Early adopters – the approximately 15 percent of
doctors who use an electronic health record (EHR) system successfully –
are hitting a major speed bump ahead of their peers: EHRs can slow physicians
Too much emphasis is being placed on EHR
deployment and not enough on utilization. In the rush to
computerize patient information, per Obama’s five-year goal that all
medical records go digital, it has been assumed that once EHRs are widely
deployed, patient records will automatically be more complete and shareable, administrative
costs will be cut and that universal quality of care will be enhanced. First,
we have to get doctors to meaningfully use the system…
In a study conducted by Fallon Clinic comparing EHR
technology used as is vs. with speech recognition technology (replacing traditional
transcription and keyboard-only control of the EHR), not only did speech
recognition prove to help doctors capture the patient encounter in more detail;
it helped to save $5,000 per doctor per year in transcription costs and generated
additional reimbursement per encounter.
Additionally, a recent survey of 1,241 doctors who use
speech recognition (Dragon
Medical) to capture dictation and populate the EHR found that:
- 81 percent said that it significantly reduced
- 69 percent said it made their EHR faster and
easier to use
83 percent said that it improved the quality of
their patient notes
According to Dr. Dean Carr, a primary care physician at Group Health Cooperative and Epic EHR user in Seattle, WA,
speech recognition “has really shortened the process of creating
notes. It shortens my day by 1–1.5 hours because I no longer have
to log in from home to complete chart.”
Studies have shown that in some cases, the adoption of EHRs
themselves does not lead to higher quality care. An April 2008 article in
the New England
Journal of Medicine by Groopman and Hartzband
cited the dangers of over reliance on EHR automation: “This capacity to
manipulate the electronic record makes it far too easy for trainees to avoid
taking their own histories and coming to their own conclusions about what might
be wrong. Senior physicians also cut and paste from their own notes,
filling each note with the identical medical history, family history, social
history and review of systems.”
By pairing speech recognition technology
with EHR deployment, doctors are more likely to successfully use the
system. By dictating their patient notes directly into the EHR vs. typing
their notes, doctors can spend less time on clinical documentation and more
time on patient care. The return is huge – patient notes are more
detailed and no longer rushed or sloppily-typed. Moreover, traditional
medical transcription costs can be slashed and insurer reimbursement is often
higher because of more detailed “medical decision-making” – such
as history of present illness, review of systems, assessment and plan.
A solid ROI example of the value of speech
recognition improving the patient documentation process is Beth Israel Deaconess Medical Center in Massachusetts. They
have achieved cost savings of more than $2 million and decreased turnaround
time (the time from when a patient is seen, to the documentation and
communication of their visit) by 90 percent because of their investment in background
speech recognition technology. By leveraging speech recognition to support
clinicians’ “free dictation” which can be stored in the EHR, healthcare organizations can speed EHR utilization,
while cutting healthcare costs and improving patient care.
Some physicians have even deployed speech recognition prior
to EHR implementation to generate substantial cash savings which helps pay for
the EHR itself. “Before we began implementing
the EHR, we leveraged speech recognition to nearly eliminate transcription
costs,” commented Douglas Golding, MD, Medical Director and Chief Medical
Information Officer at Lifetime Health
Medical Group in Buffalo,
NY and NextGen EHR user.
Golding’s practice covered the full EHR cost with
transcription savings generated by speech recognition. “As we
implemented the EHR, we found that [speech recognition] facilitated free text
entry where needed. Now providers create and edit patient reports in one
sitting through free-form dictation. Report turnaround times have gone
from as much as three weeks to a matter of minutes,” Golding detailed.
“Best of all, we started saving money almost immediately. The
payback on our investment took only six weeks. Most technology payback
periods take two to three years to break even.”
Recent physician surveys, anecdotes and testimonials confirm
that speech technology is a complementary tool to make EHR systems user-friendly.
So while the EHR transition
addition to making sure doctors’ efficiency isn’t compromised by
learning a new system, we need to ensure they don’t take shortcuts by
creating cookie-cutter medical records that are void of expression and patient
detail. Speech let’s the patients’ records maintain their
doctor’s voice (literally) and doesn’t force physicians to be a
slave to the keyboard.
Keith Belton is VP of Marketing at Nuance Communications.
We have an “employment guarantee” so our hospital thankfully cannot “eliminate” transcriptionists with “The Dragon”–instead they must either give them a retirement package they can’t refuse or pay for sometimes costly schooling and retraining and place them elsewhere within the hospital in other positions–so not so “cost effective” from that end.
And from reading the above blog, it appears that doctors have now become medical transcriptionists.
The U.S. military recently embarked upon use of Nuance Dragon Naturally Speaking, Medical, for front end speech recognition with AHTLTA, DoD’s Electronic Health Record. Initial data shows positive results in improving clinician satisfaction with AHLTA use; improving perceived quality of clinical encounter notes; and in some cases, improving productivity ( a few more patients seen), by using macros. Army is also engaged in research of natural language processing, using Language and Computing’s TeSSI NLP engine, to turn text into coded facts. Kaiser is doing similar with Epic to support billing. Robert E. Connors, FACHE, PMP, Executive Health Manager, Henry M. Jackson Foundation for the Advancement of Military Medicine.
There is no debate that voice recognition is a great tool. I agree with Dr. Flashner, however, that it is nothing more. As a physician using EMR and voice recognition for the past 6 years, it is safe to say that I have considerable experience. I am not EMR and voice recognition is a time saver for me but the quality of my documentation is indeed improved. The time spent in reviewing and correcting documents could easily be spent seeing two more patients/day and the cost issue is a total wash. EMRs using custom designed templates (I use MedTemps EMR)can be very useful. Most physician visits are somewhat routine (consider the post-operative visit, the preparation for surgery discussion, initiation of drug therapy and discussion of toxicity/monitoring, as well as routine exams in a stable chronic disease (joint counts in rheumatoid arthritis, diabetes, valvular heart disease) and the list could go on an on. Templates make dealing with these issues easier and an EMR that allows you to access prior records (on the fly) allows real time comparison. Templates are not that useful when seeing an unknown problem (except for documenting completeness of investigation)and here is where voice recognition really does well. To talk of technological advances without incorporating both aspects, is indeed a very narrow minded viewpoint. Jeff Kaine MD
Dr. Rowley writes:
> However, the disadvantage is that the notes were
> narrative, and thus not in any consistent structure
> from one practitioner to another, and no consistent
> lexicon was used. Hence, searching text entries was
> not really feasible.
Slee’s point is that the narrative notes ARE the record and that coding is an abstract of the record. When Natural Language Processing and vocabularies/lexicons are sufficiently developed, then the abstracts can be automatically created and the EMR begins to be useful for all the applications foreseen. I have noted that one can make multiple abstracts of the record for different purposes. Billing requires one abstraction, epidemiology requires another abstraction, efficacy of treatment research requires yet another, operations research (i.e. waiting times) still another, and so-forth. But they are all abstractions of the same record. Right now we only have the billing abstraction, but that only in theory.
> we wanted to have a robust set of templates that
> include a collection of “text snippets” that can
> be called into the chart note with a single click
> (and can be modified or appended on-the-fly).
I have seen systems like this: the doc assembles a note from standard paragraphs instead of choosing a complete canned note or a canned complete note or something. Sometimes the standard paragraphs can be edited individually, sometimes he’s made to leave out the standard paragraph and add his own when the standard won’t do.
The main problem I have with template-driven charts for Hx, PE, ROS, and so-forth is that it is too easy to check “no abnormality noted” without actually looking or having any particular need to look — this then is used to upcode a visit. Not that I have any love of current reimbursement systems, mind you.
This idea extends to automated transcription systems: the doc dictates “Standard Report C1” and have that be the end of it. Maybe “Standard Report C1” is “normal CXR” and it is several paragraphs long. One hopes the read was really as thorough as the report suggests it was.
I know I’m asking for a lot 😉 but we should not pretend we’re ready for more than we’re ready for. On the other hand, if we don’t start, we’ll never be ready. `tis a quandry.
The answer to Margalit’s earlier question is: Yes, this was an informercial. Not clear that THCB should simply be a publicity vehicle for vendors, regardless of how groovy their technology. Matthew take note . . .
Actually a well-designed template can improve care by prompting the physician to perform a complete physical examination, for instance. It used to be that taking a history and physical exam followed a well-worn sequence taught to all medical students. Today this sequence is becoming increasingly truncated. With a template, one has to either fill in the blank with the proper data, or admit that it was “ND” (not done).
Of course, the downside is that templates discourage reporting of unusual findings that do not appear on the template, if one is too lazy to add them.
The bottom line is, and has always been, you can’t legislate good medicine.
As a practicing physician who has used EHRs for a number of years, I have used speech-recognition add-ons (like Dragon Systems, for example) to create text in chart note fields. The advantage was savings on transcription costs (we were able to get rid of our transcription service). The disadvantage was the occurrence of correctly-spelled nonsense that needed proofreading before saving. This was certainly an advantage over hand-written (read that: hand-scribbled) notes, and had faster turnaround than with a transcriptionist. However, the disadvantage is that the notes were narrative, and thus not in any consistent structure from one practitioner to another, and no consistent lexicon was used. Hence, searching text entries was not really feasible.
I have basically abandoned the use of speech recognition in my own practice, in favor of better template-driven “point and click” tools in our EHR. It is simply a little faster, and more consistent in terms of note structure.
With this in mind, as I have been involved in developing a hosted, web-based EHR, we recognize that multiple inputs of chart notes are useful. Thus, allowing speech recognition software to input text is something we include, but also we wanted to have a robust set of templates that include a collection of “text snippets” that can be called into the chart note with a single click (and can be modified or appended on-the-fly). Of course, manually typing, or having a transcriptionist type chart note entries are also options. A well-designed EHR should have the ability to have multiple methods of input, depending of user preference, but I think a flexible template structure is quicker than speech recognition.
Robert Rowley, MD
Chief Medical Officer
Practice Fusion, Inc.
It is unfortunate that Mr. Belton has the serious misconception that speech recognition in and of itself will somehow magically generate more complete charts, higher reimbursement, and yield more time for patients. This notion is just as misguided as the notion that EMRs will somehow immediately yield much lower drug/medical error rates, allow for seamless information sharing among many clinicians/hospitals, and lower the overall cost of healthcare.
Speech recognition is a tool—period. A stethoscope is also a tool. Both are very cheap and very powerful. Both require training and practice in order to realize their benefits to the clinician, and ultimately, the benefits to the patient. The results of effective training and ongoing practice include speed and accuracy— critical elements in the day to day practice of medicine.
A stethoscope has one big advantage over the current status of speech recognition. That advantage is that, worldwide, heart sounds, abdominal auscultation, lung sounds, and the sound of blood moving through major blood vessels will sound pretty much the same. There is no “translation” necessary.
Speech recognition is at a disadvantage compared to the stethoscope in that two major types of “translation” are required:
1. Subspecialty lexicons. Does speech recognition “out of the box” accurately recognize all of the special words and phrases that may be used/dictated/spoken on a regular basis by neonatologists, ophthalmologists, pediatric oncologists, thoracic surgeons, pulmonologists, urologists, etc? The answer is “No”. Fortunately, there are a couple of third party vendors that have spent years developing inexpensive, subspecialty lexicons that address this issue.
2. Integration with EMRs. Does speech recognition “out of the box” enable one to dictate directly into any and all EMRs that exist on the market thus making this tool “seamless” in its integration with these products? The one mentioned by Mr. Belton cannot accomplish this unless one chooses to purchase the most expensive “version” of the product. I am only aware of one speech engine that accomplishes this feat at NO additional expense, and that is the Windows Vista speech engine. The solution here was intuitive but was a long time coming— integrate speech into the operating system, and if almost all EMRs are grounded in that operating system, then those EMRs are automatically “speech enabled”.
As an end-user of speech recognition since the days of IBM VoiceType Dictation 1.0, I have seen a very long series of incremental advancements in computerized speech recognition. The most dramatic advances have been mostly in the form of the hardware catching up to the demands of the software. The more recent introduction of Windows Vista speech as an integral part of the operating system is also a major advancement as it is finally no longer necessary to “bolt on” an external engine in order to take advantage of what this technology has to offer the interested clinician.
Gary M. Flashner, M.S. M.D.
VP, Medical Content
Tom Leith summed it up pretty well. There is technology and then there is appropriate technology. Unfortunately, the current crop of EHRs are focused on preserving a bunch of unnecessary billing/malpractice defense boilerplate crap rather than collecting and generating usable information. As a primary care physician, I never, ever look at old physical exam data for anything other than blood pressure and weight. I record the data so that I can bill insurance and so that I have proof that I checked blah, blah, blah, if I ever get sued. The EMR that I use now is a badly designed word processor whose only function is recording that data and generating semi-readable text. Work flows and communication within the office were an after thought at best. There is, for instance, no way to redirect my lab results to another doc if I’m out of the office for a vacation (flying to and from a family funeral on the east coast I recently learned that airplanes have WiFi!)
We have used voice recognition in our office for several years prior to the addition of the EMR. Some people in the office are good at it, some not too successful. However, it’s still a slow, cumbersome method for recording useless data.
As a former applications programmer, I type fluently, but scribbling with a pen on paper is still quicker (and in my case emminently readable).
Healthcare professionals have alternative solutions available in speech recognition. Windows Vista now has speech recognition built into the OS and third party medical topics can be added from Trigram Technology. The topics allow physicians and healthcare users to dictate into any program on the PC.
Also, since the medical topics seamlessly integrate into Windows Vista developers can take advantage of existing Microsoft code to link their program with speech recognition. In fact, many programs written in Microsoft Visual Studio require very little additional development since the speech engine is embedded into the Vista OS and can read menus and make applications ‘speech aware’ without additional coding. The Microsoft speech engine also shares informtation with the handwriting recognizer, so, tablet PC users benefit from that additional input.
Microsoft Vista speech engine and API licenses are free. Developers and doctors can contact Trigram Technology for more information on adding medical topics to their PC or application.
contact firstname.lastname@example.org or http://www.trigramtech.com for more information.
> I think physicians are educated enough
> to make their own decisions regarding
> their own documentation.
Left on their own, they’d do a lot less of it: a great deal of documentation is about billing third parties and defending lawsuits — two activities docs aren’t all that thrilled about doing.
EHRs aren’t meaningfully used because they don’t add any particular value from the doctor’s point of view.
Even if EHRs were no more onerous to use than dictation for transcription (do not underestimate the difficulty of doing this properly), this basic fact doesn’t change. But until EHRs are no more onerous to use than dictation for transcription, they won’t be meaningfully used.
Speech recognition is an enabling technology, but all by itself changes nothing. I take that back: speech recognition is an engineering refinement to a problem we have not earnestly begun to tackle. The enabling technology are Natural Language Processing methodologies and coding systems (plural). I am a big fan of Vergil Slee’s analysis of this whole topic.
I happen to be an IT person with 26 years in the business, and I think EHR is a byproduct, not the driving force.
Rethinking how we render medical care is at the core. The processes are key, and the people who go along with it essential. Technology is the artifact we have AFTER we figure out the rest.
Kindly refrain from generalizing. Not all IT people are propeller heads. 🙂
Is this an infomercial?
Speech recognition is important and so is digital inking and scanning and macros and templates and any other methodology for documenting a visit.
I just don’t think it’s the IT folks job to “ensure they don’t take shortcuts by creating cookie-cutter medical records that are void of expression and patient detail” – they, being the doctors.
I think physicians are educated enough to make their own decisions regarding their own documentation. Let’s concentrate on providing them with usable tools to do their job and not presume that IT knows the “right way” to perform that job.
Only IT people think that EHR is save all solution. In the current form, it is at the most a warehouse with some positive. For it to be usefull as IT folks are saying, there need to be new strategies and policies. We need new EHR system. For example, one of the article I wrote on my blog, I proposed it more patient controlled, and so on.
speech recogninition is easrier for physicians but then it is also just a repository…it does not add any intelligence of use. Unless of course, the recordings are being passed to LCCs for transcription. But then, that is UnAmerican to ship jobs overseas during the economic crisis.