The Pros and Cons of Deploying AI to Confront Physician Burnout-The HSB Blog 10/13/23
Artificial intelligence (AI) has the potential to alleviate physician burnout significantly by reducing the amount of time spent on bureaucratic tasks like documentation and reviewing old medical records. AI that utilizes large language models (LLMs), speech recognition, and natural language processing (NLP) can help transcribe conversations between physicians and clinicians into formatted clinical notes and prepare clinical summaries of a patient’s medical history.
This is especially crucial as physicians spend increasing amounts of time utilizing the electronic health record during patient visits and after-hours for documentation purposes. In light of a projected shortage of 124,000 physicians by 2034 and heightened levels of burnout post-pandemic projected by the Association of American Medical Colleges, leveraging AI to minimize bureaucratic burdens is an essential next step. The use of AI to reduce physician burnout through decreasing bureaucratic burdens can allow physicians to dedicate more time to addressing patient concerns, leading to improved patient satisfaction scores and health outcomes.
Physician burnout costs the U.S. approximately $4.6B/yr. due to reduced hours, physician turnover, and expenses of finding and hiring replacements (Harvard Business School)
60% of physicians agree that bureaucratic tasks, including note writing, are the top contributor to physician burnout (Medscape)
Physicians spend almost 50% of their time on the electronic health record (EHR) and desk work with 1-2 hours of after-hours work each night dedicated to EHR tasks (Annals of Internal Medicine)
AI utilizing voice-enabled technology saved clinicians 3.3 hours per week and reduced the amount of time physicians spent reviewing old notes by 60% by producing a clinical summary (AAFP)
Artificial intelligence utilizing LLMs can play a significant role in reducing physician burnout, especially by assisting with the burden of clinical documentation. The advent of new tools from vendors such as Augmedix, Regard, Nuance, and Botco.ai allows healthcare organizations to significantly reduce the administrative burden on physicians. For instance, Nuance’s Dragon Ambient eXperience (DAX) software does this by recording the physician-patient encounter and transcribing it into a formatted clinical note through the utilization of speech recognition and NLP. However, several challenges exist that need to be addressed before further integrating this new technology into the workday of physicians.
One issue is privacy. These novel tools require access to a patient’s protected medical record and will also consolidate new medical information during the patient’s visit. Therefore, there is a potential risk to patient privacy rights, and mistrust in these systems may hinder implementation. Moreover, patients may be wary of sharing information, fearing legal repercussions due to recorded data. According to a global survey by UIPath, only 44% of respondents from the baby boomer generation hold favorable views on AI in the workplace. Physicians, too, may be wary of this new technology as they fear for their job security and that recorded audio may be used against them in malpractice cases. For example, a recent survey of 1,500 physicians by Medscape found that only 19% of physicians would be comfortable using voice technology during a patient consultation.
Another pressing issue is the possibility of error and misinformation. Language generation models can produce inaccurate information which is particularly concerning when it comes to incorrect medical information. “Hallucination” is the term used to describe when an NLP model conjures false information. This phenomenon is more pronounced when multiple languages are used during a clinical encounter or when the technology infers information that the patient did not explicitly verbalize. AI can also omit facts from the patient visit. Dr. Shravani Durbhakula, a pain physician and anesthesiologist at the Johns Hopkins School of Medicine, expressed her reservations, stating, “The major concerns I would have here is I’m not sure the computer would be smart enough to know what is important [enough] to pull out into the note.” She stated the world-class hospital does not use [ambient intelligence tools] to automate clinical notes. “You could miss critical information."
Lastly, artificial intelligence trained on large datasets of text may inadvertently reflect biases present in the training data, perpetuating medical bias. When it comes to note writing, this can be seen in the form of emphasizing certain diagnoses or symptoms for different patient demographics. For example, a New England Journal of Medicine study highlighted this bias when a generative AI model, GTP-4, ranked panic and anxiety disorder higher on its list of potential diagnoses for female patients as compared to male patients. Furthermore, when GTP-4 was asked to generate clinical vignettes of sarcoidosis, the model described a black woman 98% of the time, reflecting a significant bias in its output.
Physician burnout is a huge source of burden on the healthcare system. It can lead to increased medical errors, lower quality of care, worse patient outcomes, and higher attrition rates. For physicians themselves, it can lead to increased rates of depression, substance abuse, suicide, and overall work dissatisfaction. Nearly two-thirds of doctors experience symptoms of burnout following the pandemic according to results published in the Mayo Clinic Proceedings.
A major contributor to physician burnout is the increased administrative burden placed on physicians, namely in the form of documentation required for electronic health record systems (EHRs). A perspective piece published in The New England Journal of Medicine found that for every hour spent on patient interaction, physicians spend an extra 1-2 hours completing notes, ordering labs, prescribing medications, and reviewing results, all without extra compensation. In another paper published in Annals of Internal Medicine, the authors discovered that physicians devote almost 50% of their time to the EHR and desk work, allocating an extra 1-2 hours nightly to EHR tasks. 60% of physicians agree that bureaucratic tasks, including note writing, are the top contributor of physician burnout as per a report published in Medscape.
AI presents a promising solution to significantly reduce the time spent on documentation. An American Academy of Family Physicians (AAFP) report found that AI leveraging voice-enabled technology saved clinicians 3.3 hours per week thereby helping to reduce burnout. Furthermore, the AI was able to reduce the amount of time physicians spent reviewing old notes by 60% through creating clinical summaries. For example, in one case study, Regard’s CEO found that their AI tool reduced measures of burnout by 50% and reduced documentation time by 25%. Augmedix’s website similarly boasts a 40% improvement in work-life satisfaction and a 3-hour per workday reduction.
Despite the burden of bureaucratic tasks like documentation, this is increasingly important in the United States healthcare landscape where the government ties reimbursement to the quality of the medical record. Without proper documentation, physicians do not get paid for the services that they provide patients. The AAFP report found that AI integration resulted in a 25% increase in diagnoses sent to insurance companies that were previously unrecorded in the EHR. Beyond the financial incentive of proper documentation, it also serves to facilitate communication with other healthcare providers, reduces risk management exposure, and captures value-based case metrics.
Conversely, inadequate documentation can lead to adverse treatment decisions, expensive diagnostic studies, repeated studies, unclear communication, inappropriate billing, and poor patient care. While proper documentation is necessary, physicians face significant challenges due to the physician shortage in the United States and the high volume of patients. AI stands as a potential tool to mitigate the aforementioned shortfall of physicians by alleviating physician burnout and attrition. A 2021 JAMA Network Open study found that an AI tool extracting relevant patient health data and presenting it alongside the patient record reduced EHR use time by 18%, a promising tool in reducing burnout. Nuance’s DAX software also showcases promising outcomes, claiming to reduce documentation time by 50% and reduce feelings of burnout and fatigue by 70%.
Physician burnout is a pressing issue that needs to be addressed, especially with nearly two-thirds of doctors experiencing symptoms of burnout according to the New York Times and the impending physician shortage. In addition, physician burnout has significant economic consequences. For example, a study by Harvard Business School found that the economic toll of physician burnout is staggering, amounting to approximately $4.6 billion annually in the United States alone. This financial burden arises from reduced physician work hours, physician turnover, and expenses associated with finding and hiring replacements.
Artificial intelligence can play a significant role in reducing physician burnout by reducing the amount of time physicians spend on documentation, the top contributor of physician burnout. This reduction in burnout can lead to a noticeable improvement in the quality of patient care, enabling physicians to dedicate more time to patients without distraction and consequently improving healthcare outcomes. Furthermore, minimizing documentation-related work during after-hours, often termed “pajama time,” can help mitigate medical errors, a significant concern when physicians' recall and alertness may be compromised.
As AI integration has been found to increase documentation of previously unrecorded diagnoses, physician reimbursement may also become more accurate. While addressing concerns of privacy, error, misinformation, and biases are essential, AI services focused on enhancing the documentation experience are continuously evolving and are poised to play a pivotal role in alleviating physician burnout. As these AI-driven technologies progress, they are bound to enhance the healthcare landscape, ultimately benefiting both healthcare providers and patients.