HP
Copyright 2009
Firstly, let us look at a simple example of AHP in action before we discuss how it has been applied to a medical situation.
Suppose you had a dilemma as to where to send your child to school. School A has better teachers but School B has a better range of subjects. So the options are the schools and the criteria are range vs teacher quality.
We have to assess the schools against each of these criteria. Suppose we prefer school A with respect to the teachers? criterion by a margin of 70% to 30% and we prefer school B with respect to subjects by 60% to 40%. Now we have to assess the relative importance of the two criteria. Suppose that we take them to be equally important, i.e 50%-50%. It is then evident that school A would be selected since its margin of preference on equally important criteria is greater. You can also see that changing any one of these ratings would provide potentially a different outcome. The ratings of criteria are called weights. This is the essence of AHP except that there some very innovative ways of arriving at these judgments which we won?t cover here.
In general, the criteria can be decomposed into sub-criteria, a necessary step for more complex situations. For example, in a recruitment decision, the candidates are the options and the criteria could be ?communications skills?, ?presentation skills? and ?experience?. ?Experience? might be decomposed into ?technical? and ?non-technical? while the ?technical? could be further decomposed into ?hardware? and ?software? etc. You can see how a hierarchy (the ?H? in AHP) is developed to represent the structure of the decision. This provides a common focus for discussion and deliberations.
Now for the medical diagnosis model: Here the options are the potential illnesses that might be afflicting a patient and the criteria are the symptoms to which particular illnesses relate. Here is the part that really interested me and I would be keen to get the reaction of Ash and others. The presenter (an animated Italian brimming with enthusiasm for his work) of the model for this application made the following claim: Traditionally doctors diagnose by means of a process of elimination. They perform tests relating to certain diseases and if the test is negative that disease is discarded and another hypothesis is examined. And this is not the best way to do it, the danger being that once discarded (possible due to the misreading of test results or other errors), a disease is unlikely to be revisited. Rather, he claimed it is best to continue to add potential diseases and keep doing so. The reasoning is that the weights of AHP will ultimately reflect the relative likelihood of these diseases being responsible for the illness. Now where do the weights come from? This was not made clear in the presentation but I think it has to be through a collection of data from many patients over a long period of time.
He further claimed that medical diagnosis as currently performed is only within the range of 40% to 60% successful whereas using the AHP model, success rates have been found to be 87%.
At a later time I might discuss the concept of personalized medicine which I think holds the key to improved future treatments once a diagnosis is made.