Large language models (LLMs) have generated buzz in the medical industry for their ability to pass medical exams and reduce documentation burdens on clinicians, but this emerging technology also holds promise to truly put patients at the center of healthcare.
An LLM is a form of artificial intelligence that can generate human-like text and functions as a form of an input – output machine, according to Stanford Medicine. The input is a text prompt, and the output is represented by a text-based response powered by an algorithm that swiftly sifts through and condenses billions of data points into the most probable answer, based on available information.
LLMs bring great potential to help the healthcare industry center care around patients’ needs by improving communication, access, and engagement. However, LLMs also present significant challenges associated with privacy and bias that also must be considered.
Three major patient-care advantages of LLMs
Because LLMs such as ChatGPT display human-like abilities to create comprehensive and intelligible responses to complex inquiries, they offer an opportunity to advance the delivery of healthcare, according to a report in JAMA Health Forum. Following are three major benefits LLMs can deliver for patient care:
LLMs have opened a new world of possibilities regarding the care that patients can access and how they access it. For example, LLMs can be used to direct patients to the right level of care at the right time, a much-needed resource given that 88% of U.S. adults lack sufficient healthcare literacy to navigate healthcare systems, per a recent survey. Additionally, LLMs can simplify educational materials about specific medical conditions, while also offering functionality such as text-to-speech to boost care access for patients with disabilities. Further, LLMs’ ability to translate languages quickly and accurately can make healthcare more accessible.
- Increasing personalization of care
The healthcare industry has long sought to find avenues to deliver care that is truly personalized to each patient. However, historically, factors such as clinician shortages, financial constraints, and overburdened systems have largely prevented the industry from accomplishing this goal.
Now, though, personalized care has come closer to reality with the emergence of LLMs, due to the technology’s ability to analyze large volumes of patient data, such as genetic makeup, lifestyle, medical history, and current medications. By accounting for these factors for each patient, LLMs can perform several personalization functions, such as flagging potential risks, suggesting preventive care checkups, and developing tailored treatment plans for patients with chronic conditions. One notable example is a recent article on hemodialysis that highlights the effective use of generative AI in addressing the challenges that nephrologists face in creating personalized patient treatment plans.
- Boosting patient engagement
Better patient engagement generally leads to better health outcomes as patients take more ownership of their health decisions. Patients who exhibit better adherence to treatment plans obtain more frequent and effective preventive services, which creates better long-term outcomes.
To help drive better engagement, LLMs can handle simple tasks that are time-consuming for providers and tedious for patients. These include appointment scheduling, reminders, and follow-up communication. Offloading these functions to LLMs eases administrative burdens on providers while also tailoring care for individual patients.
LLMs: Proceed with caution
It is easy to get swept away in all the hype and enthusiasm around LLMs in healthcare, but we must always keep in mind that the ultimate focus of any new technology is to facilitate the delivery of medical care in a way that improves patient outcomes while protecting privacy and security. Therefore, it is imperative that we are open and upfront about the potential limitations and risks associated with LLMs and AI.
Because LLMs generate output by analyzing vast amounts of text and then predicting the words most likely to come next, they have potential to include biases and inaccuracies in their outputs. Biases may occur when LLMs draw conclusions from data in which certain demographics are underrepresented, for example, leading to inaccuracies in responses.
Of particular concern are hallucinations, or “outputs from an LLM that are contextually implausible, inconsistent with the real world, and unfaithful to the input,” per a recently published paper. Hallucinations by LLMs can potentially do harm to patients by delivering inaccurate diagnoses or recommending improper treatment plans.
To guard against these problems, it is essential that LLMs, like any other AI tools, are subject to rigorous testing and validation. An option to help accomplish this is to include medical professionals in the development, evaluation, and application of LLM outputs.
All healthcare technology stakeholders must recognize and address patient privacy and security concerns, and LLM developers are no different: LLM creators must be transparent with patients and the industry about how their technologies function and the potential risks they present.
For example, one study suggests that LLMs could compromise patient privacy because they work by “memorizing” vast quantities of data. In this scenario, the technology could “recycle” private patient data that it was trained on and later make that data public.
To prevent these occurrences, LLM developers must consider security risks and ensure compliance with regulatory requirements, such as the Healthcare Insurance Portability and Accountability Act (HIPAA). Developers may consider anonymizing training data so that no person is identifiable through their personal data, and ensuring that data is collected, stored, and used correctly and with explicit consent.
We are in an exciting time for healthcare as new technologies such as LLMs and AI could lead to better ways of delivering patient care that drive improved access, personalization, and engagement for patients. To ensure that these technologies reach their full potential, however, it is critical that we begin by engaging in honest discussions about their risks and limitations.
Photo: Carol Yepes, Getty Images
