Innovation trends in healthcare point towards a future of improved communication between care teams.
For Augmedix Co-Founder Ian Shakil and Chief Technology Officer Saurav Chatterjee, focusing on human connection and the clinician-patient relationship is key to transforming medical documentation. Ambient medical documentation products are helping healthcare systems, physician practices, and hospitals save up to three hours per day, increase productivity, improve work-life satisfaction, and increase patient satisfaction.
Discover Shakil and Chatterjee’s unique perspectives on the health system of the future and the essential technology required to improve and optimize communication between care teams.
Tell us more about the state of medical documentation today and the problems healthcare systems have faced.
Saurav Chatterjee: From my perspective, much of today’s medical documentation paradigm was designed for one clinician to convey to another the status of the patient and to power passive billing and coding needs. Digitizing documentation, which started to progress more than a decade ago, is a big step forward with numerous advantages. But the big leap forward from digitization arises from taking a passive documentation paradigm and allowing for digital platforms to trigger opportunities to actively improve patient care. This is made possible due to abilities to analyze massive amounts of information (e.g., being able to identify potential defects in diagnosis or medication).
All that said, the state of current notes, even post-digitization, is haphazard. Some notes are terse and others are comprehensive, some are missing information or listed out information in different layouts, making it hard to use for any purpose other than for another clinician to use. It is hard to feed this haphazardly created, unstructured data into analytics platforms.
Ian Shakil: A typical clinician spends more than one-third of their day on the computer, taking them away from what they care about most: time with patients. This has severe ramifications for the health system and leads to mounting burnout among clinicians, spiking attrition, and reductions in patient throughput which has the overall effect of reducing overall patient access.
Double work is another common problem. With many touchpoints, a patient may have several versions of the same medical note instead of one holistic view that each clinician can add to. Clinicians across the team feel this burden as well, as they redo notes from intake, to care journeys, to discharge, just to name a few areas of massive redundancy. Double work isn’t only wasteful; it creates opportunities for discrepancies and outright errors.
Tell us more about medical documentation of the future and how these problems will be resolved.
Chatterjee: To improve care in the long term, clinicians need to be able to analyze large amounts of information to guide patient care and identify gaps that need to be filled. We are going to see more health systems hire healthcare data teams, but they need structured data to analyze. This is where the medical note needs to evolve from the current unstructured sentences format to a form that is more conducive for analytics. By utilizing both real-time and post-visit data analytics, healthcare teams are able to make higher quality care decisions.
Augmedix uses technology to create the medical note. However, instead of directly creating the unstructured medical note sentences, our technology platform actually first creates a structured data tree, which it then renders into sentences. We can upload the finished sentences to the EHR, but we can provide the structured data for analytics purposes as well.
Shakil: The next era for EHRs data is…to harmonize, reduce double work, structure data so it’s actually useful, and turn the EHR from this ‘feed the beast’ exercise into a copilot to unburdened clinicians so they can focus on patient care. Technology of the future will be able to perform this structured note taking burden passively, in the background, on a ubiquitous basis.
What’s more, the bi-directional pipes that we are laying that can perform this ambient documentation will also prove useful pushing data back in the other direction, from the system of record back to the point of care. I forecast these ambient systems of the future to be elegantly nudging clinicians in real-time to remember to close care gaps they were otherwise forgetting, surfacing helpful patient education, etc. But unlike nagging EHR flags that queue up, these ambient nudges will be responsive based on what’s actually covered at the point of care, and they will be perfectly timed to land before/during/after the right points during the visit.
Tell us more about generative artificial intelligence (AI) and large language models (LLMs) and how they will be utilized in the future.
Chatterjee: LLMs are a large step forward in AI technology and we are very excited about this development. The advantage of large language models (LLMs) is that they take lots of input information to then generate a summary of what was said. With LLMs, alone, unstructured data inputs result in unstructured outputs, which are nice to read, but not always good for maintaining structured data sets that are meaningfully analyzable. Looking ahead to the future of documentation with Augmedix, we’re taking unstructured input to generate structured output. We utilize LLMs as a portion of the Augmedix solution, to add more color to the structured data we provide, ultimately resulting in more color and context while maintaining the fidelity of the core structured note elements. Without advanced LLMs, a lot of the color had to be added by back-end human quality reviewers (which we employ) or by the clinician during final review.
Another consideration about LLMs is how unpredictable the AI can be. There are a lot of cases where it will take input and create the perfect medical note or summarization, but there could be one small change in the input and it generates a completely incorrect, nonsensical output. Since all AI systems are statistical, they face this potential dilemma. It’s very unpredictable and hard to determine how it will perform each time. It’s important that we tune our models to avoid false positives and optimize for real-world de-burdening vs. slick demos that appeal to lay people; Augmedix is on a mission to do precisely that.
Shakil: I’m so enthusiastic about the entire AI and automation boom in healthcare. Within that, there are so many technology modules that are crucial to making Augmedix possible, including automatic speech recognition (ASR), which we’re seeing improve technologically so much every quarter. The various natural language processing (NLP) and natural language understanding (NLU) capabilities are essential in accelerating medical documentation.
Right now, we’re seeing a lot of conversations around LLM and generative AI. For Augmedix, the impacts won’t be as obvious right away on the surface, but the initial implications on the back-end will be huge. Saurav mentioned how LLMs can help us with the context pieces of the note, which ultimately means Augmedix can produce more of the note with more automation and less human intervention. Which means we can provide faster note turnaround in a more scalable and affordable way for our health system customers.
Another area where LLMs come into play for Augmedix involves what I call “re-rendering.” In other words, LLM technology can help:
- Convert a medical note we generated to a patient facing rendition of the note, an after visit summary (AVS)
Convert a standard note that a health system mandates system-wide to a more personalized note that an individual clinician prefers (which may be more verbose, or have more stylistic differences) - Summarize a series of ED notes into a unified discharge note with minimal incremental human burden
- The non-obvious fact about LLM technology is that they are going to be necessary but insufficient for the next frontier of an ambient documentation system. They will help with rendering and contextual automation, but they need to be paired with structured data systems that continue to maintain high confidence discrete data to power robust EHR required fields. Augmedix is laser focused on bridging between these two paradigms in a way that is seamless and delightful for our end users.
Is there anything else you would like to add?
Chatterjee: With healthcare, mistakes are very costly. You always need to check that what is being generated is correct, which is why we’ve implemented guardrails to protect against mistakes. In other domains, if there are mistakes it’s not as big of a deal, but here we are dealing with patient lives, and we want to make sure things are done correctly so clinicians can focus on what matters most— high-quality patient care.
As we look ahead to the future of healthcare, ambient medical documentation and data solutions will seamlessly fit into workflows helping clinicians and patients form a human connection at the point of care without the distraction of technology. By extracting data from natural clinician-patient conversations and converting them into medical notes in real time using ASR, NLP, and LLMs, burnout is reduced, and both clinician and patient satisfaction is enriched with actionable insights that elevate care.
Interested in continuing the conversation about innovation in healthcare? Get in touch with Augmedix representatives to learn more.