Posted by: zyxo | May 5, 2010

The wrong diagnosis ? Blame your doctor !

Expert knowledge or guessing ?

In the old days, when the animals were still talking and I was in my first couple of years doing commercial data mining I used to ask to the marketeer : “If I would not make a model for you, what would your selection criteria be for this campaign ?” (gut-feeling, experienced guesses …). Afterwards I compared the results of my model with the results of his “guess” and with a control group. It sometimes occurred that the control group was even better than his experienced guess.

So much for experienced guesses.

Ever since, not one campaign evaluation could pinpoint a data mining model that did not substantially perform better than the “gut feeling” target group selection from experienced marketeers.

What would you expect ? A marketeers’ brain can consider at most a dozen customer characteristics without even being able to combine them in a multivariate model.
A data mining model on the other hand considers hundreds of variables in combination …

So I only trust in God, others must have data !

Doctors are often wrong

Amongst my family and friends I have seen several cases where doctors ended up with the wrong diagnosis. Consequence : a lot of pain, misery and sickness that could have been avoided.
Why do they fail so often ? They rely on gut feeling, they do not even use a simple check list (You know, before a airplane pilot lifts off, he works his way through an impressive checklist, even for a small, automobile-sized airplane. I once told a friend of mine who’s a pilot :” good lord, do you really need to read the user manual ?” ).

So why do they not use a computer program, built upon a real disease/symptoms database? Do they really think they can play God ?
Or are they afraid they might loose their “status” ?

Computers outperforms the doctor

This article describes the fact that a new computer application is better than a “clinical judgment” for diagnosing fever in young children. The data show that urinary tract infection, pneumonia and bacteraemia (bacteria in the blood) occur in about 7% of young children with a fever, but only 70-80% of these children are prescribed antibiotics on initial consultation and 20% of children without an identified bacterial infection are probably over-treated with antibiotics.
OK, this is a very extreme IT/medicine example, but I am convinced that a lot of sickness could be avoided if doctors would be a little more humble an use some sort of computerized aid when diagnosing a patient.

Reblog this post [with Zemanta]


  1. Medical Expert Systems was a hot research field in the 1970s. A good example was MYCIN. There are reasons why such systems haven’t yet been widely adopted, one of them being that medical doctors have little incentive to adopt them.

    One should not underestimate the difficulty of automating medical diagnostics reasoning. If you look for papers on medical expert systems on Google Scholar, you can find papers dating back to the 1950s, the very dawn of AI.

  2. Rod.
    Thanks for visiting my blog.
    Concerning “medical doctors have little incentive to adopt them”. I fear that is absolutely right.
    The question is : why not ? Why is their paycheck not linked to their medical results in stead of the number of treatments they administer ?
    The latter would probably be a good thing for the patients, but who would officially judge the results ?
    I wonder if there have been some experiments or test projects on this matter.

  3. Isn’t it true that interest groups linked to the medical profession manipulate the number of admissions at medical schools in order to control the supply of doctors and, thus, keep their privileges? Imagine how much opposition they would offer against wide adoption of medical expert systems!! I can’t blame them for acting in their best interest, but their interests are not necessarily the ones of society.

    Performance-based compensation is always a good idea, but one needs a reliable and accurate mechanism to measure performance, and that can be hard. The problem with “social systems” is that it does not matter what rules are put in place, humans will always find a way of gaming them. “Engineering systems” are much easier to deal with…

    If you want to read more on medical expert systems, I suggest the following:

    Reasoning Foundations of Medical Diagnosis (1959)

    A model-based method for computer-aided medical decision-making (1978)

    An analysis of physician attitudes regarding computer-based clinical consultation systems (1981)

    Rule-based expert systems: the MYCIN experiments (1984)

    Medical Expert Systems – knowledge tools for physicians (1986)

    Why expert systems for medical diagnosis are not being generally used (1987)

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s


%d bloggers like this: