Scribes, AI or otherwise, are Useless in Healthcare
The problem with EMRs isn’t the time it takes to type; you need discrete values from a database; language is inadequate.
Introduction
Normally we talk about healthcare and health insurance and today is no different. Today we are going to talk about scribes, the things that turn voice into text, AI and otherwise, and how the entire medical records industry has gone and is going down the wrong path. We will present this article in the usual problem and solution format.
Before we get started let’s define a couple of terms. Discrete values are the data points you might pick out of a database; or elements in a drop down list box. For example vanilla, chocolate or rocky road. Continuous values are those that don’t fit into a particular category, like the value of pi (π), or the text of “War and Peace.”
We welcome comments and even arguments in the comments.
The Problem
In 2009, Meaningful Use became the law. Meaningful Use is the Healthcare Information Technology for Economic and Clinical Health (HITECH) was signed into law on December 30, and was aimed at promoting adoption of Electronic Medical Records (EMR) Across all providers. The problem is that this was a government initiative and politicians don’t understand healthcare nor the technology they mandated.. We got a wholly misguided attempt to solve a problem the solvers didn’t and don’t understand and who never think about the unintended consequences.
What we got was paper records, just without the paper.
Language isn’t Adequate to the Task
The way humans transfer information is imprecise at best. I’ll posit that more than half the time we have to ask for clarification of something someone said to us. That wastes time. Then there is interpretation. We wonder what the underlying meaning is behind the words that we hear. “I can see that we have different perspectives on this matter” might become “you’re an idiot.” Finally, minds wander and words just slip out. We say things we don’t mean, we say things that are off topic and generally language is just a bad way to convey meaning.
Let’s think about it this way: medicine is science and language is art. Art is subjective and as such can’t be quantified nor judged. Medicine is science and thus, every action has a definite, reasonable and quantifiable outcome.
Unnecessary Documentation
It might make us feel empathetic to write or type “patient presents with screaming and crying” in the encounter documentation but that phrase is wholly worthless otherwise. We know that you came to the doctor because something hurts or is broken or both. We don’t need to document that, even in the case of a screaming and crying patient.
Nobody is Going to Read the Novel you Typed in as Notes
You are amazing at documentation. You love your patients and they love you. You make sure they get the best care that they can get anywhere and at any time. You spend hours out of your day writing copious, detailed documentation of every encounter.
Nobody will ever read your notes.
You are amazing, but you are wasting your time. Maybe in a soft science like talk therapy, how we feel about a thing is relevant (and maybe in the world of quantum physics too, but that is a discussion for another time). In medicine we are only interested in doing the thing that needs to be done to achieve the outcome we desire. How we feel and think about a thing is irrelevant to the outcome and also how we end up with the antivaxx crowd where thinking and feeling are more important than research and double blind studies.
We have had Speech to Text for Decades
This is a direct challenge to AI. In point 3 of the article “17 Reasons Why We Can’t Have AI” we pointed out that code is not only faster and cheaper, but it is better. Actually writing the application to do the necessary thing is just better. First, AI is a black box so it is as good as it is ever going to get. Sure, you can add “training” but what you are doing is just muddling the already murky waters trying to get a system you don’t understand to do something you can’t define. If you want the system to record your words as text, we have that without AI, have had it for decades and can implement it easily. If you want AI to write something you didn't say, that is a little like saying to your husband “if you don’t know why I am mad at you, I am certainly not going to tell you.”
Stop it.
Workflow Thinking
There is no such thing as workflow. Sure, there are dependencies, you can’t go to surgery before you are prepped, but who, how and where the prepping gets done is a completely different subject. For example, let’s say that Ford makes F-150s (they do). In order to have an F150 to sell, each truck needs a transmission. This is a dependency. Then who, when and how are all open to various methods of being accomplished. Maybe Ford wants to make transmissions. They have and still do produce them. But they also collaborate with GM (!) on the new 9 and 10 speed automatics and they buy transmissions from Tremec, ZF, Getrag and others. There are any number of ways to satisfy the transmission dependency, and none of them include workflow.
I hear doctors every day saying this or that doesn’t fit my workflow. Exactly where did your workflow come from? Did you hire an ergonomics expert to come in and show you how to do it? Why are you so married to it? Do you expect vendors to come in and examine what you are doing and then write software, like the EMR, to fit your workflow? What if another practice or hospital has a different workflow? Here there be monsters. Now you have the expectation of every vendor in the world coming to visit to see how you do things, then modifying their offering for you or writing something from scratch. That doesn’t sound very efficient does it? That sounds like a way to have a large, slow, unreliable, unmaintainable software package that costs $300,000,000+ per installa… I just realized that is exactly what we have with the Epics and Oracle/Cerners and Athenas of the world.
…and they are ALL going to AI scribes.
Specialties
You don’t need specialties in your documentation system. We go to the hospital and there are 130+ specialties, each with their own software. There are pediatricians, dermatologists, neurologists, ad nauseum, and they all have their specialty software. As an illustration, let’s say you bought the F-150 discussed above. When you take it into the dealership, you find they have an engine documentation system and a transmission documentation system and a suspension documentation system and a paint and trim documentation system, and others.
If I found that, I would get rid of that truck immediately.
If having separate documentation systems is dumb at the dealership, why is it smart at the hospital?
I’ll bet those documentation systems cost $300,000,000+ per dealership and are modified or written from scratch for each one..
The Conclusion of the Problem
You can’t use free text to document medicine. Language is too imprecise a tool to achieve the necessary results. Scribes, both AI and traditional code, simply translate speech to text. AI adds an unnecessary layer of complexity over an unnecessary layer of functionality.
The current batch of vendors deal in unnecessary complexity as their stock in trade. We have evidence of this in workflow thinking and the catering to specialties that simply don’t need to exist. They, the big vendors, have not thought about the problem enough to come up with a viable solution. Now they are selling scribes to cover up their mistakes and only complicating and automating the wrong solution.
Scribes are not the answer.
The Solution
The solution of course is to stop and start over. We don't need free text, and therefore we do not need scribes in medicine.
Nomenclature
In 1966 Neil Pappalardo wrote COSTAR in the MUMPS language, (COSTAR and MUMPS research is left as an exercise for the student) and a few years later Judy Faulkner got hold of it and founded Epic. These two really really bright people (sarcasm: the sign of good literature) are more responsible for the morass that medicine finds itself in than any other. They advocate and use free text to document the patient encounter. We find the digitized version of paper notes that have to be typed in, we find language integrated database, we find proprietary code sets, we find specialties in the software.
What we need is a standardized nomenclature suitable for documenting the entire patient encounter. Designed by Arnold Pratt in 1965, SNOMED_CT, the Systematic Nomenclature of Medical and Clinical Terms. was exactly what Judy and Neil needed, and chose not to use. Today SNOMED is included with the Unified Medical Language System (UMLS) and is designed in such a way that anything can be searched for hierarchically, and in a database you only need one query to get a medical term for about anything you can do in medicine.
Values that you can pick out of a database, like UMLS, are called discrete data. Remember that from the introduction? This is as opposed to continuous data. Continuous data can be exemplified with free text, a scribe can’t help you with discrete data.
Why this Matters
Once you eliminate specialties and proprietary ‘code sets’ or text entries (like scribes, AI or otherwise) and put all medical information in one database, then you can start doing things like relating what the patient says to actual symptoms, and those to diagnoses and those to treatments and those to outcomes.
Let’s break this list down individually.
Patient Presents With
Hey, we are patients. We don’t know the difference even between dorsal and ventral. We don’t really know how to describe where or how much it hurts, maybe even only pointing, with no language at all. With discrete values, we can differentiate what the patient says with what the patient means in a very related and structured manner. Then when the patient says “the pain moves around” we can relate that to a diagnosis with a positive outcome.
Symptoms
Now that we have a way to translate what the patient says with what they mean, we can start doing meaningful tests and relate those with diagnoses that have positive outcomes. Imagine having a set of symptoms that just befuddle you and everyone you know. You can type those into the EMR and find a list of patients with those same symptoms and their diagnoses and treatments along with outcomes.
Diagnoses
You are probably noticing a trend here, as we go down the chain. With discrete value based documentation you can search for anything you like, and its outcome. You could even look at any diagnoses and see the average life expectancy afterward.
Treatments
Maybe there is an alternative treatment you get in your list of patients with your selected symptoms and have something to fall back on that worked at least once when all else fails. This is all about the access to information and being able to collate and search it easily and quickly.
Medical Coding
If you document your patient encounters with discrete values, you don’t have to have medical coders. As the documentation is entered in the EMR via the UMLS, it is coded. No, you don’t need to know any codes. Use English, or German, French Swahili or any of several other languages. The average practitioner spends about $77,000 per year, each, on the combination of EMR, compliance reporting and medical coding. We at Sentia can cut this down to zero. We haven’t talked about the economics of our solution, we have done that many times before. Check our channel for that. We offer the EMR for no charge to the practice and integrate compliance reporting so that it is all one click.
Relationships
With free text the typed in notes are associated with a single patient encounter. With UMLS discrete values, they are related to 14 million other clinical terms, the individual patient encounter, the symptoms, the diagnoses, the procedures and he outcomes. It is these relationships that we are after and that provide the value we are looking for.
Conclusion to Why this Matters
With free text and scribes and anything that anyone is doing today, except us at Sentia, you can’t do searches and research as described above. Nothing is related except by being on the same sheet of paper, even when it is virtual paper behind the glass of your monitor. You and your colleagues are doing the same things in the same way, sitting around pondering what the problem could possibly be exactly the same way that doctors have been doing for 4000 years. This is why we say that scribes are useless.
Further, the silos that medical data is kept in because of specialties and the requirements of documentation exacerbate this problem to the point that it becomes unsolvable. If you have a different proprietary ‘code set’ you can’t communicate between silos in the same program, much less between Epic and Oracle Cerner or any other free text or scribe based EMR.
Finally, with the extermination of medical coding, we eliminate one of the biggest hurdles in treatment. That is communicating what was done to the medical coder so he or she can communicate it effectively to the insurance company for adjudication so you can get paid. That is a lot of monkey motion.
There are dozens of other reasons to choose Sentia and we invite you to read all about them. Our goal today was to demonstrate how the big players in the EMR industry are simply clueless and on the wrong track so fundamentally, that they not only can’t solve the documentation problem, they can’t even adequately define it.
Scribes produce text. Text is not what you need. What you need are discrete values from the UMLS and an easy interface to pick them from. See that interface in our article “Sentia Health’s Electronic Medical Record Demonstration.”
Call to Action
We have this system in prototype now, fully functioning.
Contact us here or on our site and we will be happy to provide a demonstration of the fully functional prototype.
If you liked what you read, please like and subscribe, click on the notification icon, subscribe to our newsletter, and follow us on all our social media and blog sites.
We have built a comprehensive health information system to keep the patient healthy and on the right track with the ability to incentivize healthy living. Implementing this system should be fairly simple and will completely revolutionize the way healthcare is paid for, saving countless lives. We have shown a way to use this system to make the best healthcare system in the world also the most efficacious and the most affordable, and a way to move toward value-based care.
