It’s an open secret within the physician community but may not be widely known: Doctors spend more time documenting and updating a patient’s electronic health record than we spend examining and speaking with the actual patient. According to a 2020 study in the Annals of Internal Medicine, doctors devote an average of 16 minutes and 14 seconds per patient encounter, with chart review (33%), documentation (24%), and ordering (17%) taking up the bulk of that time – leaving us less than 5 minutes for direct interaction with the patient.

This is a problem because while creating documentation is time-consuming, it’s crucial to our patients’ health. And our existing solutions still aren’t up to the task. Worse, many of the proposed solutions may misunderstand the problem.

Consider a woman in her early 40s, who recently checked into an emergency room where I practice, complaining of fatigue and difficulty sleeping:

As I examined her, the story of her reported symptoms became far more complex: Cough, fevers, and shortness of breath. It expanded into her family’s history with health and diet, her life insurance policy, the funeral of an aunt who recently died due to lung cancer and past visits with her primary physician, who diagnosed her as having atypical pneumonia.

On further questioning, the real reason for her visit to the emergency room finally emerged: She was worried that she also had lung cancer and was concerned that her primary care doctor had not thoroughly examined her for that.

In her case, I had enough time to drill down to the core of her concerns. Exhaustively documenting my patient’s entire visit and all her reported symptoms would have produced an accurate description of what transpired. However, it would also include several red herrings, and might not clearly convey the crux of her visit: She wanted her physician’s assurance that she does not have lung cancer.

Technology has increased documentation time, with doctors providing more attention to the computer or tablet than the patient. To mitigate the demands imposed by technology, physicians hire scribes, develop macros, build off templates, and leverage speech recognition to create a patient’s encounter note. There are a number of solutions that try to address the problem through sophisticated ambient scribing technology that listens to the whole encounter, and then automatically creates a clinical document for the physician.

I believe these ambient scribing solutions are technical marvels, but in the end, don’t address the root problem:

The fundamental goal is to provide an easier way for physicians to take the mental model in their head and put it into the patient’s digital file.

Ambient scribing, by contrast, provides a transcript of what occurred rather than a representation of the physician’s mental model of the patient. The distinction I’m making is comparable to a court recorder versus the judge issuing an opinion. The recorder creates a raw transcript; the judge uses that transcript to create a mental model – effectively an expert narrative that discards facts that are extraneous to the case.

Ambient scribing is often proposed as a way to improve documentation, but it also solves the wrong problem. Video cameras and auto-dictation, for instance, can theoretically do that, but the burden would still be on the doctor interpreting it all. (Assuming, that is, they had enough time to review the full transcript.)

So the real challenge with more efficient documentation is taking relevant subjective information from the patient, objective data from medical records, examination, and testing, and using expert clinical opinion to formulate a cohesive medical narrative of the entire encounter. And no matter how amazing AI becomes, ambient scribing can only get better at creating a more accurate transcription—it does not include the unspoken thoughts and expertise of the clinician.

As examples of startups that tackle the documentation challenge from different directions, here’s several with significant promise:

• Augmedix provides a hybrid solution by combining artificial intelligence, natural language processing, and a virtual scribe who can ask the clinician questions to clarify their mental model for the documentation.

• Suki uses a combination of machine learning and natural language processing to convert physician speech into full clinical documentation, with an optional review of notes by a transcriptionist to make any necessary edits. Essentially, their system learns from each physician how to transform a physician’s short verbal expressions into a fully documented mental model.

• Abridge deploys an ambient listening technology that effectively acts like a physician’s court reporter. It listens to the conversation, similar to ambient scribing – and then creates a structured summary that the physician can refer to while writing their clinical documents at the end of the day.

• Decoded Health has a sophisticated clinical artificial intelligence that thinks like a doctor. Based on the patient’s symptoms, medical history, and objective findings in the encounter, it generates a graphic interface of the mental model of the patient that the physician modifies to accurately reflect what they are thinking. This model then allows the software to develop a physician document that sounds like the physician and is a coherent narrative.

In full disclosure, the latter two are being cultivated at Inflect Health, an innovation hub, where I am an advisor.  As I say, it is very hard to find players in this space addressing the problem from the mental model framework, but we hope to see promising results from startups like these soon.

As for my patient? Fortunately, her symptoms resolved after 3 days of antibiotics, and her chest x-ray and other exams were grossly unremarkable, showing no indications of malignant tumors. And through my documentation, her primary care physician now knew that cancer had been the unaddressed concern of hers all along.

The task ahead now is understanding the true problem, so that a solution that automates the physician documentation task can be adopted by our industry – giving us more time to devote our attention to where it really matters: With our patients.

Photo: megaflopp, Getty Images

The post has been updated with a more accurate description of Suki’s technology.