Talking to Myself

When I was a Paid Professional Liberal doing public health in Kansas, every year at Christmas time I would take a week off and work a series of shifts in my “home ED” back in Florida. It gave the folks in the ED physician group a bit of a break around the holidays, I reminded myself that I could still do doctor stuff, and, to be frank, it was the best paying hobby that I know.

Given the changing pace of medicine and of medical practice, it was not surprising that every year there were a few new things to learn. There were new drugs, new lab and x-ray methodologies, and new physicians in the community with whom you’ve never worked. And it was always fun to play that year’s version of Admission Roulette, where the primary care physician is not on-call for himself, but is being covered by someone else; that physician refers you to someone else to admit the patient, and that person referred to is also being covered by someone on-call. Throw a consult or two into the mix, such as in an elderly patient with chronic medical problems needing admission for a hip fracture, and things get even more fun.

It was during one of these vacations that I discovered the newest (and continuing) bane of my existence. It is a particularly savage device known as the Automatic Towel Dispenser. This infernal creation requires you to pass your hand underneath a sensor to generate the provision of a single square of paper the texture of finely pressed chipboard. If, like me, you are one of those people who desires a second sheet because there’s just not enough surface area to soak up all the water in a single piece, you then have to wait until a timing device somewhere within its’ electronic bowels decides you can access an additional square. You spend these ten seconds or so frantically waving your still-dripping hands under the sensor, because you never quite know when it will give you the towel you want. It’s like playing a slot machine where you know you’ll win something, but you’re not sure when or at what cost. Nonetheless, you know it’s important to show some kind of purposeful movement at all times, a committed demonstration of intent so that all who pass by will know, “There is a man who wants a towel.” It drives me bats, partly because of the waste of time (if I see twenty patients in a shift, and the only time I choose to wash my hands is after seeing each patient for the first time, I still spend a minimum of three minutes waving my hands about in a frenetic mode), but mostly because of the indignity of it all. And in some of the patient rooms, the Automatic Towel Dispenser is located immediately behind the counter I tend to lean on while talking to patients. This means that if I get to close to the dispenser, the electric eye assumes I’m frantically waving my hand and sends off a virtually endless stream of towels until I remove my body to the other side of the room.

This is not to say I have any problem with hand washing as a concept. The Austrian physician Ignaz Semmelweiss should be praised daily for his 1847 discovery that deaths from puerperal fever could be virtually eliminated by hand washing between work in the morgue and patient care. That notion led to Sir Joseph Lister’s work on antisepsis and the emphasis on cleanliness in medical practice that benefits us today. As a past-life public health professional, I truly believe that it’s the personal, common-sense things, like washing our hands and covering our coughs and sneezes, that will do more to contain infectious disease than any other policy or program. So when I put on the long white coat and dust off the stethoscope, I’m probably more into hand washing than I need to be. Given where your hands wind up going and what they end up doing in the ED, that’s not a bad thing. But it adds nothing to the clinical effort to just stand around dazed like Amy Winehouse, wringing your hands in despair as you wait for a towel that’s still going to be too small, so you’ll need another. And it’s not like the alternatives are all that bad. If the towel dispenser is the kind that has a crank on it, you’ve got to figure that the last person who used it washed their hands before turning the crank, so the crank should be a contagion-free zone. If it’s the kind where you just reach toward the center of a bulging pile of towels and pull one out, it’s not like you have to caress the entire dispenser before reaching into the slot for that one elusive towel. (Clearly, I have way too much time…and water… on my hands thinking about this. Towel-waiting time is thinking time.)

It’s these sorts of events that make me recognize that I’m just not a technology-friendly guy. While it’s true that I have come to rely on my laptop for writing, e-mail, and Fantasy Football (The Screaming Leemers are now 3-0, while the Vocal Meerkats are 1-2), I’m otherwise pretty clueless about how most of modern life works. For example, last year it was decided that I needed a new cellphone to replace my six-year old model that finally failed to recharge. My problem was to find a phone that did what I wanted it to, which is to make and receive phone calls. A device that does so, however, turns out like failure in Apollo 13 (“Not An Option”). You have to get a phone that has text messaging, wireless capabilities, and multiple permutations of scrolling menus. And that’s just the free one that comes with the service…no upgrades for me. To be fair, I probably shouldn’t be surprised at my lack of comfort with things novel and new. I’ve been accused of failing to recognize that there has been any decent popular music written since 1990. This, of course, is not a case of willful ignorance, but a mere reflection of the fact that with the possible exception of “Love Shack” by the B-52’s (technically released on a 1989 album, but I’ll cut it some slack and round up), the proposition is absolutely true.

What bothers me about the Automatic Towel Dispenser is not that it does what it does. It’s kind of like trying to get mad at a computer game. The game does what it’s programmed to do, and there’s really very little you can do about it. (While I believe this to be true, I am equally convinced that these programs has some kind of sensor in it that detects the tension level in my mouse hand and causes an internal pixie to rub his hands together with glee just before making the screen go dark. The pixie is why my son’s “Bad Words” quarter collection keeps growing every time I play Age of Empires.) The problem, I think, lies with the people who design the systems and those who purchase them. Technology sells, but too often it sells for the sake of technology, and unless the buyer is the user the ultimate consumer is left out. For me, I’m not looking for things to be user-friendly. The towel-dispenser is user-friendly, in that it takes no effort or training to operate. I want them to be user-useful.

Let me give you an example. The hospital I work out added voice-recognition software to their medical records capabilities a few years ago. Voice-recognition software is designed to allow you to dictate your note directly into a computer, and the computer will print out your dictation in real time without the need to go through a transcriptionist. On the surface, it looks ideal. Real-time data can be generated in print and is available immediately for clinical use, or can be stored indefinitely in an electronic format. Clear, legible, typewritten documents lead to more accurate billing and provide medicolegal security. Costs are lower in the long run because you remove the human typewriter in the middle. What could be better?

Technically, nothing. It’s a great premise, and the software indeed does what it’s supposed to do. If you give it words, it will take what it “hears” and put them on paper and in an electronic form for all the purposes just described, and it does a great job at it. And it’s a fairly “user-friendly” format…it takes some time to get used to the system, but the learning curve is not bad. In some ways, it might even be able to speed up the documentation effort; for example, you can program it with a standard exam so if you tell it “lung exam” it will print out a preset response (“Lungs are clear to auscultation bilaterally without wheezes, rales, or rhonchi”).

But it’s sometimes not user-useful. There are times it takes longer to get the system to transcribe your words correctly than it does to simply dictate a chart. The system is said to have a recognition accuracy of 90-95%. That’s really very good for this kind of software. But what it really means in practice is that the system will misunderstand, and transcribe incorrectly, one out of every 10-20 words it hears.

That may not seem like a big deal. For example, let’s look at one the Introduction to the Declaration of Independence with every tenth word removed (starting with word #5):

When, in the course human events, it becomes necessary for one people to the political bonds which have connected them with another, to assume among the powers of the earth, the separate equal station to which the laws of nature and nature's God entitle them, a decent respect to the of mankind requires that they should declare the causes impel them to the separation.

The study of linguistics and human factors engineering tells us that humans are pretty good about filing in the blanks, especially knowing the rules of usage of a particular how language and the cultural context of the communication. But the computer can’t do that, because while it may know some of the structural components of language, it knows nothing of context. So if read at face value, the dictation may seem disjointed at best and confusing at worst.

The other key factor is that for the system to transcribe a word, it has to “hear” it. That is, it has to be able to clearly delineate a word from other speech on either side of the word and from any background noise. In many ways, this is operator-dependent. People who speak slowly and clearly, using short words and brief phrases, have less problem than people like me who tend to talk fast and use lots of punctuation, causing the computer to run words together and transcribe what it thinks it heard rather than what was said. Earlier this week I was asked about a record where the computer had documented that, “A ride was injured by the radiologist,” rather than the intended, “MRI was interpreted by the radiologist.” (To be fair to the machine, this is no different than what we do when we think that Manfred Mann was blinded by the light while wrapping up a feminine hygiene product in the middle of the night, or when we attempt to attribute any lyrics whatsoever to “Louie, Louie.”).

This effect can lead to some pretty funny moments. For example, when I dictate a physical exam, I tend to use the word “cardiovascular,” as in “cardiovascular exam shows there to be a regular rate and rhythm without murmur.” The software has continually interpreted the word as “cremaster.” The cremaster muscle covers the testes, and raises or lowers the scrotum in order to regulate the development of sperm. According to the software, I’ve been listening to scrotums for well over a year, even on women. (Here’s a home experiment for you. The Cremasteric Reflex occurs when you stroke the inner part of the upper thigh and the scrotum and testicle on that side are pulled up towards the groin. Try it with someone you love. This is, of course, to be distinguished from Peter’s Sign, the radiologic finding that on pelvic x-ray, the male member always points to the side of the problem.) And knowing that these things happen, we sometimes set the system up to fail. A few weeks ago I started to dictate some phrases I always thought were funny into the system to see what would come out. “Ball peen hammer” emerged as “balls penis ham her.” (Don’t worry, Risk Management…it’s already deleted).

Fast talkers like me also cause the software to often not recognize shorter words such as articles and prepositions, and for whatever reason when I use the system it tends to leave out the word “no” quite often. This can produce radical shifts in meaning; there is a world of difference in the phrases “patient has chest pain” and “patient has no chest pain.” Some physicians with stronger accents can also cause the software problems, although the program does have the capability to be “trained” to the particulars of an individual over time. But while a human transcriptionist can slow down a tape to listen again to what is being said, and through training and experience has some sense of context, the computer has no ability to do either. And while you could legitimately say there is a responsibility for the physician to go back and check his work, you might also reflect on how much time it would take away from the work of the department to go back and correct by hand every tenth word as well as any unusual phrases like I’ve noted above. So we tend to trust in the software, and use both the known limitations of the program and the clinical context to clarify any confusion.

And then there’s the Law of Unintended Consequences (I could make a joke here about marriage and children, but I demur). The ED where this software was first installed was a busy, noisy place, and the station where the physicians and clerks sat was right in the middle of the floor show. The docs sat on one side and the clerks assigned to work with them sat on the other, and usually there’s a pretty lively back and forth banter as orders are given, phone calls are transferred, and both printed lab work and rubber bands fly across the table. That background noise, on top of the unique sounds associated with the ED and other twenty-four hour establishments such as psychiatric holding facilities and fraternity houses, interfered with the software’s ability to recognize specific words and phrases. However, you couldn’t have a doctor continually leave the ED to go to a soundproofed area. So a glass partition was installed that divided the desk in half, docs on one side and clerks on the other, with some additional extensions that enclosed the physicians from both the front and the sides. That left open the back, which happened to be immediately in front of rooms used for patients whose clinical condition required close proximity to the core of the ED, but were unruly about their visit to our facility. These noises would echo within the canyon of glass, which in turn made dictation even more difficult when the party was really going strong. The dictation effect was irrespective of the fact that it was harder to talk to the clerks over the wall, and sometimes it took a running start to make sure the printed lab work got over the partition when flung at the doctor for review. To counteract this problem, a three-inch-wide slot was cut in the glass so you could order a Big Mac Value Meal and a CT scan of the abdomen and pelvis at the same time. (I will confess that, at the end of a night shift when the caffeine was good for one last burst of creative energy, I actually printed out a sign that said, “Mr. Gorbachev: Tear Down This Wall!” and taped it to the glass.) We have now moved into a newer, and much more tranquil, ED that features modern, state-of-the-art design and mood lighting. Working there feels a lot like being in the booth with Venus Flytrap. It’s just a matter of time before we light the incense and install a lava lamp to liven up the place.

While frustration with systems such as voice recognition software inevitably leads to calls for culpability, in reality this state of affairs is no one’s fault. The system designers and vendors made the system to do exactly what they said it would. The purchasers did not buy the system in order to impose on others. They saw a need and found a way to answer it. The health care system has more silos that Kansas, and it shouldn’t be surprising that a system purchased for the benefit of one part of the system might hurt the function of another.

I’ve been speaking here of a hospital-based system, but the use of technology is exploding within prehospital care as well. And while my example is of an”administrative” use of technology, patient care technology has the same limitations. Point-of-care testing may be technologically feasible, but what benefit is there if the result doesn’t make the paramedic provide a better level of care? If the ultimate user of medical technology is the patient, how do a few extra minutes on scene to use a device add value in an urban EMS system?

The solution to the problem of user-utility is ultimately one of communication. For whatever reason, designers don’t talk to purchasers, buyers don’t talk to users, and users seldom provide feedback to those who build or buy the systems foisted upon them. Agencies and institutions making large capital investments in technology should as a matter of course convene multidisciplinary teams to evaluate the products for not only their ability to get the job done, but to do so in the most clinically effective way. This is especially true when new technologies force changes in established patterns of work between different organizations. For example, while I value electronic EMS records for billing, review, and research purposes, it drives me nuts as a clinician that in many cases I can no longer get a copy of the EMS record from the prehospital crew before they leave the hospital.

So if you’re one of those people who’s developing a new application for health care, ask your friendly neighborhood doctor, nurse, or EMT what it might mean to him. If you’re buying a system, make sure it not only does what you want it to do, but also helps your staff do their job the best way that they can. And if you’re the one who’s had some new device dropped in their lap, accept what the technology is supposed to do and find a way to help it work better. It’s not helpful to rage against the machine unless you actually understand what you’re raging against. And only when the batteries die can you rage, rage against the dying of the light.

(With inadequate apologies to Dylan Thomas, “Do Not Go Gentle Into That Good Night” should be required reading for all. There’ll be a test.)