Challenges of voice recognition software in Medical Transcription
All of us, at some point in our life, would have played Chinese whispers. Decades ago, this was predominantly how healthcare documentation was being taken care of. Physicians recorded their notes on the audio, and Healthcare Documentation Specialists (HDS) converted the audio files to medical records. In due course of time, we realized that, just like in the game, words tend to get misinterpreted when conveyed from one person to another, and this affected the accuracy of the document. Also, with the HDS taking care of every report from start to finish, the physicians had to be a bit more generous with their expected turnaround time.
With the advancement of technology and the introduction of Artificial Intelligence (AI) into the healthcare industry, it was believed that voice recognition software would be the solution to all our documentation woes. The big question, however, is if this software has completely replaced the human workforce. The answer to that is a definite NO. Let us investigate the reasons why.
Man, or man-made, both have an inherent need to be taught. Neither of them can function without the basic domain knowledge. Medical records prepared by HDS need to be fed to the software until it is adequately trained to correlate the spoken language with the transcribed text. The software could take days, weeks, or months to get started and create physician profiles. Until then, reports will need to be prepared in a good old-fashioned manner by the HDS. Once physician profiles are created, is it a smooth drive with the software? Unfortunately, the journey is strewn with dents that arise from dictator habits.
Warehouse (where/how’s) Damage
Dents are usually cosmetic and do not affect the functionality of a vehicle or product. Similarly, where and how physicians choose to dictate their notes can determine how unappealing the medical records turn out to be when filled with multiple blanks.
Background noise: No matter how crystal clear the physician’s style of dictation is, it will not be heard if he/she chooses to dictate in a noisy background. And what is not heard cannot be documented, neither by the software nor the HDS.
- Multi-tasking is not always a welcome skill: While dictating notes, when physicians choose to multi-task by cross-talking with their staff, the software captures every spoken word that is heard, irrespective of whom it is spoken to. Sometimes, it could be a challenge to identify the portion of the audio that belongs to the medical record and the one that is part of the conversation.
- Thinking aloud: The intelligence of the software is applied in associating every syllable that is heard to a word that has been initially fed to it. Whenever physicians think out loud, the intelligent software tends to prefix the thinking sound to the next word that is dictated, resulting in a word that could alter the meaning of the sentence.
- Clarity of thought: It always helps when physicians refer to their well-prepared notes rather than recall from their memory. Frequent corrections to dictated portions are seldom incorporated by the software. It cannot go back and forth to edit portions of a document to accommodate such corrections.
- Slow and steady wins the race: Clarity of speech can go a long way in helping the software capture every dictated word accurately. Speed and pronunciation are key to communication and speaking comprehensibly undoubtedly helps. Sound-alike words are the bane of any voice recognition software. Human intelligence is what it takes to identify and correct such similar-sounding words.
- Run-on sentences: Punctuation is completely alien to the voice recognition software. Pauses are the cues for punctuation, but software can rarely incorporate the right punctuation unless it is spelled out.
- Right equipment usage: The device used by physicians, to record their notes, matters a lot. It determines the sharpness and continuity of the audio that is picked up by the software. The more damaged the equipment, the greater will be the distortion to the audio and the lower the accuracy of the medical record.
Maintenance and Repair
Prevention is always better than cure. The sooner physicians realize the importance of their role in healthcare documentation and change their dictation habits accordingly, the easier it will be for the voice recognition software to provide neater and more accurate documents that require minimal to no human intervention. As part of the continual improvement process, physicians or physician assistants could correct words and phrases that have been repeatedly captured incorrectly and feed them back to the software. This will help fine-tune the process and complete the learning cycle.
The HDS tribe is here to stay as long as we continue to have challenges with the voice recognition software in medical transcription.