Top ways to use AI in patient portal messaging

Patient portal message triage

A full patient portal inbox is a source of stress for healthcare providers confronting heavy workloads. Complicating matters is the reality that not every patient portal message is intended for the physician or best handled by them.

Indeed, some patient portal messages are about billing, while others pertain to low acuity concerns better handled by a nurse or advanced practice provider. Instead of letting the task of portal message triage continue to overburden providers, some have tapped AI and machine learning to segment out the messages to the right respondent.

In June 2020, researchers from Vanderbilt University Medical Center took this a step further and found that machine learning can also help patient portals classify messages by urgency. They tested two machine learning models and found that they were effective at assigning risk to different types of messages, which can help providers prioritize their inboxes.

This ensures patients get their needs met in a timely manner, a separate study from NYU Langone Health found. When the researchers installed a large language model into the EHR, the system was effectively able to triage patient portal messages that needed more urgent review and response. This, in turn, sped up the response time, which could have positive impacts on patient experience and outcomes.

Streamlining patient portal responses

Perhaps most notably, gen AI has proven effective at responding to patient portal messages. The technology can effectively assess a patient query or request, summarize a provider response and deliver the response with empathy and health literacy in mind.

In 2024, researchers from Mass General Brigham said gen AI can provide a "best of both worlds" scenario in which patients can get satisfactory answers to their patient portal messages without adding to clinician workloads.

The researchers, writing in The Lancet Digital Health, compared AI and human responses to 100 patient portal messages, finding that the AI-generated answers were comparable to those written by humans. In fact, about a third of the time, human reviewers couldn't tell the difference between answers written by AI and those written by a person.

More than half the time, AI responses did not need human editing. Meanwhile, reviewers said AI responses were usually more detailed than those written by humans.

However, the researchers cautioned against solely using gen AI to craft patient portal messages. Although data has shown AI can send empathic messages to patients and is careful to consider health literacy levels, the tools still are not immune to algorithmic bias. That risk, plus risks for inaccurate information, indicates that gen AI patient portal messages need human review before sending.

Still, gen AI could be an answer to a critical clinician workload problem: time in the EHR. In a 2023 JAMIA analysis, researchers said that patient portal messages skyrocketed during the pandemic. Each patient portal message was tied to a 2.3-minute increase in EHR time each day, on top of already huge documentation burdens.

Although experts caution against fully relying on gen AI to respond to patient portal messages, the technology could streamline provider workflows and reduce provider burden.

Accommodating varying patient health literacy

AI isn't just useful in generating a patient portal message response; data shows that AI can translate medical jargon that's often confusing for patients parsing through patient portal messages. This is key as healthcare providers work to meet varying levels of patient health literacy.

Data shows that patient portal messages and clinical notes riddled with medical jargon and complex terminology can be confusing for patients. This can have downstream consequences of disengaging patients or even misinforming them.

The good news is that simply removing medical jargon or medical abbreviations can boost the readability of patient portal messages, separate data has shown. Medical notes that use plain language are more accessible for laypeople and patients.

But those jargon-free notes are easier said than done. Healthcare providers need to document in clinical notes and often give medical advice using complex medical terminology. Translating those terms is important for patient engagement but can add to already high provider workloads.

AI has proven useful at translating medical jargon. Researchers from NYU Langone Health used a version of GPT-4 to translate 50 patient discharge summaries, lowering the reading level of the notes from an 11th grade reading level to a sixth grade reading level. Those translated using AI also scored higher on the Patient Education Materials Assessment tool.

Ideally, similar systems can be implemented for patient portal messaging.

What's more, gen AI can generate patient portal response that avoid the issue of medical jargon in the first place. A 2023 assessment of ChatGPT showed that it can respond to patient queries in a way that is understandable for most patients, taking into consideration patient health literacy levels.

Assist patients in crafting patient portal messages

In addition to helping healthcare providers write patient portal responses, some researchers have found that AI prompts can also help patients better compose their own messages.

In particular, AI prompts can help patients ask more specific or clearer questions when messaging their providers. This is essential as healthcare professionals seek to cut down on time spent in the EHR and patient portal. By knowing exactly what a patient needs, healthcare providers can get to the heart of the issue more quickly.

Researchers from VUMC tested this theory by designing a custom-trained gen AI model that could produce follow-up questions to a patient query. Reviewers regarded these follow-up messages as clear and concise, the researchers said.

In theory, the tool could be adapted to help patients refine their messages before hitting send, similar to how AI has been engineered to help people compose emails or text messages.

"Follow-up questions can be adapted to the patient’s specific situation through prompt engineering to provide more personalized care, such as education, familiarity with medical terminology and the patient’s own medical history stored in the EHR," the researchers wrote in JAMIA.

This approach of creating efficiency on the patient's side has some benefits over using gen AI to draft provider responses, the researchers argued. For example, AI draft responses could disrupt the systems clinicians already have in place for managing their patient portal inboxes and replying to patients. Moreover, AI-drafted responses from providers carry some risk for inaccuracies.

Do's and don'ts of using AI for patient portal messaging

Using gen AI for patient portal messaging is very new, meaning there's little information about how the technology can actually affect patient satisfaction and clinical outcomes. As such, healthcare providers should practice caution when utilizing the technology.

Foremost, nearly every study investigating AI use in the patient portal urges healthcare providers to review any information generated by the AI. This doesn't just ensure the information provided is accurate and relevant, although that is important. It also adds the human touch most patients want in their healthcare interactions.

For example, providers who review AI-composed portal messages can edit the response to be written in their own voice. It is important that patients feel empathy from their healthcare providers and to know that their clinicians are involved and engaged in their care.

To that end, healthcare providers might consider telling patients that a portal message has been composed or aided by AI.

This is an ethical conundrum challenging healthcare providers right now. Patient satisfaction with AI messages might go down if they know they were, in fact, written by the technology. However, not disclosing that fact might feel dishonest or ethically murky for some providers.

A 2025 review by researchers at Duke Health showed that it won't significantly harm patient satisfaction if providers disclose their use of AI to patients.

Patients like messages written by both their providers and AI, with perhaps a slight preference for the AI, the researchers wrote. Although that preference diminishes a bit when patients are aware that AI wrote the patient portal message, the researchers said it's not enough to caution providers away from disclosing that fact.

"Ultimately these findings give us confidence to use technologies like this to potentially help our clinicians reduce burnout, while still doing the right thing and telling our patients when we use AI," said Anand Chowdhury, M.D., an assistant professor in the Department of Medicine at Duke University School of Medicine and the study's senior author.

AI is evolving at a breakneck speed, meaning the impacts it can have on healthcare and the patient experience are only just being uncovered. As healthcare providers and health IT developers continue to experiment with the tool, it is likely it will continue to refine patient portal utilization and messaging.

As those patient portal use cases continue to come to light, healthcare providers will need to learn how to best implement strategies into their own clinical workflows while maintaining a positive digital patient-provider relationship.

Sara Heath has reported news related to patient engagement and health equity since 2015.