Using Artificial Intelligence to Improve Clinicians’ Experiences with the Electronic Health Record

The electronic health record (EHR) is a digital diary of a patient’s health. In one location it contains all the materials that health care providers have ordered and assessed as a vital part of helping patients maintain health, manage illness, or plan for the future. These materials can include images such as CT scans and MRIs, reports about biopsies and surgeries, test results of blood and other body fluids, and the critical narrative by doctors collecting all the materials into an assessment and a plan.

Health care providers spend an enormous amount of time on EHRs every day, documenting their actions, thoughts, and orders. It is known to be burdensome, and there is widespread interest in making it less so. For every hour providers spend face-to-face with patients, they spend two more hours in front of the EHR, largely on documentation.^1-2

Beginning in September 2019, Stanford Medicine and Google Health began a thorough study of the experiences of primary care providers (PCPs) with the EHR both at Stanford (50 PCPs) and across the U.S. (204 PCPs). The goal was to learn how these PCPs felt about the possibility of having artificial intelligence (AI) help them with their documentation duties, as this can help inform the design and development of AI technology for  clinicians; technology developers and artificial intelligence scientists; and health systems and their administrators.

Before beginning to determine which activities might benefit from AI, it was essential to understand the precise components of the documentation workflow. “I was surprised how incredibly complicated the documentation process is once you truly dissect it,” says Steven Lin, MD, senior author of the whitepaper³ that resulted from the study, Executive Director of the Stanford Healthcare AI Applied Research Team (HEA₃RT), and clinical associate professor of primary care and population health.

He continues: “We took the process of documentation from end to end, breaking it into distinct tasks. Many tasks are cognitive and many are clerical. In general, clinicians want assistance for clerically burdensome tasks, not for cognitive tasks, although there are some areas of overlap. The assessment and plan, for instance, is a meaty part of the document that clinicians say is cognitively helpful but also clerically burdensome.”

Lin explains that the whitepaper’s authors knew from prior studies and from human scribe experiences that “most physicians are happy to hand over parts of the process to a clerk. We dove deep into what physicians want to offload and what to keep, assuming that the level of granularity might help inform the design of next-generation AI technologies.”

Providers thought that AI-enabled documentation could save them time, improve the quality of care, and provide accurate, high quality clinical notes, when focused on the right types of tasks for assistance.

The idea of having assistance with documentation has been around for years. Since 2015, Stanford has had a unique program using scribes who work closely with primary care providers who provide mentoring and research opportunities to the scribes, almost all of whom plan a career as medical providers themselves, in return for their help with transcription, which greatly reduces the in-clinic and after-hours workloads of the physicians. There are also transcription services offered by companies that, according to Lin, “basically document the entire doctor/patient conversation word for word, which is not helpful.” Cost is a consideration for both of these types of documentation assistance.

Some challenges with designing AI assistance to PCPs are specific to the population being surveyed. Lin points out that “doctors are about how they do their documentation. To satisfy such a heterogeneous group of individuals in general is very hard.” The Stanford-Google Health survey was designed and given to the PCPs to ferret out specific challenges  with the EHR and the level of assistance they would accept from AI technology. There is overall agreement, according to Lin, that “the technology has to be totally inconspicuous.”

In healthcare, it is vital that AI research be guided by human considerations

The research lead for the project at Google, Lauren Wilcox, PhD, saw the opportunity to partner with Stanford as a way to identify goals for AI assistance applied to healthcare. As she explains, “In healthcare, it is vital that AI research be guided by human considerations: what clinicians and patients need, how they work together to meet patients’ goals, and where the greatest opportunities to support care are. Partnering with Stanford and Dr. Lin’s team was a unique opportunity to research the possibilities of combining high-quality care and AI assistance.”

AI assistance in documenting the EHR is still evolving. Lin describes the steps in this way: “The current state – the first generation -- is simply record keeping. In the second generation we will get specific about what we can and want to automate. It will be intentional about helping patients better understand what was done during their appointment and what their doctor said. The third generation will be a true virtual assistant, an intelligent sidekick that gets smarter with each patient encounter, custom tailored to an individual physician’s practice style and preferences, not just with documentation but also with orders.”

Results from the study can be found here.

¹ Sinsky C, Colligan L, Li L, et al. Allocation of physician time in ambulatory practice: a time and motion study in 4 specialties. Ann Intern Med. 2016;165:753-760.

² Lin S. The present and future of team documentation: the role of patients, families, and artificial intelligence. Mayo Clin Proc. 2020;95:852-855.

³  Hong G, Wilcox L, Sattler A, Thomas S, Gonzalez N, Smith M, Hernandez J, Smith M, Lin S, Harrington R. (2020). Clinicians’ Experiences with EHR Documentation and Attitudes Toward AI-Assisted Documentation.