This week I want to breakdown why notes are so hard. And more specifically, why building a note-taking app for doctors is not as simple as it may seem.
Here’s a glossary to start with:
Example SOAP Note
Pt is 87 yo woman, highschool teacher with past medical history that includes - status post cardiac catheterization in April 2019.She presents today with palpitations and chest pressure.HPI: Sleeping trouble on present dosage of Clonidine. Severe Rash on face and leg, slightly itchy Meds: Vyvanse 50 mgs po at breakfast daily, Clonidine 0.2 mgs -- 1 and 1 / 2 tabs po qhs HEENT : Boggy inferior turbinates, No oropharyngeal lesion Lungs : clear Heart : Regular rhythm Skin : Mild erythematous eruption to hairline Follow-up as scheduled
When you see “SOAP Note” below. Just picture the above.
Now that we’ve got that out of the way…
EHRs have historically been about using structured clinical treatment data and unstructured clinical treatment data to get doctors and hospitals paid. Believe it or not, the primary purpose of most EHRs is not to facilitate care. For years now, EHRs have been trying to get narrative-based notes to be more like structured data inputs. The way they have accomplished this is by providing checkboxes, radio buttons, dropdowns, and other inputs to clinicians to record things they would traditionally write out in a SOAP note. As you can imagine, this is a very time-intensive process and requires a lot of focus on the part of the user to fill out completely and accurately. The answer to this has been to develop Clinical Decision Support systems and features to “assist” a clinician in filling out these forms. Unfortunately, the primary way this takes shape is through additional modals, checkboxes, and inputs that get in the way of the real work.
Why all the hubbub about structured data, you ask? Because of two reasons: meaningful use and ICD-10-CM codes.
Meaningful use was established in 2009 by the HITECH Act that basically laid out how and why U.S. healthcare facilities should modernize and upgrade their systems. A big part of that was giving Medicare and Medicaid (the biggest healthcare payors in the U.S.) incentives to healthcare facilities using a Meaningful Use certified EHR.
Note: In 2019, the Medicare Access and CHIP Reauthorization Act (MACRA) rolled meaningful use up into the Merit-Based Incentive Payment System (MIPS). Most organizations still refer to it as meaningful use, I suspect because MIPS sounds like a weird disease you got on a college binger that one time.
The other reason for such an intense focus on structured data is ICD-10-CM codes. These are nationally and internationally recognized codes for different problems. These codes are quite complex as they don’t just list one code for everyone with heart palpitations (as in the patient above). They vary depending on body location, age, severity, and how much treatment needs to be done. For the above SOAP note alone, there are 50 possible ICD-10-CM codes, depending on how specific and actionable any single problem is. Yeah, I know. It’s a pain.
The problem is that EHRs have been optimized for meaningful use and ICD-10-CM because that’s how everybody gets paid. Yet, all of this has been done at the expense of clinicians’ ability to use their computers well and in a way that promotes their relationship and care of their patients.
So what are we going to do about it?
Our goal is to blend the narrative-based SOAP note that clinicians across the board prefer(and learn in med school) with the computer’s ability to, well, compute. In other words, we’re going to take unstructured notes and make them into usable data.
The first step for this is to build a flexible system that helps clinicians build SOAP notes quickly. And, crucially, one that can be customized depending on which terms or structure a clinician prefers. Our system, which I’ve previewed here before allows a clinician to add “blocks” of content that they can fill out. They can do this as easily as we can type an email or post a Tweet.
Further, they can save commonly used structures and text to quickly copy and paste to their EHR of choice to save time on documenting their encounters.
In the second iteration of this product, we intend to template-ify the setup. In other words, we intend to let clinicians define variables that they can fill in on the fly with, you guessed it, structured data. And they can use pre-defined templates for specific problems. Take a look at that SOAP Note again:
Pt is 87 yo woman, highschool teacher with past medical history that includes - status post cardiac catheterization in April 2019.She presents today with palpitations and chest pressure.HPI: Sleeping trouble on present dosage of Clonidine. Severe Rash on face and leg, slightly itchy Meds: Vyvanse 50 mgs po at breakfast daily, Clonidine 0.2 mgs -- 1 and 1 / 2 tabs po qhs HEENT : Boggy inferior turbinates, No oropharyngeal lesion Lungs : clearHeart : Regular rhythm Skin : Mild erythematous eruption to hairlineFollow-up as scheduled
This patient has a “problem” of palpitations and chest pressure. That is commonly referred to as the chief complaint. Many other people have chest pressure and palpitations, so the commonalities between them can be pre-saved and filled in.
Many EHRs do have template functions already. However, because the base UX is quite bad and hard to use the templates are often hard to use as well. That’s why we’re starting with Step 1. To get the base UX correct.
What I’ve bolded above is all a doctor really needs to type out. Everything else is already stored in the patient’s chart. If the doctor has a simple, easy to use way of customizing the bolded information either through typing, speaking, or even using an Apple pencil, the time it takes to document goes way down. Add on to that the potential for Natural Language Processing to suggest patient history, relevant medications, etc and the doctor practically doesn’t have to do any typing at all.
And, you guessed it, our third step is to integrate Natural Language Processing and Machine Learning to do just that. Natural Language Processing can understand what a clinician is typing and why it’s relevant. It can then find other information in a patient’s chart to bring alongside the current note or to even add in information so the clinician doesn’t have to find it themself. Additinally, it can suggest ICD-10-CM codes to make billing and coding faster. And, ultimately, can do much of the heavy lifting of documentation for a clinician.
While our ultimate note-taking tool won’t be MIPS certified at first, we think it’ll give us the right foundation to build HeatlhcareOS around a very powerful way for clinicians to work with patients. And it’s all coming very soon. We hope.