AI in Medical Documentation: Solving Physician Burnout and Improving Patient Outcomes
Last year in the U.S. investments into the AI health sector increased by more than 30% from 6 billion USD in 2020 to 8 billion USD in 2021. Investors recognize the enormous potential that artificial intelligence (AI) solutions can offer to improve every aspect of healthcare. AI can reduce process inefficiencies, reduce costs and improve patient care.
At the same time we are seeing an explosion of data coming not only from EHRs, but also from patients themselves through consumer wearables and health apps. An average of 250 health apps are released every day, and over 350k health apps are currently available. AI thrives on data. The more data it can access and the more accurate and contextual that data is the better the results it will have.
Will AI solve the burden of overworked and burnt-out clinicians?
AI has the potential to decrease charting-related administrative waste by freeing up clinicians’ time to focus on their patients.
Even before the COVID pandemic, almost half of U.S. physicians reported feelings of burnout. Because of COVID, burnout is now more common among intensivists. However, historically, clinicians who work in primary care specialties are the most affected. These include family medicine, obstetrics and gynecology, internal medicine and pediatrics.
Developing solutions that address physician burnout is very important. Burnout can lead to depression and suicide. In the U.S. the prevalence of depression among physicians is two times higher than in the general population. On average, a physician commits suicide every day in the U.S.
There are many contributors to physician burnout, but one of the main causes is medical documentation. Entering data into EHRs is time-consuming and shifts the focus from patient care to administrative tasks. Clinicians have become professional data entry clerks, which reduces their ability to practice at the top of their license.
During an average office day, physicians spend about two hours on EHR and desk work for every hour of direct clinical face time with patients. They also spend another one to two hours of personal time each night doing practice-related computer and other clerical work.
Why is the patient not the primary focus of attention during a typical office visit?
To have a better understanding of what medical documentation entails, let’s dissect the process of charting. The average office visit in primary care lasts about 20 minutes. The doctor typically spends more than three quarters of that time (16 minutes) interacting with the EHR. A third of this EHR time is spent pre-charting, reviewing the chart and trying to understand what’s going on with the patient in front of them (reviewing previous visits, notes from other doctors, lab results, etc.). Almost a quarter of this EHR time is spent on actual documentation: writing the note and entering other data, such as diagnosis and billing codes. Once the doctor has an assessment of the problems being addressed during the visit and has a treatment plan, the remaining time is spent placing orders on the EHR.
As you can see, primary care clinicians spend more time “data mining” and finding information than writing the note itself.
Even before EHRs were fully adopted, physicians relied on third parties, from staff to transcription services, to complete their medical documentation. While medical transcriptionists no longer are mainstream, medical scribes are widely used. In fact, almost 20% of physicians in the U.S. use scribes. And speech recognition apps have become even more popular. More than 60% of physicians in the U.S. use these tools to dictate their notes. Another low-tech tool to expedite medical documentation is problem-oriented templates. They not only decrease charting time, but also improve the quality of the note.
Scribes and transcription services not only are expensive, but also have unique limitations. They both are labor-intensive and hard to scale. Transcription turnaround time for completed notes can take up to 24 hours. Onboarding a scribe is time-consuming, and these positions have high rates of turnover, which can make the quality and consistency of the service unreliable.
Dictating a note using speech recognition has many advantages over typing a note using a keyboard. For example, a physician, on average, can type 35 words per minute (WPM). On the other hand, speech recognition software can easily transcribe over 150 WPM.
AI has had a tremendous positive impact on the accuracy and quality of speech recognition by leveraging deep learning to achieve voice-to-text conversion and include some natural language understanding. This next-generation voice recognition has the ability to determine the user’s intent, and it also can understand the user’s voice commands.
This high-level natural language processing (NLP) task enables these tools to act as a virtual doctor’s assistant by using speech recognition and artificial intelligence to populate a note in the EHR without navigating the EHR user interface. Many companies are now offering solutions that use deep-learning NLP. However, most of them still require a human in the loop to assure quality of medical documentation.
Ambient clinical intelligence (ACI) represents the future of AI in medical documentation. It removes the keyboard and mouse from the doctor-patient interaction (assuming the patient speaks English), it captures a multi‑party conversation, and it creates clinical documentation automatically by using text summarization. Reports from companies offering these tools show more than a 50% reduction in documentation time and physician burnout. Although these are obviously the results of small studies, there is real potential in this elegant solution.
ACI does have limitations. For example, the nuances of facial expression, body language and other non-verbal cues are aspects of a clinician-patient conversation that cannot be captured by ambient listening. However, not everything that transpires in a visit needs to be documented. ACI would still need cues from the clinician to flag what’s relevant and document any assessment and plan that the clinician might be considering but didn’t communicate to the patient. The environment itself can also be a barrier; trying to capture audio in a busy emergency room can be a nightmare.
A key challenge to this innovation is the willingness and infrastructure of EHR companies to support integrating third-party applications. Traditional EHRs are well-known for limiting innovation with their restrictive interoperability.
And although EHRs connections with outside systems have improved, there are different levels of interoperability. The basic level is foundational, where a completed note can be delivered to the EHR. However, a structural integration, where discrete data elements can be exchanged, is needed so that AI assistants can pull out patient context and complete tasks.
For example, a doctor could give a voice command to send a prescription or order a lab. This can save precious time during the visit and enhance the doctor-patient interaction. A study on EHR usability showed that it can take up to 62 clicks for a doctor to place an order for Tylenol.
What are the three main goals of medical documentation?
Medical documentation has three main goals. One is to support continuity of care by communicating to the rest of the team and staff what’s going on with the patient. The second is billing; the documentation includes the necessary ICD-10, E/M and CPT codes. The third one is legal protection, including defense against malpractice claims.
Medicare, with its “patients over paperwork initiative,” updated its office-visits coding guidelines last year to make them simpler and more flexible. Revisions included eliminating history and physical exam documentation as elements for code selection. A key change in the guidelines is that now ancillary staff or even the patient can record the chief complaint or the history of present illness; the patient could complete a questionnaire, and the answers could become part of the note.
This opens up possibilities. You could use an electronic form with questions for the patient to fill out prior to the visit. But what if instead of a simple form, you could use AI conversational agents to collect patient information? These agents could be text-based, like chatbots, or voice-based using smart speakers. Almost everyone now has a smartphone, and up to 35% of Americans use some sort of smart speaker on a daily basis. There are limitations; Alexa does not always understand what we want, and we have to repeat ourselves. However, natural language processing is advancing rapidly.
Another aspect of medical documentation in which AI is having an impact is using NLP to analyze unstructured medical text from a note and to automatically extract information like medications or medical conditions. When we link these entities to medical ontologies, we can suggest actionable insights to the clinician. These insights can support clinical decision-making and improve patient care.
By analyzing and extracting meaning from the clinical notes, we can automate content population in EHRs (for example, ICD-10 codes). This can have a great impact on the revenue cycle.
Ultimately, the goal is for AI to help clinicians with all these administrative burdens so they can focus on the patient and improve the patient’s experience. AI is transforming healthcare and is giving us the chance to re-imagine medical documentation. Because medical notes by themselves don’t provide healthcare, humans do.
Chartnote is revolutionizing medical documentation one note at a time by making voice-recognition and thousands of templates available to any clinician. We know first-hand that completing notes while treating patients is time-consuming and an epic challenge. Chartnote was developed as a complementary EHR solution to write your SOAP notes faster. We partner with Augnito to provide next-generation AI voice recognition.
Focus on what matters most. Sign up for a free account: chartnote.com