Thursday, January 22, 2009

Less is More, More or Less

Much of our decision making process in based on our perception of reality. We may have a “gut” feeling about something one way or the other but what exactly causes this? Gerd Gigerenzer describes simple rules of thumb which become the basis for our decisions . The rules of thumb ignore other information. For example, we have a recognition rule of thumb or heuristic that enables us to process something familiar to us very quickly. Studies have shown that many times keying in on a couple of the most important variables is a much more valuable predictor of a future outcome than many weighted variables. The art is knowing which information to ignore.
An example is an ingenious study design in which used a very sophisticated set of computer generated variables to try and predict which Chicago schools would have the highest rate of dropouts. The data from the first half a of schools were used as a predictor for the second set of schools. It turned out that asking two questions fared better as a predictor of the future “What was the attendance rate” and if that was more or less the same moving on to the next question which was “what were the test scores”.
The healthcare arena has many examples. One of the most compelling was some research done at the University of Michigan on coronary care. It was found that predicting heart attacks as a basis for ICU admission was no better than chance. Much of this was attributed to the spector of malpractice ans so called “defensive medicine”.This meant that you had a 50/50 chance of being placed erroneously in the ICU with the associated risks of infection etc. It was discovered that too much emphasis was place on medical history questions such as a history of diabetes or hypertension. After constructing a very elaborate logistical regression instrument results improved as expected. Many of the doctors did not like using this method because they found the charts cumbersome. When the instrument was taken away surprisingly doctors were able to predict about as well because they were armed with a better internal algorithm for making a choice after being exposed to the more complex instrument.
It turned out that the most important question was whether or not there were ST segment changes on the EKG and if so the ICU was statistically the right choice.. The next most important question was whether the admission complaint was chest pain and if not a regular nursing unit bed was more appropriate.

No comments: