Can AI Compose Clinical Documentation? Absolutely, but with Limits.
In the fast-paced world of healthcare, clinical documentation is a crucial aspect of patient care and the overall efficiency of medical institutions. However, doctors and medical professionals often find themselves burdened with extensive paperwork, taking away valuable time that could be better spent with patients. Enter Language Model-based Generative Artificial Intelligence (LLM AI), a cutting-edge technology that promises to revolutionize clinical documentation processes and significantly enhance the productivity and accuracy of healthcare professionals.
OK, you’ve probably already guessed that the introductory paragraph was written by LLM AI, specifically OpenAI’s ChatGPT. It’s close enough to what I would have written myself, but was composed in less than 1/100th of the time. That kind of productivity improvement makes the case for deploying LLM AI to handle clinical documentation nearly irresistible, and the world’s tech titans have piled into the opportunity.
For those of us who are not technicians, generative LLM AI uses algorithms and an enormous sample of existing, coherent information to machine-learn the connections and relationships between words. With an adequate set of human instructions called “prompts”, LLM AI can construct sentences, write research papers, create art, develop computer code, and pass a US Medical Licensing Exam. When you combine that kind of intelligent composition power with automatic speech recognition (ASR), LLM AI tools have the potential to dramatically improve the efficiency of clinical documentation processes.
Clinical documentation and provider burnout
While clinical documentation serves a number of functions, four are paramount for most practitioners:
- Patient Care: the record of the medical history, current diagnosis and treatment plan for the patient
- Care Team Communications: a mechanism for the treating physician, nurses, specialists, diagnostics providers, and others to communicate asynchronously about the patient’s care
- Billing Support: accurate documentation to support full and fair reimbursement.
- Legal Defense: a contemporaneous account of a physician’s conclusions, recommendations and disclosures to the patient.
Because thorough and accurate clinical documentation is critical to both patient care and the business of healthcare, getting it right can create a heavy burden for clinicians and medical staff. A 2022 JAMA Internal Medicine report confirmed that physicians are spending nearly two hours per day outside office hours completing clinical documentation. In simplest terms, that is an extra work day every week.
This is not just an inconvenience; it’s having disastrous effects on physician quality of life and mental health. A 2022 New York Times report suggested that one critical factor in physician burnout is the incongruity between the deep mission of personal care and engagement that caused many bright, ambitious students to become doctors, and the impersonal, bureautic, business side of healthcare on the other.
Is AI the answer? Yes – but with limits.
Can LLM AI support effective clinical documentation? The short answer is absolutely, but with limits. A recent academic study of the leading LLM AI models’ ability to read and interpret clinical information demonstrates the incredible power of these tools to support healthcare, but concludes that LLMs should be used as “a supplement to existing workflows rather than as a replacement for human expertise.”
LLM AI offers incredible promise in easing the burden of clinical documentation for doctors and practices, and there are a multitude of reputable companies offering AI driven documentation solutions. In order to evaluate whether AI is right for your organization, there are a handful of questions to ask:
- How exactly does the solution integrate with my existing practice technologies? This is more than just a yes or no question relative to EMR integration. Is the AI clinical documentation solution able to pull information from the EMR for a complete SOAP note (Subjective, Objective, Assessment and Plan elements of a patient encounter)? Does it synchronize with patient schedule information so that patient demographics are automatically included in clinical notes in real time?
- Will an AI derived clinical note be a “Siri-esque” rough first draft that requires a substantial amount of time editing, or will it be a complete SOAP note subject only to physician review and approval? The former output will not provide nearly as much time savings or efficiency as the latter. In today’s state of the “AI art”, however, the latter output likely involves a human-in-the-loop within the service provider to assure “final SOAP note” quality and thus could have a higher initial cost.
- How does the AI solution empower billing and coding effectiveness? It is no longer necessary to settle for, “the physician will draft a summary note of the patient encounter and the coders will figure out how to bill for it.” An integrated AI solution can be designed to capture, near perfectly, every relevant aspect of the patient encounter such that complete and accurate coding is an assured part of the practice workflow.
- Does the output of AI-generated clinical documentation improve the physician’s ability to practice and defend good medicine?By no means am I suggesting that AI LLM can write a better SOAP note than the physician could. But, the majority of physicians do not have the time today to write what s/he would consider their best possible note for every patient encounter. With AI’s ability to learn the individual preferences and styles of each physician, it can empower the doctor to create a near perfect note every time, in a fraction of the time.
The rate of advancement of AI technologies over the past year has been astounding. While philosophers and ethicists may debate whether AI could replace medical decision making in the future, there is no doubt that AI is improving how clinicians and patients experience medical care delivery today. Because AI-supported clinical documentation solutions can eliminate a key driver for physician burnout, we will see health systems and practices using these technologies as competitive differentiators in the effort to attract and retain clinical talent. The power and limitations of LLM AI demand careful assessment to assure that a proposed solution will deliver both the essential functions of clinical documentation and a material savings in physician and staff time.