personalized discharge letters for patients, composed in easily understandable, nonmedical language, effectively narrating their hospital stay and providing clear postoperative instructions pertaining to medications, physical/occupational therapy, and follow-up appointments. Furthermore, as AI-driven language translation improves, these bespoke discharge letters would be immediately available to patients in their preferred language, improving health equity in regions with diverse ethnic backgrounds.

Certainly, AI will never be able to understand all the nuances of the human-to-human urologist-patient relationship, but perhaps AI can be used to strengthen this relationship. By addressing patient comprehension and understanding of urologic disease, increasing the patient’s accessibility to and satisfaction with high-quality urologic care, and enhancing patient-urologist communication and face-to-face time, AI can and will be used to bridge the gap between patients and urologists.