By Eric Fishman, MD
In the beginning there was memory. The physician’s memory was the original repository of the patient medical record. Memory was supplemented by handwritten notes on papyri in ancient Egypt and Babylon and on paper from medieval times to the 20th century. With the advent of recording devices in the 20th century, handwritten notes gave way to the infinitely more time effective practice of dictating patient notes into a recorder which were then transcribed into a typewritten or word processed document. Of course, that practice introduced an extraneous third party into the medical record keeping process: the transcriptionist with the attendant additional expense to the physician and loss of privacy for the patient.
In the 1990’s, speech recognition software came to market and products such as Dragon Medical® enabled physicians to either speak directly to their computers to produce patient records and reports instantaneously or to continue dictating on recorders which could then be ported onto a computer so that a digital voice file could be transcribed by the software to produce the patient record. The transcriptionist was often eliminated from the process and efficiency increased in terms of time and money spent to produce the patient record.
At the current time, 21st Century technology offers physicians and health care providers a medical record paradigm that will not only vastly upgrade the process of producing, maintaining and safeguarding medical records but will, in a direct and fundamental way, actually improve the quality of medical care. The technology is Electronic Medical Records (EMRs). EMRs produce the most accurate and complete patient health record possible to date and help physicians practice better medicine as well. EMR technology is available in a plethora of shapes and sizes with a great variety of possible features. The technology can change the way you interact with your patients, from before they make their first appointment to after they’ve left your office, and have questions about their visit in your office.
As a practicing physician you are aware of the repetitive nature of some aspects of your practice, specifically with regard to patient diagnosis. It is very likely that you and/or your staff have asked the same or at least very similar questions to each of the thousands of patients you have treated. Unless you are practicing in a tertiary referral center, and never see the same condition twice, the patient answers likewise tend to be repetitive. Similarly, physical examination findings fit into certain categories that are seen over and over again. For this reason, most of the current high-end Electronic Medical Record products very capably utilize ‘pick lists’ or ‘click and point’ methodology to complete large portions of the patient medical record.
These point and click systems are particularly adept at documenting, for instance, allergies to medications, medications that are currently being taken, past medical history, family history, social history, and major portions of the physical exam. This is the case because of the narrow range of options which are available as patient responses. For instance, your patient either smokes or doesn’t smoke. And if he/she smokes, it is probably 1 ppd, or 2 ppd, or some other value that can reasonably easily be foreseen by the experts who have designed the point and click system for your office.
However, the historical portion of the patient medical record typically has a great deal of information that cannot be easily foreseen by the developers of the point and click templates. For instance, as an Orthopedic Surgeon, my patients frequently find themselves in automobile accidents. It is not likely that the author of whichever EMR may find its way into my office has contemplated the various street names and intersections in my community. Therefore, in a typical point and click system, there will be a scarcity of relevant information concerning the specifics of the accident. And I find that these specifics are important for a wide variety of reasons, not least of which is that they remind me of the particulars regarding this patient when they return to the office. Utilizing templates for the historical portion of the report, while feasible, tends to produce extraordinarily repetitive reports, each of which sounds not only vaguely similar to the previous patients, but in many cases essentially identical to other patients. This certainly makes it difficult to recall the characteristics of this particular patient.
Based on extensive, personal experience I can unequivocally state that production of a medical record with a system that restricts the user to point and click templates exclusively is far from perfect. At one point in time I undertook the academic exercise of trying to put together a decision tree regarding the mechanism of injury for the occupant of a motor vehicle who was involved in an accident. I started with a Microsoft Access database and started branching. First, was my patient the driver, front passenger, or rear passenger? For each of these 3 options, I then wanted to know if there were 1 car, 2 cars, or multiple cars involved. We now have 9 options. Then we need to know if the car was stopped, traveling slowly, traveling at a moderate rate of speed, traveling quickly, or traveling at an unknown rate of speed. With these 5 options we are now up to 45 mechanisms of injury. Was a seat belt in use; was the headrest properly positioned. 4 more options, and now we are at 180 choices. Was the vehicle impacted from the front, rear, left side, right side, left front, right front, left rear, right rear. Another 8 options to consider. Next, did any portion of the patient impact any portion of the vehicle? This gets a little more difficult to enumerate, but I classified the body into 8 different areas that were likely to have been injured. Then I wanted to know was there a substantial amount of damage to the vehicle, etc.
Eventually, I had an untold number of different mechanisms of injury for an automobile accident, each of which could very easily be documented for each of my patients. It would merely take a few clicks of a mouse, and I immediately had paragraphs of text on my document which covered, or so I thought, essentially every possible option.
Did it work? My answer is a resounding NO. This was, as I mentioned, a ‘home grown document management system’, and I could customize it at will within my office. I found that in at least 50% of the patient visits, I needed to change the templates since I had not, in fact, taken into consideration every possible mechanism of injury.
So, what did I do? Like any physician who knows that he can make the system work, I modified the templates each and every time that I found it necessary to do so. I figured that after seeing ‘enough’ patients, the percentage of new patients who didn’t fit into the predetermined potential mechanisms of injury would diminish, and when it was below 10%, I’d stop customizing it. One flaw persisted; the number of new patients who didn’t fit into the mold never went below 50%. It just stayed there. No matter how many different options I placed into the template, my patients just hadn’t reviewed my EMR before having their accidents, and therefore hadn’t complied with my predetermined mechanisms of injury!
I either needed to choose a different patient population – one that was more considerate of the templates that I was using – or I needed to discontinue the use of the templates. I didn’t like either option, and I quickly determined that the middle ground was best. I kept my patients, and I kept my templates. However, I utilized the templates only for the portions of the reports for which they are most appropriate. I need to test the range of motion of the spine on all of my patients who complain of spinal pain. The AMA published a book which tells me what the ‘Normal Values’ are. I made templates out of the MS Word macros I had been using for years, and in just a few short clicks, the entire range of motion was documented.
But what did I do for the pesky history that was so difficult to document properly? I resorted to the state of the art of the 1990’s and implemented Dragon Medical speech recognition software. I dictated the history using Dragon, and used templates for the rest of the medical report. It worked perfectly.
Therefore, I recommend to any physician that he or she integrate Dragon Medical speech recognition software into their EMR system to produce the best and most complete patient health record. This is because the pre-determined templates of most EMRs simply cannot anticipate the full spectrum of facts presented by the patient which must be incorporated into the historical portion of their health record. The historical portion of the patient medical record should not be truncated to fit a narrow range of template options and speech recognition technology enables the physician to dictate the entire patient history and any other patient data into the EMR.
One of the advantages of an EMR is that it allows physicians, hospitals, insurance companies, pharmaceutical companies, medical societies, and other parties entitled to view the patient data for legitimate, permissible purposes, to do so. Legitimate, permissible purposes include coordinating patient treatment, accessing diagnostic procedures and results, preventing adverse drug reactions, and ensuring medical practice within clinical practice guidelines. One particularly high priority purpose from the physician’s standpoint is that the data be accessed by third party payors to streamline reimbursement for services.
An issue that arises with the integration and use of speech recognition software in an EMR system is that the voice dictated text may not be maintained adequately as ‘data’ and therefore may not be appropriately parsed with current technology. It is therefore not easily accessed by third parties, and may not be able to be utilized, for instance, for E/M coding. However, the dictated text is generally not information that third parties require. For example, in the context of the particular facts of an automobile accident, I do not believe that the E/M code will be different for a front-end, 1 car collision than it is for a 2 car rear-end collision. Similarly, it is highly unlikely that there is any difference in medical care for a patient who develops chest pain at their favorite local restaurant, as opposed to developing it while they were at the fair. Again, if the child’s earache started after they swam in their backyard pool, as opposed to the local municipal swimming pool, the treatment provided is likely to be similar.
Even though the historical portion of the medical record which is dictated does not constitute ‘data’ that can be captured by third parties, given the current state-of-the-art, this is a slight disadvantage with no substantive impact. Any perceived disadvantage is significantly outweighed by the benefits to the treating physician and the patient of having a complete patient history on file which the physician can access and utilize in rendering proper medical care.
There is a trend towards utilizing artificial intelligence to parse the dictated historical portion of the record so that the EMR can capture it as data. However, I do not believe that there are any reasonably priced, commercially available products which do this with sufficient accuracy at this point in time as to be practical. However, in the not too distant future, this is likely the direction that dictated text will take.
One of the limitations of speech recognition technology in the past was that the level of accuracy only reached 96 or so percent. However, the latest versions of the software, combined with modern hardware configurations, has resulted in a 99% accuracy rate except for those physicians who speak with particularly unusual and heavy accents. It is generally accepted that a 98% accuracy rate for a human medical transcriptionist is the standard. Most physicians should be able to obtain at least 98% accuracy without any significant difficulty, and 99% accuracy is more and more common using the current version of Dragon Medical. In fact, I have seen published reports of physicians with accuracy rates of over 99.5%.
Another factor which may concern a physician is the need to wear a headset microphone which is physically attached to a computer. While this is the ‘standard’ mechanism of utilizing voice recognition software, there are a variety of options. One is the use of a wireless headset, or even an array microphone which can be attached to the computer, but not to the physician. A more and more common alternative is the use of a high-end digital recorder, such as those by SONY®, Philips®, or Olympus® to capture the spoken voice, which can then be ported to a computer that then analyzes and transcribes the dictated text. In situations such as this, the medical office will frequently utilize the services of a ‘revisionist’, to review the text, as opposed to a transcriptionist who classically types it from scratch.
In step with hardware that permits computing in a mobile environment by means of a tablet PC or laptop, the newest release of Dragon Medical includes a roaming user profile. This allows physicians to dictate, using their own personalized user profile, from any of a variety of networked computers, PCs, tablets, etc.
The need to integrate speech recognition technology into EMRs has not escaped the notice of the EMR manufacturers. It is no longer the exception but has become one of the standards by which EMR products are evaluated. Does the EMR have a speech recognition feature? Is it compatible with Dragon Medical?