MEDICAL DICTATION: Crossing the Chasm from Paper Charts to Intelligent Records
The concept of a paperless chart has been around for years now and usually refers to the creation of an electronic patient record (EPR). An EPR at the most basic level involves the physician typing in all the patient information and going through a series of screens that force him or her down a certain path. (Some organizations may use the term CDR, EMR, or CPR, however for the purpose of consistency in this article, we will use EPR throughout.) Some of the more sophisticated EPRs use a click and point method which speeds up the process somewhat, but the physician is still encumbered by a programmer's concept of screen flow that may not suit his clinical approach. Then, at the top of the list are those EPRs which offer a voice recognition input system. These are one of the better alternatives as long as all physicians are willing to spend the time to train the system to recognize their voice. But no matter which system a physician chooses, the final product will still simply be a paperless chart. Have you attained any significant benefit? No, simply because what you end up with is a lot of information entered as text and no way to use that data without reading each document. In essence you have moved to a paperless chart, without gaining the benefits of a computerized system. The data will just sit there in the electronic record. Proponents of such systems are correct to point out that they eliminate the paper chart, and they are right. But that is simple replacement of the paper chart. No extra functionality has been added and that is not enough. You might as well just have typed the document, with no information storage, retrieval or reporting capability. The benefits of an EPR system are many and within the medical community are quite widely recognized. However, there is one major flaw in the concept of EPRs. Until now no one has provided practitioners (physicians, nurses, etc.) with a practical non-computer phobic methodology to input their information (data) into the system. It is wonderful that many hospitals are spending millions of dollars (on average 30-90 million dollars over a three to five-year period to implement an electronic clinical data repository) to tie together all their patients' information; such as demographics, billing, laboratory results, radiology reports, etc. But, what do all these things mean without the physicians input and analysis from the physical examination? They mean very little, except to document that the test has been done for billing purposes. It appears that little thought was given to capturing physicians' data as input to the electronic medical record. One exception was a government agency, the National Institutes of Standards and Technology (NIST) that awarded a number of technology research and development grants aimed at decreasing the 40% of the costs of health care associated with paper. In particular, a two million-dollar grant was awarded to Berdy Medical System, Inc. in 1996 for research and development of new technology to facilitate physicians input to electronic medical records by way of natural spoken language. Physicians and most people born before 1980 display a wide spectrum of computer phobia. They represent a large market for voice input software. This software has gone through an evolution - first only recognizing discreet speech (with pauses between each word), and the most current developments allowing continuous speech. With continuous speech, an attempt has evolved to allow doctors to input directly into electronic medical records simply by speaking. Again, another great concept, but something is missing. Voice technology today only provides a text document from the speaker's voice. Sounds reasonable at first, but let's look at this in greater detail. The purpose of an electronic chart is to be able to glance at information rapidly, and in many different ways. In part it is to get away from wasting a physician's time flipping pages and frantically looking for the last report or test result, but more importantly being able to view a group of results over time periods to evaluate a patients progress. Discreet Data Elements
To achieve this end result, and EPR focuses on a database repository (usually a relational structure) of discreet data elements that can be related to each other and presented to a physician to appropriately evaluate and treat a patient. The key words here are discreet data elements as opposed to plain text documents. For example: Today's average physician, examining patient J. Doe, transcribes as part of the physical examination a cardiac finding of a murmur, the transcriber types the note for the physical examination and notes murmur under cardiac examination., and the patients note is placed in the paper chart for J. Doe (a text document). Now if the physician wants to know if the patient had a murmur in the past, the physician needs to read through all prior text documents to see if one was ever recorded (a time consuming process), which may not even be feasible if recorded by another physician with illegible handwriting. Let's add voice software to this scenario. The end result is the same text document as returned from a transcriber, but there is a benefit of reducing transcription costs. Now if the physician wants an answer to the question, "Is this murmur new?," the physician still has to go through the same process of searching all the typed documents to see if it was ever recorded on a previous examination. Next, let's add an electronic medical record to this scenario with voice input. This means that the physician was able to use voice software to dictate and transcribe the examination, and this text document has now been stored in the electronic medical record. Again, the physician wants to know if the murmur is new? And again, he has to read each text document that has been stored to see if it was previously recorded, albeit changing pages on a screen is somewhat easier than fumbling through paper, the physician still has to read each document. The technology lacking here is the storage of information as discreet data elements rather than text documents. If this murmur was stored as a discreet data element under the category of cardiac findings and further stored with a data flag noting this is an abnormal finding the physician could easily get an answer to the question of this finding being a new murmur. Additionally, many questions could be answered, even if they were posed in different ways. Eg: Show me all abnormal cardiac findings for J. Doe. Show me all cardiac findings for J. Doe. List all prior examinations for J. Doe with a cardiac finding of a murmur. Or the question can be asked in relation to other information such as - Show me all examinations for J. Doe with a murmur and his hemoglobin as of each examination. The missing piece is referred to as "gisting." To understand the concept one must only look at everyday examples of incomplete or poor speech. If your child says "a cookie want I now," you would get the gist of what he wanted and the meaning of the sentence even though it may be incomplete and out of order. That is because you know the meaning of the words and the context of the situation.
Meaning plus context equals understanding = the gist of what is being said.
Voice to text simply is not enough to obtain patient outcomes, and the return on the investment from the costs of implementing an electronic medical record the data must be stored as discreet data elements. How can unstructured physicians jargon be converted to discreet data elements? The answer is understanding the gist of the speaker or applying the process of gisting. Gisting, a term coined by Berdy Medical Systems, Inc., describes a new technology of understanding the gist of the speaker. Medical gisting software is software which understands medical language that is either spoken by a health professional, or recognized by looking at text. From existing language and grammar models, speech is classified into categories, then based on statistical probability and the context of the industry jargon, the gist - the meaningful intent of the speaker - can be decided. Gisting is the bridge across the chasm from speech input to electronic medical records. Gisting, packaged as SmartGistTM, parses physicians spoken input, or even text documents, into discreet data elements. Then based on the existing language and grammar models that define understanding of physician's use of medical grammar, information is categorized and associated with the appropriate components of a patient's medical record and stored as discreet data elements. Actually, SmartGist produces a Health Level-7 transaction, which is given to the EPR or data repository of your choice. Then it is the EPR that actually stores the information in the data repository. SmartGist takes the unstructured physician's natural spoken language, converts it to discreet data elements, and places these elements into appropriate transactions. Let's take a look at the following example: You are a physician seeing a patient in the clinic. You take a history and perform an examination, then speak into a telephone like handset as if you were normally dictating. Order doesn't matter, and you don't use punctuation. "Very pleasant man in mild distress secondary to chest pain weight one hundred seventy pounds blood pressure is one hundred thirty over ninety four temperature is ninety seven point two skin is pale"
SmartGist First Normalizes the Text
- Punctuation is removed
- Patient identification confirmed
- Numeric data is changed to text
- Abbreviations are normalized
- Headers and trailers are removed
Normalized Text is then Gisted
- General Status Px:VERY PLEASANT MAN IN MODERATE DISTRESS SECONDARY TO CHEST PAIN;
- Body Weight: 144 lbs.;
- SYS BP: 112 ;
- DIAS BP: 74 ;
- TEMP: 97.2 ;
- Skin Px: PALE
Gisted text in then made into an HL-7 Transaction and passed to the application via API OBX||ST|3000.01^General Status Px^AS4||VERY PLEASANT MAN IN MODERATE DISTRESS SECONDARY TO CHEST PAIN ||||||F OBX||ST|1010.1 ^Body Weight^AS4||144|lb^lb^ANS+|||||F OBX||ST|1002.2^SYSBP^AS4||112|||||F OBX||ST|1002.3^DIASBP^AS4||74|||||F OBX||ST|1000.2^TEMP^AS4||97.2|||||F OBX||ST|3000.03^Skin Px^AS4||PALE
Gisting is a stand-alone technology that gives electronic meaning to medical spoken language. Any commercial speech technology can be used at the front-end to provide the verbal or text input and any commercial HL-7 clinical data repository can be used to store the data.
Dr. Jack M. Berdy, MD, founded, along with Richard Holtmeier, Berdy Medical Systems. With the goal of combining computer and medical expertise to facilitate physicians in this era of medical reform. Dr. Berdy also was Founder, Chairman, and CEO of On-Line Software International Inc. (NYSE:OSII) for over twenty years. OSII was a $100 million dollar company specializing in IBM mainframe communications and systems software and was acquired by Computer Associates Inc. in 1991.