What can ChatGPT already do for doctors and patients?

A survey published in the International Journal of Retina and Vitreous compiled 68 studies to map out how generative artificial intelligence, especially ChatGPT, is being used by professionals in healthcare, particularly in ophthalmic care. The review, which included the participation of the D’Or Institute for Research and Education (IDOR), shows how tools based on large language models (LLMs) are gaining ground in medical routines, especially for triage, preliminary diagnoses, and medical education.

In recent years, platforms like OpenAI’s ChatGPT have increasingly been part of discussions about the future of medicine. Powered by large databases, these AI systems can respond to commands and questions using natural language, simulating human conversations with impressive fluency. But what, exactly, is AI already capable of doing for doctors and patients?

In an effort to answer this question, researchers conducted a systematic literature review on the use of ChatGPT in ophthalmic care, with a special focus on retinal diseases. The study considered 68 articles published on major scientific platforms, including publications that addressed the application of ChatGPT by healthcare professionals, researchers, patients, and family members, as long as they involved the care of retinal diseases.

The articles were classified into eight main categories: diagnosis (44 studies), disease management (49), patient counseling (33), medical education (23), academic research (8), clinical documentation (5), literature review (2), and medical coding/bureaucracy (1). Many publications were included in more than one category, highlighting the cross-cutting nature of AI across different stages of healthcare.

AI in Triage and Diagnosis

Among the main highlights of the survey is the use of ChatGPT as a triage tool, for initial contact between a patient and the healthcare system. At this stage, the patient inputs their symptoms and receives initial guidance on whether or not to seek specialized care. Although challenges still exist, such as the correct interpretation of reported symptoms, the results are promising: the AI showed the ability to recognize language patterns associated with clinical conditions and indicate appropriate levels of care.

Regarding diagnostic support, ChatGPT showed inconsistent performance. In some cases, it was able to offer guidance aligned with clinical guidelines, even for complex diseases like age-related macular degeneration and diabetic retinopathy. However, the model still provided incorrect answers, especially when confronted with rare cases or imprecise descriptions. Furthermore, it was possible to confirm the risk of hallucination, a term used to describe convincing but false responses generated by the AI.

Medical Education with AI Support

Another field where ChatGPT has excelled is medical education. Several studies included in the review tested the AI's performance on theoretical ophthalmology exams and multiple-choice questions similar to those used in medical boards. In many cases, ChatGPT performed comparably to or even better than professionals in the field.

However, this does not mean that AI can replace medical training. There have been reports of conceptual inaccuracies, inappropriate use of references, and difficulty handling specific contexts. For this reason, experts argue that ChatGPT should be used as a complementary tool, capable of providing explanations, reviewing concepts, and assisting with clinical reasoning, but always under human supervision.

Other Uses Under Evaluation

In addition to the areas of triage, diagnosis, and medical education, the study also identified applications of ChatGPT in administrative tasks, such as filling out medical records and generating procedure codes; in reviewing scientific literature; and even in drafting academic texts. However, in these cases, the AI's performance is more unstable and still requires rigorous validation.

What to Expect in the Future

The study offers a valuable overview of the current and potential uses of ChatGPT in medicine, with a focus on retinal diseases. Although significant limitations still exist, especially regarding the accuracy and verification of information, generative AIs already represent a tangible tool for supporting medical practice.

With the advancement of technology and the development of more robust models, these tools are expected to become useful in processes such as automated triage, clinical decision support, and continuing medical education. For this to happen, it will be essential to invest in professional supervision, integration with reliable sources, and continuous improvement of the AI, remembering that generative intelligence is not a reliable source of information, and it is the user's responsibility to check the data it provides. When it comes to patient health, the need to turn to traditional scientific sources is even greater, which reinforces the irreplaceable role of healthcare professionals in the technological advancement of medicine.

Written by Maria Eduarda Ledo de Abreu.