Tuesday, June 22, 2010

Turing Fail

A Turing Test is used to determine whether a computer can sufficiently mimic a human being in a conversation. The test has a judge conversing with two partners, one a computer and the other a human. If the judge cannot accurately distinguish the human from the computer, then the computer is said to have passed the test.

About 18 months ago (give or take), I read a letter that presents a modified version of the Turing Test. The letter bears the signature of a physician. I didn't examine the signature with a magnifying glass, but if I had, I would not have been surprised to see signs that it was a digital copy. The question this test raised is "Did a human actually write and/or sign this letter?"

The answer, I fear, is not good in either case, as you might conclude yourself. Here is the content, with PHI redacted inside [brackets]:

Dear [patient]:

At the [institution], we are committed to providing the highest
quality of care and to ensuring that our patients are kept informed of all
aspects of their care.

As the Hospital’s Infection Control Director, I oversee our Hospital’s
efforts to prevent, detect, and control infections. Despite our best efforts,
some times infections do occur in the hospital. During your current
hospitalization, you experienced a [infectious disease] which was managed
in accordance with standard practice. We regret any discomfort or inconvenience
you may have experienced as a result of this infection.

Although you already may have discussed the infection with your physician,
we wanted to ensure that you were aware. If you have any questions about your
infection or the treatment you received, please contact your physician



[physician], MD
Director, [Infection Control Department]

Most of the letter reads as if it were boiler-plate text. The purpose of the letter seems a little strange at first. Why not tell the patient that they've aquired an infection during their stay? In this particular case it was not feasible, as the patient not aware. That might also explain why no letter was recieved after their first infection, when they were.

This appears to be a letter sent out by the institution to all patients who encounter a healthcare associated infection (HAI) during a hospital stay. It isn't clear to me whether this is institutional policy or some sort of effort to comply with a local law or regulation. However, in this particilar case, the patient died a couple of days before the letter was mailed. Unfortunate timing does occur, but it made me think about the details of this case and others that must be like it. I'm told by collegues anecdotally that this situation is not rare. I am aware of more than a few similar situations personally. I find myself wondering how many family members of patients who die shortly after an HAI recieve such a letter from this institution.
My prediction in this case, is that the letter was crafted by a computer system. The impersonal style, lack of awareness of the patient condition or prognosis demonstrated, and other tell-tales from years of doing computer automation all seem to point in that direction. I also lean towards the signature being a digital copy based on two assumptions:

  1. I find it hard to believe that a provider who reviewed the patient's chart would have sent this letter.
  2. I expect that a provider signing such a letter would look at the chart.
Whether my predictions are correct or not, this process should be examined in more detail, as it can certainly be improved.

For my own part: This letter makes me even more aware that computer automation requires a great deal of thought. It may not be a pleasant reminder, but a necessary one nonetheless.


Post a Comment