Big data in action: How CompSci theory helps medics fight cancer It’s all very well sucking in exabytes of data from snazzy new sensors, but what can you actually do with it all? Genetic medicine is coming along nicely but in computational terms that is the easy job, since we have a decent model of how genes work. Joining big biomedical data sets together is stupidly hard since the points …
...the best treatment is often no treatment at all. It is massively over-tested for and over-treated in the US leading to completely unnecessary impact on quality of life to treat cancers in men which (left to themselves) offered a lower prospect of harm than the treatment.
Yup. Leading to Rudy Guiliani (?sp) getting all confused about relative risk of prostate cancer mortality between the US and UK. (Everyone and his dog gets tested in the US; understandably most live for a long time. If predominately the symptomatic patients are diagnosed - a much smaller proportion - they will appear to have a much higher mortality.)
...will ask her about this paper:
http://www.bmj.com/content/352/bmj.i493
Essentially routinely collected data is heavily biased, to the point at which the direction of effect doesn't agree in ~30% of cases... that is, it is unclear whether drug (or treatment or procedure or whatever is being tested) is helpful or harmful with routinely collected data.
Don't trust observational data; it's not reliable for hypothesis confirmation. Hypothesis generation yes, but that is a very different beast.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
