"OUTLIERS" Are they telling us something?

Linda waits patiently for me outside the British Museum

You know what an outlier is.  It's a piece of data that just doesn't fit.  It's a hockey player with all his teeth.  It's a short, swarthy, jovial Norwegian.  It's a McCain sticker on a Prius.  It's something that makes you look  again, and say "Huh"?

Back many years ago when I was doing real research I encountered lots of outliers.  Everybody in my field – most fields, I suspect – encounter their outliers.  Most outliers are nothing more than experimental error. Examples abound.  For instance, consider paleomagnetism.  A paleomagnetic outlier might be a direction that seems wrong; it doesn't match the rest of the measurements.  It might represent lots of things, all errors:  the rock you sampled had rolled down the hill  and you didn't know it; you wrote the wrong orientation numbers in the book; you read the wrong end of the compass needle; you recorded the wrong time of day; you gave it to the wrong graduate student to measure – etc., etc., etc.  I once wrote a paper describing a way to objectively identify a true outlier.  The purpose was to scrupulously avoid the temptation to retain – or pitch – an outlier depending on its influence on whatever you are trying to prove or disprove.    My simple little paper got trashed by two statisticians, with ill-concealed disdain.  I still think I’m right, but I’m afraid to say so in print.

The thing is, we all recognize that the outlier might not merely represent a mistake.  It might be trying to tell us something.

Well, an outlier spoke to researchers at Memorial Sloan-Kettering Cancer Center.  They investigated a drug called everolimus for its effect on advanced bladder cancer.  They concluded – using the conventional  blindingly confusing statistics – that it was a failure.  Into the dumpster with you, everolimus!

Fortunately, they noticed that one woman – out of 45 – showed a remarkable improvement.  In fact, she has now been in remission for 2.5 years and is going strong.  So they sequenced the genes of her tumor cells, and found “inactivating mutations” in two genes – TSC1 and NF2, if you are curious.  Then they tested other people with advanced bladder cancer and found that those with TSC1 mutations fared better than those without.  None of these other patients had the mutated NF2 gene. 

So, I wonder – what is this trying to tell us?  It clearly tells us to keep a little everolimus on the shelf, for that unfortunately rare case where a bladder cancer patient has the two mutated genes.  (They are doing further testing, of course.)  To my simple way of thinking, it also tells us that they should study just what in heck those two genes do.  Basic biology, guys. 

  

I got this from the 9/4/12 edition of the National Cancer Bulletin.