Speech Recognition and Clinical Work

Guest Blog by Emergency Room Physician, Thomas Lang, MD

At first, when I used speech recognition in creating clinical notes, I didn’t like it at all. At that juncture in time, there was a transition from paper and what I call ‘straight dictation’ (that is, a transcriptionist would type out for me what I had just rattled off into the dictaphone) into a more digital and automated world. At that time, SR simply wasn’t very good, requiring way too much editing by the user. Accuracy of SR has dramatically improved recently.


Because of my work as a locums ER physician, I am frequently moving in and out of different environments. SR is now an invaluable tool that I simply can’t live without. That said, SR is not perfect, but I am quite optimistic that there are many ways left to discover that SR can help me more optimally care for patients.

A significant proportion of physician SR tool use is done sitting in front of a desktop computer, holding a microphone of some sort. This might be fine for the radiologist, but for the person taking care of patients, this just doesn’t ‘cut the mustard’. In my view, the next horizon for big improvement will be in innovations to support clinician mobility. One significant recent advance here is the big improvement in microphone technology. It is really quite interesting to think of the amazing performance of the cheap tools sitting in my living room that I can tell what music to play. Parallel advances the external mics of tablet devices are helping to move us closer to the mobility we need to support efficient patient care.

One SR initiative that I am quite skeptical about is current efforts I am seeing to ‘listen in’ on a patient encounter using AI (artificial intelligence) and then produce a note for physician documentation. After using all types of methods to create notes (SR, scribes, ‘straight dictation’, paper, etc.), I have come to find that the most important thing to me is to create a concise note that ‘tells the story’ I want to tell ……. rather than a note that ‘has everything’ to ‘check off all the boxes’ for billers, coders, legal, and all other administrative purposes. I just can’t get my arms around how it would be possible to create a note with AI that would satisfy me without a lot of editing, and anything that requires more than a bare minimum of editing is a non-starter as far as I am concerned.

I think that the biggest problem with speech recognition today is that it is very much underutilized by physicians. Looking over the shoulder of colleagues at work who are typing away at their desktop computer workstation, if I ask them why they aren’t using SR, the most common answer I get is: ‘I tried it in the past and it didn’t work’ …… which, as someone who has gone from initially disliking SR to now considering it indispensable …. I can understand how this is the current reality.

To recap, the accuracy of speech recognition has greatly improved recently. It has great potential to further improve clinician workflow, but, rather than grand solutions in the AI realm, I am looking at how I can improve mobility for what I consider the ‘holy grail’ of physician work tasks ….. all my work tasks (notes, ‘orders’, data access) done at ‘the point of care’ and the ‘time of care’ ….. that is, while I am ‘at the bedside’, giving the patient as much of my time and attention as possible.

Tom Lang, MD

Leave a Reply

Your email address will not be published. Required fields are marked *