Large Language Models in Hearing Care
This scoping review of 110 reports indicates that while large language models show promise for specific hearing care tasks, current evidence lacks proof of clinical benefit, prospective deployment, or robust privacy safeguards, highlighting an urgent need for future research focused on real-world impact, safety, and comparative effectiveness.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
More than one and a half billion people around the world live with some degree of hearing loss, yet there are not enough trained specialists to help them all. In clinics, telehealth services, and even at home, the gap between the need for care and the available professionals is widening. To bridge this gap, scientists and doctors are looking toward a new kind of computer program called a large language model. These are systems trained on vast amounts of text that can read, write, and answer questions in a way that mimics human conversation. They can process complex medical notes, explain difficult concepts in simple terms, or help organize patient records. The hope is that these tools could act as a force multiplier, allowing a single audiologist to support many more patients or helping people manage their hearing health without waiting for an appointment. However, just because a computer can write a convincing sentence does not mean it can safely guide a medical decision.
A team of researchers from universities in Hong Kong, the United States, and Europe set out to find out exactly how these computer programs are being used in hearing care and whether they are actually helping patients. They did not simply ask if the technology works; they looked for proof that it improves real-world outcomes. The researchers gathered and analyzed 110 different reports published between 2022 and 2026. These reports covered everything from how well the computers could answer questions about hearing loss to whether they could help write medical notes or create rehabilitation exercises. The team sorted these studies into groups based on what the researchers were actually testing: some were just checking if the computer could pass a written exam, others were building small prototypes to see if they worked in a lab, and a few were trying to see how the technology fit into actual patient care.
What the researchers found was a landscape full of potential but short on proof of real-world success. The vast majority of the studies were like a driving test where a car is driven on a closed track with no other traffic. In the hearing care field, this meant that most studies tested the computer's ability to answer specific questions or generate text in a controlled setting, far away from the messy reality of a busy clinic. For example, some studies showed that these programs could answer questions about hearing aids or tinnitus with a high degree of accuracy, or that they could score a speech test almost as well as a human expert. Others showed the technology could help draft medical reports or organize research data. These are useful skills, but they are only the beginning. The researchers noted that while the computer programs could pass these tests, no study in their review had actually tested whether using these tools led to better health outcomes for patients, such as improved hearing, better communication, or higher satisfaction with care.
Crucially, the review highlighted a significant gap between what the technology can do in a test and what it can do in practice. None of the 110 reports described a situation where a hearing clinic actually deployed these computer programs to treat patients over time, nor did any study compare the results of using the technology against the usual way of doing things. The studies that did involve people were small and short-term, often limited to a few weeks or a small group of volunteers. Furthermore, the researchers found that the studies rarely checked for the most important safety issues. Very few reports tested whether the computer programs could be tricked into revealing private patient information or whether they could handle the complex, incomplete data that often appears in real medical records. The privacy safeguards that exist in the real world were not being tested in these studies; instead, researchers mostly asked people if they felt the technology was safe, rather than proving it was.
The researchers also pointed out that the way these computer programs are built and used changes very quickly. A program tested in January might be fundamentally different from the same program in June because the developers update its underlying code, change how it searches for information, or adjust its safety rules. Many of the studies did not report exactly which version of the software they used or when they ran their tests, making it difficult to know if the results would hold true today. This is a major problem because a system that works well in one specific setup might fail in another. The review concluded that while these tools show promise for specific tasks like generating text or scoring simple tests, we do not yet have evidence that they improve the overall quality of hearing care. The path forward, the authors suggest, is to move beyond simple tests and start designing studies that measure whether these tools actually help patients and doctors in their daily lives, while rigorously checking for safety, privacy, and fairness. Until then, the technology remains a powerful tool waiting to be proven in the real world.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.