Mirages and moving classifiers

The joys of new technologies are many and varied. There are now ways of communicating even within NHS hospitals which were apparently designed to block all mobile phone signals, to see veins where none appear to exist and to allow a child within a radiotherapy fixation mask to become a superhero. Technologies are brilliant; they brought us in primate days from being able to see the tasty ant in the tree crevice to eating it, and from Victorians witty after-dinner conversation to millennials Twitter scrolling silently.

They also let us see things we couldn’t before. Smaller, subtler. The first glimpses. And in this way, they may accidentally open up a range of misdiagnosis, extended diagnosis and well meaning mismanagement. The phenomenon of ‘too much medicine’ we have explored before (like here) but there’s an even subtler variant in the way prognosis can shift with a changing lens.

Imagine you have six children with a malignancy. Two have definite local disease; overall survival (OS) 80%. Two have obvious metastatic disease; OS 40%. The other two have subtle, low level ‘oligo’-metastatic disease; OS 60%. OS for the whole population is 60%. If you only have the imaging to detect the big lumps, the met group has an average OS of 40%; the ‘local disease only’ group is 80 + 60 / 2 = 70%. Now add a new technology, don’t change the treatment success at all, and you have improved the ‘local disease only’ group to 80% OS; this makes sense, you’ve got rid of those with spread disease. You’ve also made the ‘metastatic’ group better (40 + 60 / 2 = 50% OS) by diluting these with the better-player oligometastatic group. Your new SuperFerretGlowatron 300 has improved outcomes in both groups! But …whole population survival remains unchanged at 60%. There is no real improvement. It’s a mirage.

Trust nothing. Question everything. Stay appraised.

  • Archi

(Visited 27 times, 1 visits today)