A research letter published in JAMA earlier this month describes the use of artificial intelligence for something novel: helping sort through scores of messages from patients to their physicians. In this study, researchers worked with the Kaiser Permanente Northern California (KPNC) Research Determination Office to deliberate how AI and natural language processing algorithms can help label message content and sort patient messages to improve the workflow and process of clearing physician inboxes. The goal of the study was to streamline a process by which messages could be labeled into generalized categories (e.g., medication refill request vs. medical query) and also to divert messages to the appropriate role (e.g., pharmacist vs. a physician). The study, which included more than 4 million patient messages, found that leveraging advanced algorithms to manage patient messages had tremendous impact, including providing population-level insights regarding commonly recurring issues and even allowing for faster response times for potentially acute and emergent situations.
Why is this important? With the wide-spread adoption of electronic health records (EHRs) and with more healthcare systems providing their patients with online portal services, the volume of patient messages to physicians has sky-rocketed. The significant benefit to providing these portals is that they enable patients to communicate directly with their clinical teams; however, strenuous physician workflows often do not account for the significant amount of time that it takes to respond to these patient inquiries and messages. Studies indicate that the rate of inbox messages received by physicians has increased by nearly 157% from pre-pandemic levels, posing an incredibly challenging workload effort. Initiatives like this aim to help alleviate some of these burdens.
Rest assured, Kaiser is not the only organization piloting this concept. An American Medical Association article published earlier this month describes how Ochsner Health is working with Epic to leverage Microsoft Azure and OpenAI’s GPT-4 large language model to analyze patient messages and provide better insights to physicians. The algorithms take into consideration important data points from the patient’s record to provide physicians with consolidated insights and potential draft responses. Dr. Jason Hill, MD, MS, and Clinical Innovation Officer for Ochsner Health, explains how systems like this can significantly improve physician workflows by helping “make sure we’re [physicians] drafting a good message to our patients” and by collating scattered information, since the tool “allows us to access the entire health record.”
Dr. Hill admits, however, that these systems are far from being flawless and require human validation: “AI itself is a great tool, but it is not perfect. There are times when it does make mistakes…That is why you must make sure that you have a human reviewing these messages.”
Indeed, given how immature this technology is and how much clinical validation is still required, the quickly growing AI industry has coined the phrase “human in the loop” for this very reason. As these systems are capable of significant hallucinations (the generation of fake results to queries), it is essential to proceed with them cautiously and have human validation of all results and steps. Especially given that these systems are increasingly being leveraged in administrative contexts that have real impact, accuracy and consistency are crucial. Nevertheless, if developed correctly with the appropriate validations and safety measures in place, this technology has immense potential to significantly transform clinician workflows.
link