Pancreatic cancer really pisses me off.
I’ve known too many good people struck down too early by it, and in two generations we haven’t really improved cure rates significantly. Besides being highly resistant to most forms of chemotherapy, pancreatic cancer is almost never diagnosed at an early stage. By the time patients develop the typical belly pain radiating to their backs, jaundice, or loss of appetite, it’s already too late. We’ve yet to find a reliable blood test or routine scan that’ll help, and that pisses me off even more. And though most every pancreas cancer patient can, in retrospect, identify vague symptoms that started months beforehand, they aren’t enough for people to seek care – or much for a doctor to go on if they do. We’re badly in need of an innovative idea.
But whaddya know, Microsoft has one –and it’s way out of left field.
Turns out, most people with undiagnosed pancreatic cancer have recurring patterns of internet searches in the months preceding diagnosis. A team of Microsoft analysts parsed the internet search histories of 9.2 million people over 18 months, and identified query clusters – things like “itching”, “oily stool”, “taste changes” – that could predict who had undiagnosed pancreatic cancer and who didn’t.
(Ignore that popping noise, it’s just my mind blowing out of my ears.)
The potential applications of this concept go far beyond pancreatic cancer. Besides screening, this approach could refine our understanding of symptom patterns in almost any disease.
The approach is a long way from workable, and there are some very thorny privacy concerns: most people have a healthy skepticism of “big data”, and the intentions of those who mine it.
But the concept surprises and excites me.
The full report is here and reported in the New York Times here.
