Using Dragon NaturallySpeaking in an Orthopedic Surgery Office - February 1998
Dragon NaturallySpeaking Deluxe Edition can be used quite efficiently in an Orthopedic Surgery office. If one wishes merely to dictate text into the medical report you may start by dictating immediately upon installing the program.
However very substantially improved recognition and functionality can be obtained by taking advantage of the many features of Dragon NaturallySpeaking Deluxe Edition.This paper will discuss the various stages of implementation of voice recognition in an Orthopedic Surgery office during the first 2 months of 1998. The office is that of Eric S. Fishman, M.D., President of 21st Century Eloquence.
Stage 1 -Installation
Dragon NaturallySpeaking Deluxe Edition was installed into a Pentium 300 with 64 MB of RAM. This was utilized to dictate follow-up patient reports. Because of the significant formatting features which are used in my office reports this stage was found to be frustrating and inefficient. For instance we use many tables comparing patient's range of motion with normal values, as found in the AMA Guidelines. These tables were essentially impossible to produce without significant manual keyboard input. In addition it was found that inputting the demographic information found at the top of the chart was very tedious. The columns did not line up properly, and there were a number of misrecognitions of insurance company names.
After approximately one week of utilizing Dragon NaturallySpeaking directly out of the box, without adding any features, most people concluded that the system was unacceptably slow, laborious, inefficient, and not likely to replace our current system of dictating, transcribing, proofreading/revising and signing.
There were a large number of inaccuracies with our accuracy rates varying between 90 percent 95 percent. In addition it was quite difficult to get the names of insurance companies, referring physicians, or certain specific medical conditions to appear accurately in the text. This is a stage that many successful users of voice recognition pass through. The choices: give the program to your children to play with or try to customize it to your specific situation.
Stage two - Topic Building
After realizing that the inaccuracy rate was unacceptably high it was quite apparent that this was not a satisfactory solution for an eight physician Orthopedic Surgery and Rehabilitation office.Upon further exploration of the program it became evident that there were a large number of methods of improving the accuracy and functionality of Dragon NaturallySpeaking Deluxe.First there was the Topic Builder. This is an excellent utility that will import massive quantities of previously dictated text into your working environment. As an Orthopedic Surgeon, I had thousands upon thousands of reports that had been stored on hard disk drives during the previous decade. These were copied to floppy disks in a .txt format and then copied onto the new Pentium 300 NaturallySpeaking machine. They were then imported into NaturallySpeaking via the tools / topic builder menu.This process took me approximately 2 hours. As it turns out, I had way too many reports (I had filled 32 floppy disks with reports) The topic builder stopped when it had received 500,000 words. This process can and in fact should be done by your speech recognition consultant, or by your secretary with support from your speech recognition consultant.After processing these reports NaturallySpeaking gave me a list of just over 1000 words for which it felt it needed instructions. Some of these were clearly brand new words, such as the last names of some referral sources.However, a large number of these words were clearly very commonly used words which were being used in a slightly different context. As a clear example of the latter, it presented to me for training the words Good and Samaritan and Hospital. Clearly each of these was in the original 55,000 word vocabulary. However, they had not been included in the context of a hospital whose name was Good Samaritan Hospital. After actually dictating approximately 150 of the just over 1000 I noticed an immediate and very substantial improvement in both the speed and accuracy of my dictations. This is a process that took me less than 1 hour, but which obviously cannot be performed by anyone else. It is this process which allows the program to better understand your personal dictation style and speech patterns.After utilizing the topic builder, when I say the hospital name at which I practice, I no longer need to say ‘cap good cap samaritan cap hospital’. I merely say Good Samaritan Hospital, and because of its prior importation of my previously dictated work, it ‘knew’ that these words needed to be capitalized. Also, as should be obvious, the context building aspect of the topic builder now ‘knows’ that after saying good followed by Samaritan, that it is ‘expecting’ to hear the word hospital, as opposed, for instance, to something like ‘the good Samaritan stopped by the side of the road. This is why the accuracy rate soared immediately after performing the topic building function.Parenthetically, it should be mentioned, that when using the Orthopedic topic for general dictation, medical words would spontaneously pop up in my personal dictation. This is not surprising when you realize that the Orthopedic topic gets NaturallySpeaking in a ‘mindset’ of hearing Orthopedic words and phrases.If I had stopped at stage 2 the program would have been perfectly functional. At that stage when I said GEICO as the insurer, it would usually get this right. However, it was a bit difficult to get the ‘header’ of the report dictated in a timely fashion. There were too many ‘tab’ keys, which needed to be hit. There were too many formatting features to the report, which needed to be repeated over and over again.I was able to get a report on paper. However, it took me 2-3 times as long as it had previously taken me to dictate a report, and the formatting features were clearly not as pristine and professional as I would have liked. Nonetheless, it was clearly less expensive than my previous system which includes the use of 4 full time transcriptionists.
Stage 3 – Macros
Having sold these systems to hundreds of satisfied users, and realizing that the system had significantly more capabilities than I was utilizing, I decided to wander around the menu system and manual of the Deluxe Edition, and to call upon the tech support department of the voice recognition company. The New Command Wizard was the first place I stopped. This turned out to be a very fruitful venture. With the command wizard I was able to make macros. Macros are a series of keystrokes or other commands which can be invoked with merely one or two spoken words.As a very simple example of a macro, called a ‘text macro’, you could make a macro called ‘my name’ and this would put your name on the page. A slightly more complex macro could be called ‘my name and address. This would put your name and address on the page, but all on one line.These are the types of macros which are available in the personal and preferred versions of NaturallySpeaking. Obviously they have very substantial limitations. They do not include the ability to utilize the ‘enter’ key. Thus producing a letterhead is very tedious with just text macros. The Deluxe Edition, however, offers substantially more sophisticated macro capabilities.Slightly more complex would be a macro called ‘my letterhead’. This macro could put your name centered on one line, then would ‘hit’ the enter key and center your address on the next. Again the enter key would be ‘hit’ and your city state and zip would be added in the center of the third line. Clearly this can simplify your report production. And very importantly, once you know how to make a macro like this it would take less than two minutes to produce another similar macro. You can then put your letterhead in a different font size and style with ease to set it apart from the remainder of the report. Thus, after about 3 hours of playing around with the system, and about 1 hour of working with it I had produced a few simple macros which included the letterhead, and a few canned text passages. One example of the latter is ‘light duty status’ which produces the following: ‘this patient should be considered capable of light duty work. Lifting and carrying of up to 20 pounds is to be allowed. Continuous bending and stooping is to be avoided. Within these restrictions…’I then discussed the additional possibilities with 21st Century Eloquence. The technical gurus of the company said that I had barely scratched the surface of Dragon NaturallySpeaking’s capabilities; and they were right!One week later we had full templates for follow up progress notes. These templates have all of the following features:Full demographic headings in two columns:Patient’s name, DOB, insurance information, date of injury, date of examination, etc. are all separate fields waiting to be voiced in.
Footers:A footer and page number has been placed on each page for the ‘progress note general’ macro. It is part of the ‘progress note general’ file which is called up merely by saying ‘progress note general’.Section headings:
The report opens up with areas for dictating the additional history, response to treatment, physical examination, radiographic findings, assessment and plan. Additional heading can be added easily.
Section macros:
Within each section we have available a series of macros which will make filling in the relevant information for that section easy. For example, for the physical examination we can say ‘insert lumbar range’. This phrase then causes a file to be opened which has text explaining the examination. It also has the normal values within a chart. The cursor then stops at the first value. I have the option of voicing over the normal default with the patient’s range of motion, or accepting the default by saying ‘take one’. In fact I can accept multiple defaults by saying ‘take n’ where n is the number of normal values attributable to this patient. After finishing the ‘insert lumbar range’ macro I can go on to ‘insert leg strength’ or any one of a number of other physical exam macros.
Navigation macros:
When finished with any one section I will say ‘go next’. This will take the cursor to the next portion of the progress report such as the Assessment, etc. This macro was produced by placing small invisible ‘markers’ within the page. The ‘go next’ macro looks for these invisible markers (they are alt 166 symbols in white), and highlights them as well as everything between them. Thus, as soon as I speak, they are replaced, along with the highlighted text between them, with whatever I say.
CC macro:
This macro takes me to the bottom of the document, puts in the symbol cc: and then inserts a tab. I can then say ‘copy ‘first name’ ‘last name’’ of the referral source and their name and address will be added at the bottom of the report.
Save report:
This one was a bit tricky. Since the phrase ‘progress note general’ opened up a file it was important that this file not be over-written with any individual patient’s information. For this reason it was made a ‘read – only’ file. Many of the physician’s and physician’s assistants in my office are quite familiar with computers. Some merely know that you need to plug them in when you want them to work. In order to facilitate the usage of voice recognition for this latter group we simplified some of the command and control features with which they were entirely unfamiliar. One of these was saving reports. The computer illiterate continued to have difficulties determining path names for saving reports, etc. Thus a macro was built for saving files. Upon saying ‘save report’ a script is run which takes the patient’s name from the first line of the report and the date of the examination from the 7th line of the report and makes a file name by combining those two pieces of information. This file is then saved in the ‘mydocuments / name of voice profile’ subdirectory on the c: drive. Thus, for whomever is dictating, the file will automatically be saved in their subdirectory with the correct file name. The read only progress note general file is closed, and is thus ready to be reopened for the next patient’s report.
Additional features:
Since we have color printers in the office, and since it is helpful when reviewing reports to see the abnormal values as clearly as possible, we have all of the fill-in the blank normal values in red on the original template. When the person dictating says ‘take one’ or ‘take n’ these normal values are accepted, and are changed to black. If, however, a new numerical value replaces the normal red fill-in, then they remain red. Thus a quick glance at the page will show the abnormal numerical values in red and the normal values in black.For some of the newer people dictating some helpful hints can be beneficial to them. Thus for the newer physician’s assistants we have information in the fill-in the blanks which will be entirely overwritten by their spoken word. However, it does let them know what type of information is expected in that area. As an example, for a macro such as ‘insert lumbar palpation’ we would offer the following: {Tenderness / Spasm} is {not} noted in the lumbar region. {If it is noted, specify where}. Thus, when dictating, you are prompted for findings such as tenderness and spasm, and if present you are prompted to tell where.
Stage 4 – Networking
Dragon NaturallySpeaking needs to reside on each separate CPU. However, there are a number of shared files within the office. We now have 5 separate Pentium 300 units (each of which cost significantly less than $3000) hooked up via a Windows 95 network. It was remarkably simple and inexepensive to install. Now each of the machines has access to the same universe of .txt files which were utilized to produce the original Orthopedic Topic, and in fact the Orthopedic Topic can be copied from one machine to the next to be placed in each user’s voice profile. In addition, we merely copy the normal.dot file from within MS Word from one station to the next in order to take advantage of any new macros which are produced by anyone. We have copied each users’ voice profile onto 2 of the 5 machines so that each person dictating (we have 12 in all) will have access to a machine when they wish to dictate.
Stage 6 - Potential Future Enhancements
Medicare Compliance and billing networking are anticipated.
Conclusions
Voice recognition software is a wonderful toy to play with if you are interested in being entertained. However, it can be more than that.If you merely wish to dabble in its usage, you are not likely to find it very functional to run a professional office. However, with that mindset it can be a useful program to produce text documents.However, if you care to take the time, trouble and expense to use it properly, it can be substantially more valuable than you would expect.As a rough estimation of the requirements for a 2 station office, appropriate for 3 – 4 physicians, I offer the following:
Costs:$4000 for 2 brand new Pentium 400’s MMX or better with 128 MB RAM or better. $2000 for 2 boxes Dragon NaturallySpeaking Medical Suite, microphones included.$500 for a nice color printer$1800 for Eloquently Stated and The Eloquent Orthopedist$500 for consulting services to get your general macros in place. These will include letter-head, footers, etc.$500 for turning your specialized ‘standard documents’ into macros or for training your staff to do the same.
Optional Costs:$1200 for 2 portable dictation machines, extra memory chips and a footpedal for the now more productive transcriptionist. Please note: These are incompatible with forms based programs such as Eloquently Stated.$125 per physician (after the first 2 physians) dictating should you wish to each utilize your own headset microphone.Professional Time Required:1 hour per person dictating to install his/her voice profile30 minutes per person selecting the words to dictate after running the topic builder1 hour per person reading the words which have been selected1 hour per person choosing standard passages from your previously dictated material from which you should make macros12 hours per person of painstakingly dictating and correcting EVERY single misrecognition. Secretarial Time Required:2 hours gathering .txt files for the topic builder and installing them on the new Pentium hard drives.2 hours gathering the standard documents for the macros?? to produce additional text macros. This will depend on the number of macros you wish to utilize.
Top |
