Search This Blog

Follow adrianbowyer on Twitter

My home page

Wednesday, 6 April 2011


When our friends are ill, we often know it just by looking at them for an instant.  The lineaments of disease are writ on their physiognomy, and good doctors must become adept at making provisional diagnoses as soon as their patients walk through the door (assuming they are up to walking).

Now.  There are a lot of dead people - their number is increasing all the time; and they all died of something.  In the last century or so, almost all of them will have been photographed as well, and many will have been photographed towards the ends of their lives.  It should be straightforward to establish a truly vast database of late (in both senses) portraits linked to causes of death.  If all the portraits for each cause were then to be averaged, we'd have a typical face for pancreatic cancer, and another for a dissecting aortic aneurysm, and so on.  Non-lethal conditions could also be catalogued in the same way.

A website could then be established to which people feeling poorly could upload their picture from their webcam.  The picture would be compared with the database, and back would come the ten most likely diagnoses, in order of probability.

It would be hypochondriac heaven.

It would also automatically become more accurate with time.  This has nothing to to with the adding of extra data (though that would happen too), but is a specific instance of the general rule that people who exhibit text-book symptoms for a disease have a higher Darwinian fitness than those who don't.  They are more likely to get the right treatment, and in turn are then more likely to survive and to reproduce, passing on their text-book-symptom tendency to their children.  By this means, medical text books also become automatically more accurate with time, even if neither a word nor a picture in them changes.

It would be interesting to test this idea by adding an arbitrary unconnected symptom to a disease in medical training - say that diverticulitis often causes a slight rash along the hairline above the ears.  At the start, the correlation between rash and inflamed diverticula would be no better than random.  But as time went on, and the people exhibiting the rash lived to have more children, the initially nonsensical symptom would become a real one with diagnostic power.

In fact, it may be that many symptoms that you and I exhibit when we are ill have already been established in this way, dating right back to the preposterous diagnoses of pre-scientific physicians in the eighteenth century and before.   The evolutionary mechanism just described will have made them self-fulfilling prophecies.

All this assumes that treatments are effective, even if the diagnoses start off being based on random rubbish.  But, of course, the same evolutionary pressure will work on random treatments too.  So a witch doctor diagnosing demonic possession for what is, in reality, porphyria, and prescribing a course of viper venom is applying a selection pressure on the unfortunate patient to get cured by viper venom.  The first diagnosis and treatment will work no better than chance, but persistence and consistency will make viper venom a cure for porphyria in the long run.

Could this be the basis for much of medicine?  It doesn't matter what the diagnosis and corresponding treatment are at the start.  As long as they are applied consistently and over many generations, they will come to work effectively as the patients evolve to match them.

No comments:

Post a Comment