Editor's Note: Google Sticks Its Nose in Health Care
By Lee DeOrio
For The Record
Vol. 30 No. 1 P. 3
When the name of your company becomes a verb, you know you've reached rarified air in terms of consumer awareness and confidence. Google's powers spread far and wide, a diverse grouping where its unparalleled technology can flourish.
The company's faith in its tools has reached such heights that it appears to be ready to tackle a long-standing health care bugaboo: speech recognition. To quote from Google's research blog: "We wondered: could the voice recognition technologies already available in Google Assistant, Google Home, and Google Translate be used to document patient-doctor conversations and help doctors and scribes summarize notes more quickly?"
According to a Google research paper, it's possible to build a speech recognition model that can handle multiple speaker conversations covering everything from weather to complex medical diagnosis. This model would reduce pressure on overburdened physicians and the growing cadre of medical scribes assigned to capture patient encounters. The relief would come not only in the form of dictation but also in the natural conversation that occurs between physicians and patients.
In the study "Speech Recognition for Medical Conversations," Google researchers used 14,000 hours of anonymized training data to test two speech recognition methodologies, one of which achieved a word error rate of 20.1%, while the other lowered that rate to 18.3%. "Our analysis shows that both models perform well on important medical utterances and therefore can be practical for transcribing medical conversations," the study concluded.
Those error rates are not exactly comforting. However, it's important to note that when scribes annotated the training data, the error rates dropped to 8% and 14% for the two systems. This finding had me wondering what would be the results if HIM professionals such as transcriptionists and clinical documentation specialists were involved in the process.
And herein lies the problem when companies outside the health care realm get involved with day-to-day hospital operations—typically, the core group of users has little say in how the technology is applied. For example, many industry experts argue that a lack of user input has led EHRs to become the subject of physician scorn.
I have little doubt that should Google actively pursue a spot in health care-related speech recognition, it, too, would ignore advice from those in the trenches—a decision that even its lengthy track record of success would fail to overcome.