Human medical scribes vs automated scribes

Automated medical scribes are surpassing humans in many respects and they’re getting better every day.

What can traditional medical scribes do for you?

What exactly does a medical scribe do? Medical scribes come in two flavors: in-person and virtual. In both situations, the approach is the same: A practice (or health system) pays a human being to observe each patient encounter. They type up clinical notes by making factual observations about what is occurring in the patient encounter. Then the scribe manually types notes and other data into the EMR. Often, the practitioner dictates content of the note word-for-word and the scribe acts as more of a transcriptionist who types what they are told to type.

The medical scribe approach has several advantages, which are well tested and evidenced. Multiple studies have demonstrated that medical scribes typically save doctors time (usually 2–3 hours per day), increase reimbursement (by around 20%) by producing more detailed notes and improve satisfaction for both patients and doctors. For this reason, medical scribes are commonly employed in elite healthcare systems, emergency rooms and increasingly in family medicine clinics. The scribe industry has exploded in recent years. By some estimates, there were 20,000 scribes employed in the US in 2015 and over 100,000 as I write this today (in 2020).

The reason the scribe industry has exploded is that human medical scribes are serving an important purpose and making life easier for doctors. What the health system needs is more of it! So starting around 2019, a few software companies (like my own company, Thread Medical) set out on a mission to automate and enhance the work being done by human medical scribes.

A fresh approach: Automated scribes

These startups developed a new breed of technology, which has been given monikers like “automated medical scribe”, “self-writing clinical notes”, “fully automated SOAP notes”, “automated clinical documentation” or “a clinical documentation assistant.” (It’s so new that there isn’t an acronym or jargon for it yet!)

The basic idea of these technologies is to use “artificial intelligence” (software) to write clinical notes automatically so that practitioners and human scribes don’t have to do it. While this concept might seem far-fetched — like something out of Star Trek—I can assure you that it is very real! Over the past few months these technologies have advanced to the point where some of them can now produce clinical notes with near-perfect accuracy in real-time simply by listening to a practitioner converse with their patient.

Note that these new technologies are not dictation solutions. There is no dictation necessary. And these aren’t voice-control solutions. (The doctor does not have to say: “Alexa, prescribe X mg of drug Y to be taken X times daily orally.”) This is a new type of technology that instead writes notes intelligently simply by listening to the natural flow of patient-physician conversation. This can be done through a microphone device, an app on a phone, a tablet, or laptop.

These technologies are designed to produce the same benefits as a medical scribe. But they also do more. In many respects, they surpass the practice of hiring human scribes and provide new benefits like more detailed clinical data and powerful clinical tools that don’t exist in any EMR.

These aren’t your Grandpa’s SOAP notes. Automated clinical note-taking improves the quality of clinical data which improves the quality of clinical care.

Unfortunately, traditional note-writing practices used by medical scribes are incapable of producing detailed patient encounter data that can be used to power multi-encounter insights. Each text note needs to be read by a human in order for it to provide useful insights, so you can’t analyze the content of multiple notes at once in any useful or effective way.

The traditional SOAP note has changed very little over the last 50 years. This tried-and-true way of communicating medical observations works (mostly) fine if you want you want to draw conclusions or communicate about a single patient encounter. But if you want to draw conclusions about multiple patients (or multiple encounters for a single patient) then traditional human-typed SOAP notes are not the best way to capture or represent the information from those patient encounters.

For example, suppose a doctor has a panel of 2000 patients and wants to ask a simple question: How many of my patients complained of a migraine headache that had a severity of at least 6 out of 10? And how many of them got better after I treated them? Good luck answering that question in a traditional EMR! Most likely, you would have to read thousands of notes to find the answer! In reality, the doctor would probably never learn the answer to this question. They would be left wondering whether their course of treatment really works for their patients. This isn’t ideal for the practitioner or the patient.

AI note-writing solutions like Thread solve this problem and put all of the data at clinician’s fingertips. The technology accurately writes a traditional SOAP note with high quality. But it also does much more. During a patient encounter, it captures and saves every medical observation that occurs (HPI elements, patient history elements, exam results, etc.). This allows it to go much deeper in terms of providing new clinical tools and insights that practitioners can use to improve clinical care and patient outcomes.

Here’s another example of how this structured clinical data is useful: Many clinicians are of the opinion that SOAP notes have lost some of their value as a clinical communication tool because they are often tainted by billing information and boilerplate that clinicians don’t need to know. But it doesn’t have to be that way. By pulling relevant medical facts and events out of patient encounters, automated scribes can present notes in different ways to suit the needs of each stakeholder: a “clinical view” for practitioners, a “billing view” for coders/billers and even a “patient view” that communicates the information in a patient-friendly way. All of these presentations communicate information about the same patient encounter but are presented for the needs of each person, which is the way things ought to be!

That’s really just the tip of the iceberg in terms of what detailed patient encounter data can be used for. At Thread, we are already using it to power several useful clinical features. And we have ambitious plans on how to use this information to accelerate accurate diagnoses, enhance early detection of cancer, improve preventative medicine outcomes, and flag potential medical errors/omissions in real-time.

Clinical notes made better, faster and cheaper.

You might think that a human being writing medical notes would produce more accurate and complete results than an AI-based solution. That theory makes sense intuitively but it turns out to be untrue in practice. In my experience, the average medical scribe produces a perfect note perhaps 98% of the time and the average note quality is good but not great. We can do better.

The top AI-based solutions improve on the best accuracy that the average human scribe can produce and often surpass 99% accuracy. That’s because AI is consistent, systematic and learns from its mistakes across many thousands of patient encounters (more than most medical scribes would see during their career). Further, once the AI is taught to stop making the mistake, it will never do it again.

Software solutions also have the benefit that they aren’t constrained by human time. A scribe or doctor typically has 8-10 hours a day to write and improve their notes. This constrains how thoroughly they can write the notes. And typing on a keyboard is slow. Computers don’t have that problem. A set of note-writing tasks that takes a human 20–30 minutes can often be done in seconds by a computing cluster. A 15-minute appointment is plenty of time for an app like Thread to gather all of the information it needs to complete a note.

Another problem with scribes is that they often cannot independently write the Assessment & Plan portions of a clinical note. But a good AI-based solution can help write the Assessment because it is programmed with a medical understanding of why a particular diagnosis is supported by the medical evidence gathered in the patient encounter. For example, Thread knows that a finding of “photophobia” for a patient with a CC of “headache” is indicative of a migraine. So if “migraine headache” is one of the differential diagnoses identified by the physician, then the software will infer that the finding of photophobia is part of the evidence for that diagnosis.

A final benefit of this type of solution is that it puts the doctor in the driver’s seat. A human scribe cannot enter medical orders directly into the EMR because they’re not a doctor or mid-level provider. AI-guided systems are different. They can draft medical orders without submitting them to the EMR. Then the provider simply reviews and submits the information. This speeds up not just note-writing but other types of EMR entries, too.

The price-tag problem: The solution to administration problems is not to hire more administrators.

The largest economic drawback of hiring human medical scribes is the cost. By some estimates, the fully burdened cost of an on-site medical scribe in a health system runs at $55,000-70,000 per year. (That includes salary and benefits but not hidden costs like recruitment, training and turnover.) At this price, scribes are a viable option for many hospital systems (where the average family medicine physician drives $1.5 million of reimbursement per year). But for an independent solo or group practice (where a primary care practitioner may only be receiving $300,000/year of reimbursement), adding another $70,000/year of overhead is a tough pill to swallow. And who ultimately pays for it? Patients.

Remote scribes are slightly cheaper because they’re often located in low-labor-cost locales like India. A practice can hire one for around $24,000 per year. But that’s still a hefty price. And you get what you pay for — an offshore scribe’s mastery of English or American medical knowledge may be less than stellar — not to mention the difference in timezones, cultures and the complications of transmitting data half-way across the world to a country where privacy and data protection laws are lax. And if a monsoon or brown-out happens at the remote data-entry location, the physician may be without their scribe.

On price, AI-based solutions are superior because they use cheap technology to replace expensive human labor. Some AI scribing solutions cost roughly 1/6th as much as an onsite medical scribe and the price is continuing to drop as the technology improves.

The COVID-19 Factor. Is having a scribe in the room a good idea?

The COVID-19 pandemic highlights another reason why having a medical scribe in the exam room is sub-optimal. Every person in the practice and the exam room provides another vector through which an infectious virus (whether it’s coronavirus or influenza) can spread. It’s also another person who needs an N95 mask in a time when medical supplies are short. For that reason, I surmise that in-person scribes reduce safety for both patients and providers.

Patient encounter workflow and patient communication.

Medical scribes are sometimes just transcriptionists who simply type what the doctor tells them to type. When a doctor dictates notes during a patient encounter, it undermines communication with the patient. There is a “weirdness” factor to dictating notes that tends to alienate patients and come off as very unnatural to those who aren’t accustomed to it.

Passive listening technologies used in auto-scribes like Thread are a better alternative in this respect because they don’t get in the way of patient care. They don’t require redesigning clinical workflows to accommodate a scribe.

The privacy problem

If you care about patient privacy (and I assume you do!), then AI based solutions have advantages over human scribes. A medical scribe’s presence, whether on-site or virtual, reduces the patient’s privacy and might make them less likely to disclose embarrassing information to their physician.

Well-designed software solutions de-identify all patient data so that notes can be written while minimizing the risk that any PHI will be viewed by a human being other than a doctor and patient.

The training problem.

Anytime a practice hires a new medical scribe or assigns them to a new practitioner, it may take weeks or months for the new person to be brought up-to-speed.

Automated medical scribes like Thread don’t have that problem. They’re ready to use on Day One and they’ve already learned from the experience of completing thousands of patient encounters. The best ones also adapt to the preferences of each practitioner and learn how to generate notes that are tailored to each user’s style and habits.

The turnover problem.

Training challenges are made worse by the fact that scribes typically have high turnover. A doctor may spend 6 months training a new scribe who then quits and goes to nursing school. Then it’s back to square one: Training a new scribe.

This is a major issue in the medical scribe industry where the average tenure of a scribe is often less than 18 months. Most scribes view it as a temporary job on the pathway to nursing, PA or medical school. AI-based solutions don’t have turnover. A practice chooses one once and can keep using it for as long as they want to. And once it is trained, it does not need to be re-trained and the knowledge that it learns is stored away indefinitely.

The recruitment problem.

Experienced, top-notch medical scribes are a scarce commodity. Medical scribes typically earn minimum wage and given high turnover in the industry, it is difficult to find highly qualified candidates. The training a scribe receives is much less than a clinician (a 2–4 training course is enough to earn a medical scribe certification) and many scribes start their first job with only a smattering of medical knowledge. Recruiting is also expensive because it takes time away from staff members. And posting jobs on websites often incurs a fee to the practice.

Software-based note-taking solutions side-step these problems since there is no recruitment involved.

Human processes are complicated.

Fact: Anytime you introduce another human being into a process, it adds new complications. Scribes involve communication, supervision, sick days, scheduling, human psychology — the list goes on.

An app like Thread lives on a phone, tablet, or laptop and is free of those human complications. They’re consistent and always available when you need them.

Selling other stakeholders on the idea.

If you work in a group practice or health system, then there are many stakeholders who are affected when a medical scribe is hired. Who is responsible? Who will do the work of managing scribes? Who will onboard them? Who else needs to be involved in the decision? Who pays? Who ensures that the scribe complies with HIPAA?

Generally speaking, software solutions are lower risk and often offer free, low-risk trials. It’s easier to sell others on a new idea once you’ve tried it out yourself and can testify first-hand to its usefulness. Auto scribes are also cheaper so budgetary issues are less of a concern (since ROI is higher).

The horse and the automobile.

For these reasons, I’ll venture to make a prediction: medical scribes, dictation products, and manual data entry are the horses of clinical documentation. AI-based note-writers are the automobiles. There was a time when horses were common and ubiquitous. It was hard to imagine that they would ever be replaced by machines. But how many horses do you see walking around on your street today?

What changed? The Model T. It was an affordable technology that made more sense economically. It delivered the same benefits as the horse but also surpassed them. And cars offered many benefits that horses could never match.

At Thread, we’re hard at work building what we hope will become the Model T of clinical documentation solutions: artificial intelligence that’s both more economically attractive and more clinically effective than the current state-of-the-art. While the technology is still young, we believe it will ultimately improve practitioners’ lives for the better and help them treat patients better than ever before. And unlike many things in healthcare, it might actually make care more affordable for patients by reducing the cost structure of many providers!

If you want to express an opinion, learn more about note-taking AI, or just get to know us, drop us a line: hello@threadmedical.com. We love to meet PC practitioners!