By: Eric Fishman, MD


Physicians often complain that when using electronic health records (EHRs), they’re so busy typing that it’s difficult to make good eye contact with patients. New voice recognition software holds promise, but is it there yet?

Clicking through a database to provide exactly the type of data requested is very time consuming and arduous. It takes physicians’ attention away from patients in most instances, and may not add as much to the readability of a medical record as might be requested.

Gone are the days when the only way to enter patient data into an EHR was point and click; take your attention away from patients, and stare into an EHR screen rather than make eye contact with and relate to patients.

Now, using voice recognition in conjunction with many EHR programs solves a lot — but not all — of the problems that doctors had using EHRs.

However, there are still some variations in the way in which voice recognition works with EHRs, and there are still some downsides. If you haven’t already bought an EHR, should you only purchase one that is compatible with voice recognition software? If you already have an EHR without voice recognition, what should you do?

Speech Recognition Enhances the Relationship

There are a variety of advantages to speech recognition. First, physicians are most comfortable with this form of input because they have been using it for decades: first via an analog recorder, and more recently via a digital recorder. Next, it allows for the rich fabric of a patient’s history to be maintained rather than a tedious and difficult-to-read report produced by mouse clicks.

Possibly of most importance, however, is that it allows for maintenance of the doctor-patient relationship by minimizing the attention paid to the laptop computer inevitably and uncomfortably sitting between a physician and patient when using the point-and-click method.

There are several ways to input information into an EHR other than clicking through multiple drop-down menus: typing, handwriting recognition, scribes, and speech recognition. Speech recognition has, by far, had the widest acceptance, with some estimates that over 200,000 physicians in the United States are now using one form or another of speech recognition.

Speech-Recognition Programs Differ Greatly

However, there are tremendous differences between front-end speech recognition, in which physicians dictate and in real time review the interpreted speech, and back-end speech recognition, in which physicians dictate and transcriptionists, now frequently renamed as revisionists, will edit the speech-recognized text.

Therefore, as of July 2011, the overwhelming majority of EHRs are compatible with speech-recognition technology. In the most basic version, physicians using a desktop version of a product, such as Dragon Medical, will speak directly into the computer or tablet when various text fields open up.

In a somewhat more complicated mode, doctors use the speech-recognition program to additionally navigate between fields within a screen and to navigate between screens. More complicated still would be to use a combination of products, such as Dragon Medical and Frisbee dictation and transcription solution, to allow for back-end speech recognition, in addition to navigating between fields. There are even more complicated methodologies including the use of a combination of front-end and back-end speech recognition.

Keep an Eye Out for Degrees of Compatibility

It’s rare to find a Windows-based EHR (even the simple, Web-based EHRs) that is absolutely incompatible with any of the above methods. However, there are various levels of compatibility, and in some instances, such as when using Citrix, there may be some inefficiencies or lack of functionality when using speech and an EHR.

From a technical standpoint, one could consider that the integration between speech recognition and the underlying software could be categorized into 2 separate methodologies: embedded or alongside. Historically, by far the more common of these two was alongside, in which a physician would acquire the EHR and the speech-recognition software as 2 separate programs, frequently from 2 separate vendors. However, they would both be used, simultaneously, by a physician.

One of the major differences in a user’s experience is merely that the software will be installed onto a computer with 2 separate sets of discs. Except in rare instances, physicians using Windows-compatible EHRs are able to effectively, albeit in some instances imperfectly, utilize the two in harmony.

However, there is a trend to have the speech-recognition software embedded into the EHR itself so that physicians actually purchase only 1 product — the EHR — and it comes with speech-recognition functionality.

In some of these instances, there has been a substantial amount of integration accomplished between EHR vendors and speech-recognition vendors in order to allow for the spoken word to be placed, in some instances, as data within the data fields of an EHR.

In others, it is merely a matter of convenience that the 2 programs are provided by the same vendor and they merely sit on the same computer, as if they had been purchased separately. Your vendors will probably be very happy to extoll the virtues of a tight integration, if it has been accomplished, so merely asking them should provide an answer to this question.

Prognosticating, it would be my expectation that the combined, and in fact tightly integrated methodology, will increase materially over the next 3-5 years.

What Should I Purchase?

Because the process of documentation can be so burdensome, I would strongly recommend against using an EHR that does not function well with any form of speech recognition. Although there are medical speech-recognition programs for iOS, they are not as fully developed as the Windows-based programs.

Other signs that would indicate a relative lack of functionality would be an EHR that is able to be used only in a mobile wireless environment or only with Citrix, for instance. Some Web-based programs are also not as functional as desired.

In order to determine whether your chosen EHR is in this category, I’d recommend lurking on the online users’ forums and/or speaking with a physician who is using or who has attempted to use speech recognition with it. It would be unusual to find a program that is 100% incompatible with speech, so you may be looking at subtleties, such as impairments in navigation capabilities, rather than strictly impaired dictation capabilities.

It isn’t necessary to have the speech recognition embedded into an EHR. It just needs to be compatible with it.

Also, speech-recognition software will never make a spelling mistake. With many dozens of medical specialty vocabularies available, most physicians, whether in general practice or a subspecialty, can get started with an appropriate language module right away, improving their accuracy greatly from past years.

But All Is Not Paradise…

However, as with everything, there are disadvantages. Clearly the most relevant disadvantage is the lack of accuracy. Most production versions of speech-recognition software are capable of providing 99% accuracy, or even better.

However, with 250 words on a page, that could easily translate to 2 errors on every page. That is a meaningful problem from a medical care perspective. Physicians are strongly encouraged to review their dictations themselves and to make corrections as required. All of the currently available programs will provide for improved accuracy, over time, if each recognition mistake is diligently corrected.

From a financial perspective, in many instances, using front-end speech recognition for an entire patient record will not permit an EHR to collect the data that are needed. Therefore, a very common experience is for physicians to select a chief complaint from a drop-down list and to then dictate a few sentences of a patient’s history in their own words.

Clicking through the PMH, ROS, FH, Social History, PE, and many other components can then be done, maintaining the sought-after data. Dictating a few additional sentences toward the end of the report to discuss the actual decision-making process can then complete the note. Some early products will improve this ability, such as NoteSwift, allowing the spoken word to provide data within a database. However, this is not yet universally available.

Another disadvantage is the training period. In past years and decades, the training required to have an accurate speech-recognition profile was very arduous. Currently, in most instances, there is a trivial amount of training needed to be done to the computer voice profile. However, a moderate amount of training remains that physician dictators are required to undertake in order to dictate most accurately.

In conclusion, when choosing an EHR, you should ensure that you will be able to document as rapidly as possible. My current recommendation, in most instances, is that you use a speech-recognition program to dictate the patient-specific history and your thought processes in determining the course of treatment for your patients, and that you click through most of the remainder of a record, allowing you to collect the data that is so sought after.

To Read the Article on Medscape please click this link.