Speech Recognition (SR) Software is the magic mantra luring physicians, professionals, and corporations alike. None in the transcription/documentation vertical are spared from this whirlwind. The management can hardly resist its claims of multiplying the output exponentially. The transcriptionists can not afford the long hours of energy drain with manual typing. Every reasoning against it is dwarfed to nothing by the arguments of technology adaption, evolving with time, moving on to better tools, salvaging time, increased production, greater revenue, etc.
The voltage of this influx is so strong that it burns out every skeptic into ashes. If you pause even for a moment before bowing to its sovereignty, you are considered a troglodyte, an unbeliever of technology. You might even be stamped an outcast from its parish. So, here it comes:
We are witnessing one of the major milestones of the 25-billion dollar industry, growing every year by 15 percent. The medical transcription is an allied health profession, converting physicians’ voice-recorded reports into text. The average number of documents produced per day is simply as innumerable as the number of patients treated each day throughout the United States of America. As gigantic as the task is, the process and the product thereof are in strict compliance to state, federal, and other regulatory bodies, particularly the HIPAA and JCAHO.
This four-phase process of receiving the dictations, transcribing, quality assurance, and delivering the product is evolving into a slightly modified one. Instead of manual transcription, the SR software is used to automatically convert the physicians’ words into electronic text. This quality of the output varies depending on the efficacy of the software to some extent, although it is improving by the hour. But, as explained by the CEO of a transcription firm, theoretically the input is as varied as are the number of physicians in America. Thus, technically speaking, it is next to impossible to conceive of the possibility of a uniform and useful output when the input cannot be streamlined.
Yet, the SR software is creeping its way into almost being synonymous to transcription itself. The SR output is then given to transcriptionists, who now will have to simply go through the sentences and correct them for accuracy in grammar, medical language, spelling, etc. The transcriptionist’s job is made easy and the production time cut down at least by half. The management is elated with the prospects of revenue being doubled.
Well then, now comes the most important part of the argument. How reliable is this process? How efficiently are the transcriptionists correcting the documents? As discussed, the output can never be consistent, even for the same physician who could be in a hurry one day or suffering with cold affecting the tone of his voice. The same transcriptionist who is almost starting to trust the software suddenly wakes up to a rude feedback from the quality team. He or she missed out on some blunders. Their work is unacceptable. Explanations are called for. Their ratings are down. Their remuneration is crumbled.
The quality team generates a report listing the errors looking something like this:
Case I:
ASSESSMENT AND PLAN:
1. Hypertension.
2. Osteoarthritis.
3. Diabetes for hyperlipidemia.
The word ‘for’ above is an unacceptable error. The physician actually dictated “Three. Diabetes. Four. Hyperlipidemia." Meaning the numerical 4, setting off ‘hyperlipidemia’ as another point in the assessment. The software misunderstood this owing to its lack of reasoning abilities and of course due to various other factors, such as, the physicians speed, low volume, recording quality, etc. This was missed by the transcriptionist who made some assumptions calculating from the software’s seemingly good quality at other places. Also, remember that they are racing against time because their remuneration is directly proportional to the volume of their output.
Case II:
Material grandmother. Should have been ‘maternal’ grandmother.
Yes, the SR software by definition is capable of such errors. As funny as it sounds, the fun part disappears in a fraction when the client who paid you is made fun of by another physician or attorney or a patient who noticed the cheap quality of his medical records.
Case III:
His fear of public encounters was noticeable in how he ‘cleans’ to his wife throughout the visit.
It should have read, he ‘clings’ to his wife.
Medical records are legal documents, and are subject to the laws of the country/state. A physician/hospital is accountable for the same. These records are signed by them with date for legal purposes. The repercussions of such errors are virtually uncontrollable. The physician’s/hospital’s reputation is shackled. The quality of clinical documents, under scrutiny by federal governing bodies, is compromised. Risking the patient’s health in some cases can evoke the possibility of a lawsuit. These documents have the potential to bring down mighty insurance companies to their knees or destroy the career of a physician, per se.
So, how do we weigh the benefits of this software against its shortcomings? Should we wait and hope for it to evolve more. Can the transcription firms afford that thought? Should we go back in technology and time, and get back to manual typing. What will it take for this software to come any closer to human brain?
But most importantly, the toughest question now — Is it worth the risk experimenting Speech Recognition Software in an industry as crucial and vital as the US Healthcare System?
