When ChatGPT launched in November 2022, there were widespread collective responses of “Wow, this is cool”, and “Dear God, society is doomed”. Nearly four years later, I think it’s safe to say that those expressions remain true. It should come as no surprise that since healthcare in the US makes up about 18% of GDP, we can expect that artificial intelligence is going to have a significant impact on the system. This, of course means that healthcare IT organizations are already responding to many innovations at various stages to help with clinical, administrative, and operational functions. In my current role as a clinical analyst, I have implemented some of these technologies, and have worked closely with leaders in my organization to assess the way forward for the short and medium term. Here are some of the areas where I see AI having an impact in healthcare IT:
Clinical Documentation – AI Interpreted Progress Notes
For years, one of the biggest frustrations among healthcare providers has been the mental and administrative burden of creating clinical notes in electronic medical records systems. These progress notes have typically followed the SOAP pattern: Subjective, Objective, Assessment, and Plan. Over time, various entities (insurance companies, government health agencies, healthcare organizations) have demanded more detailed documentation of patient visits. The term note bloat has derisively been used to express the frustration with burdensome and sometimes unnecessary documentation. For some healthcare organizations that have had the money, human scribes have sometimes been used to transcribe progress notes during visits. Just about every patient has had an experience with a provider spending more time pecking on the keyboard than doing actual patient care.
In 2022, several companies released AI products that were built that process the audio during clinical visits to create progress notes that meet most of the formatting requirements for clinical documentation. This is commonly called ambient listening. Two of the major players in this field are Microsoft Nuance DAX, and Abridge. I mention them because I implemented both of them for the organization that I work for. Other AI scribe vendors are NoteMD, Athelas Scribe, Doximity, and Sunoh.ai.
The way that that the products work for clinicians usually centers around a mobile phone app that has a record button that is pressed during the patient visit, of course by first getting consent from the patient. The app just quietly records the entire conversation during the visit, that may go something like this:
“Hello Sam, good to see you again. It looks like you had an injury to your wrist?”
“Yes, Doc. I was walking in the park this morning, when out of nowhere, a crazed duck came from my right and hit me, which caused me to fall and hurt my right wrist as I was catching myself.”
“Oh, that’s really bizzare. Didn’t something like that happen to your father last year?”
“Yes, but that was a seagull at the beach, and it was his left wrist.”
“Ok, let me have a look at this. I don’t think that there is a break, but it looks like just a sprain. I’m going to give you this splint to wear for one week. You can ice the area as much as needed, and take up to 600 mg of ibuprofen as needed daily for a week. If it isn’t improving during that time, you can see me again.”
When the visit has concluded, the provider will stop the recording. Most AI scribe solutions also offer the option to process the note from a PC with a microphone in the exam room. Next, the note is transferred to the EHR system that the provider uses to complete the remaining parts of the visit. When the provider goes to the notes section of the visit, a mostly complete progress note is displayed in the EHR. One of the remarkable things about the technology is that it excludes the small talk and any information extraneous to the visit- in this case, the story of the patient’s father.
The provider will still need to review the note, and the technology will not log anything clinically significant without the knowledge and consent of the provider. The audio of the conversation is not stored on the provider’s phone or PC, and it is completely deleted from the vendor’s database after some time. The AI generated progress note will look something like this:
SUBJECTIVE
42 year old male patient Sam reports being struck by a duck while in the park this morning, causing injury to right wrist.
OBJECTIVE
Physical exam performed on right wrist. Swelling present. Range of motion restricted.
ASSESSMENT
No break detected.
PLAN
Fitted right wrist with brace to be worn continuously for one week. Patient to place ice pack on affected area as needed. Take up to 600 mg of ibuprofen PRN daily for one week. Return if no improvement within one week.
ORDERS
None placed.
DIAGNOSIS
ICD-10 W61.62XA: Struck by duck, initial encounter.
The next stage of AI scribe solutions will bypass some of the third party apps that are integrated with EHRs, and just have that technology embedded directly into the EHR. This work has already started with Epic and Cerner, so I suspect that the landscape of these solutions will be very different in a few years. I also think that there may be some legal battles that resemble the actions against Microsoft in the late 90s to early 2000s.
Over my many years in healthcare IT, I’ve developed a certain cynicism for technologies that have promised and failed to “increase efficiency so you can spend more time with patients”. I also have more than a little skepticism for AI in general. I don’t like the tech bro CEOs of the major AI companies, I don’t like the spread of data centers that we’re all seeing, and I don’t want to outsource large parts of my brain. However, I’ve always been a believer of not throwing out the baby with the bath water. I do believe though that AI scribe technology is actually living up to the promise of making the lives of providers easier. I have seen this firsthand when I rolled this out for my providers. They almost universally report that they are spending less time on the computer with patients and less time catching up on documentation after clinic hours.
Targeted Clinical Analysis
In late 2017, I started experiencing some mysterious physical symptoms that made my life miserable for months. I went to all kinds of doctors and alternative health providers, none of which provided any measurable help. Fortunately, almost all of my symptoms gradually resolved. If this had happened to me today, I could have benefitted from a new technology used by the leading medical records software company in America.
Epic Systems hosts the medical records of about 270 million patients in the US. Part of their technology is the Cosmos research database that compiles and analyzes billions of data points. One of the functions made possible by AI is called Look-Alikes. It gives the ability for a provider to request the system to match rare symptoms and diseases to the de-identified records of all other healthcare organizations that use Epic. It can then put the provider in touch with other potential providers who had patients with similar symptoms or conditions. Patients’ personal data isn’t shared- only the information about the potential matches that could help the providers to unravel medical mysteries.
AI Driven Diagnostics
Early adoption of AI tools to interpret medical imaging scans and cardiology tests have shown some promising results in identifying disease at earlier stages than what is typically detected by human observation. One example is the Enhanced Breast Cancer Detection system, which is diagnosing cancer at a 20% increased rate and at earlier stages. The system still requires active interpretation from Radiologists.
The EchoNet cardiology screening tool was developed by New York Presbyterian Hospital, and has accurately identified 77% of structural heart problems, compared with 64% for cardiologists using traditional technologies.
Which Health IT Jobs May Be Replaced?
We can’t have a discussion about AI without exploring which jobs in healthcare IT might be reduced or replaced by artificial intelligence. I can start by saying that I believe that this industry will not suffer nearly as many job cuts as some others, like pure software development companies. The skills required by analysts in healthcare settings involve technical, clinical, process, workflow, and people skills all at the same time in ways that I don’t believe AI will be able to significantly impact. However, I am starting to see some AI efficiencies being developed for some processes that I do in my job, such as the technical build for a new clinical department.
Some of the tools used by workers in data analytics are already being made more efficient by AI, so I predict a moderate effect on that area eventually. My wild guess for the long term would be to say that a healthcare IT department might be able to reduce analyst and other technical roles by maybe 10%.
I do think that the first line help desk in a healthcare IT organization could be affected at a much higher level. Many organizations have already outsourced this function to third party firms, and I believe automated ticket submission software will have a big impact on these positions in the medium to long term.
To help you stay ahead of the AI curve, check out this post on healthcare IT certifications.
