NLP patient portal add-on helps docs read messages faster
NLP helped triage patient portal messages at 81% accuracy, speeding up provider response times by nearly 17 hours.
Clinicians bogged down by a full patient portal inbox might find a solution in natural language processing, according to a group of researchers writing in JAMA Network Open.
The study, completed by a team from the Southern California Permanente Medical Group (SCPMG), found that an NLP tool can accurately flag high-priority patient portal messages, helping providers to sort through their inboxes and return urgent messages faster.
This could be good news for both patients and providers, the researchers said, as patient portal message volumes soar.
"Growing use of secure messaging adds substantially to the administrative burden on clinicians," the team wrote in the report. "Clinicians spend up to 25% more time managing inboxes, with no signs of lessening after the pandemic."
When it takes more time for providers to sort through their inboxes, it naturally takes longer for patients to get answers. And if the message was urgent, this could impede patient safety.
Notably, most patient portal platforms allow for some sort of message sorting, but it relies on patients to self-report the urgency of their query. Indeed, SCPMG's patient portal has this capability, but it can be unreliable and cause further bottlenecks, the researchers said.
But AI could be the fix.
NLP sorts patient portal messages by urgency
Using its own patient portal as a proving ground, the SCPMG researchers looked at how AI -- particularly NLP -- can help sort patient portal messages and whether it could be more reliable than patient self-reports.
Implemented in March of 2023, Smart Messaging Tool (SMT) uses NLP to analyze a patient portal message and categorizes them by topic, which the researchers said helps providers manage their inboxes.
The researchers compared the first two years of SMT's implementation to the performance of the patient portal's legacy system. Specifically, they sought to determine whether SMT could accurately categorize messages and, importantly, improve provider workflows.
In an analysis of more than 3 million messages sent by more than a million unique patients, the researchers found that SMT outperformed traditional systems. SMT achieved 81% accuracy, compared to the 44% accuracy patient self-reports of message urgency achieved.
This resulted in overall better workflows for healthcare providers, who were able to better manage their patient portal inboxes and prioritize patient queries. This resulted in a 17-hour reduction in the time it takes for a provider to respond to high-acuity patient portal messages, from 22 hours to 5 hours.
Smart Messaging Tool prioritizes patient safety
The reduction in response times is an important step forward in patient safety.
Increasingly, patients are relying on patient portal messaging -- or, increasingly, Dr. ChatGPT -- to get answers to even their most serious of medical questions. Making sure providers have the tools to address urgent patient queries quickly is essential not only to patient satisfaction, but overall clinical outcomes.
According to the SCPMG, patient safety was at the heart of SMT's design.
"For instance, it is important for messages describing serious symptoms, such as shortness of breath, to be addressed expeditiously," they explained. "Recognizing the asymmetric cost of misclassification, we configured our models to tolerate higher false-positive rates for high-acuity labels. This design resulted in lower scores on conventional performance metrics, but it supported a more clinically robust system that prioritized timely intervention."
SMT was enhanced by the rollout of the Centralized Messaging Hub at SCPMG. The Hub is staffed by clinicians who are equipped to handle a broad array of patient portal messages. This combination of the AI-powered message triage with the Centralized Messaging Hub has been instrumental in helping SCPMG manage historic levels of patient portal messaging.
NLP message triage offers another AI use case
The NLP message triage is yet another AI use case for patient portals.
While SCPMG's SMT tool helped prioritize messages for clinicians to answer, many other healthcare professionals are tapping AI -- in this case large language models (LLMs) -- to help craft messages.
LLMs have helped skim patient messages and determine an appropriate draft response from providers. Some healthcare professionals have questioned whether a message that still requires review and revision can help save providers' time. However, early research shows that AI-drafted patient portal messages can shave up to 7% off response times.
As AI continues to develop, health IT developers will continue to find various ways in which the technology can support provider workflows. It will be essential for provider users to understand the impacts of the technology and assess whether these tools realistically support better work.
Sara Heath has reported news related to patient engagement and health equity since 2015.
