Voice recognition software errors: Often silly, sometimes serious


How much is 15%?

Not that much, on paper. With any drug, at least 15% of people will get some kind of side effect. Usually they all list dizziness and headaches at the top.

But what if a commonly used drug that we all viewed as fairly benign (acetaminophen 650 mg, for example) had a 15% risk of causing death or some other serious side effect? Would you still prescribe it? Or use it yourself? Probably not. There’s a big difference between saving 15% on Amazon and a 15% morbidity and mortality rate.

Could the same be true of a seemingly harmless technology?

Voice recognition software has become pretty commonplace in modern medicine but is far from perfect. I try to be pretty careful about proofreading my dictations, but many docs, especially those in emergency room, don’t have the time to. So VR errors slip by, persisting in 71% of notes.

Most of these errors are just silly and obvious for what they are. But a recent study at a level I ER found that 15% of dictations contained one or more errors deemed as “critical,” with the potential to adversely affect patient care (Int J Med Inform. 2016 Sep;93:70-3).

Communication among doctors, nurses, and all the other key players in the hospital environment is one of the most critical areas in modern medicine. So many people often rely on the initial dictation for an idea of what’s going on that a critical error can affect the way they think about the case from the get-go.

Another issue, sadly, in today’s hospital is that no one takes (or has) the time to get a patient’s past medical history. It’s commonplace to pull the history out of previous admission notes. (Admittedly, sometimes in a demented or unconscious patient you don’t have a choice.) As a result, errors of this sort tend to propagate down the line, from an admission, to the consults, to the discharge summary, and into the next admission.

So let’s get back to that 15%.

I have to assume that 15% of people being admitted aren’t having catastrophic events from medical errors, hopefully because the doctors and nurses handling patient care are thinking for themselves, recognizing dictation errors, and addressing them appropriately.

But even if we dial it down to a tenth of that, say 1.5%, it’s still a serious concern. Bad outcomes in medicine are never entirely avoidable. That’s the nature of the job.

But bad outcomes caused by too much trust in a still-faulty technology are avoidable.

If 15% of people had a serious outcome from a medication, you’d be very cautious about using it. We need to treat these technological gadgets with the same concerns we extend to drugs and procedures. Avoidable bad outcomes, regardless of cause, are never good.

Dr. Block has a solo neurology practice in Scottsdale, Ariz.

Next Article: