AI-drafted patient portal messages save 7% on response times

Does AI make patient portal messaging more efficient?

The report, which analyzed more than 55,000 patient messages sent between October 2023 and August 2024, sought to assess general AI utilization patterns in patient portal messaging.

Overall, about a fifth of providers selected the "Start with Draft" option when beginning a message response. However, uptake was slow at the beginning of the study period. Only 12% of providers opted to use AI to draft a message response in October 2023, with that figure slowly rising to about 20% as the tool became smarter.

Using the draft responses tended to increase efficiency, at least in terms of response times, with data showing shorter response times for providers using the AI draft responses than those not using them.

However, the researchers cautioned that cognitive burden might still increase when leveraging AI-drafted patient portal messages. Best practice states that providers should review all AI-generated information, including that's sent via patient portal message. Indeed, the researchers noted that draft responses still pop up when writing a response, even when a provider does not intend to use it.

"While it is difficult to determine with certainty whether HCPs read drafts in such cases, our estimates suggest that when drafts are reviewed, they add a significant burden, increasing the time spent on each message by 135.42% compared to reading the patient message alone," the researchers said in the study's discussion.

The study authors suggested a targeted AI deployment strategy. Designing AI systems that only generate draft responses for certain types of patient portal messages -- perhaps basic messages for which providers are most likely to use an AI draft response -- could be key moving forward.

What types of AI draft responses do clinicians like?

In addition to analyzing efficiency trends, the researchers looked at the types of AI draft responses providers tend to gravitate toward.

Across provider type, AI draft responses that were shorter, more readable and informative tended to be preferred.

However, different types of providers used different types of AI responses. For example, physicians preferred messages that were concise and used neutral language. Conversely, support staff preferred messages that had a warmer tone. The researchers said AI systems could be more efficient by tailoring tone based on the user's profession or title.

Still, AI holds great potential, specifically in terms of bridging the expertise gap across providers.

Many patient portal messages get filtered to physicians and advanced practice providers, such as nurse practitioners or physician assistants/associates. The researchers observed that AI drafts allowed more messages to be filtered to support staff, potentially lessening the burden on clinicians.

AI-drafted responses have a learning curve

The researchers cautioned against overlying on AI-drafted patient portal responses. After all, they only have about 20% uptake, and there is still healthy skepticism from providers.

Instead, the health IT industry needs to continue learning more about how healthcare providers prefer to use the technology and allow that to guide development and implementation, advised Oded Nov, the Morton L. Topfer Professor of Technology Management at the NYU Tandon School of Engineering.

"[Large language models] are a new technology that can help providers be more responsive, more effective and more efficient in their communication with their patients," Nov said in the press release. "The more we understand who uses it and why, the better we can leverage it."

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