More illness = Good practice

I have had my attention drawn to this remarkable document distributed to all GP practices in London:

An Introduction to a Pan London Approach to Improve Quality, Access and Patient Experience in General Practice


It includes all manner of cunning wheezes to improve General Practice in London. On particular piece of genius includes using predicted* versus the actual GP reported prevalence of various diseases as "an indicator of general practice performance":

Organisations in the top quartile ranking within London are examples of good practice. Monitoring, intervention and support would be required for those organisations within the bottom quartile ranking in London. Monitoring and support would be beneficial for those organisations within the mid quartiles within London.

That's right. If you are fortunate enough to have a population with a lower than predicted rate of an illness then you will be judged to be an example of bad practice and failing. But if you happen to have higher rates of illness for some reason, then well done you, you're an example of "good practice"!


Now I'm no epidemiologist but population level models like this cannot be used to accurately predict the prevalence of disease at a local level** - to tell you what the real burden of disease is that those crappy GPs are failing to detect. A five variable* regression model does not capture all the variation in human disease however much you want it to. The tail is wagging the dog here.




* "Expected prevalence data are derived using expected prevalence rates provided by ERPHO which take account of age, sex, ethnicity, smoking status and deprivation score at practice level."

** They are based, of course, on taking all that local level data and then finding a best fit line between all the real data points - you can't then go back and say that those data points that don't lie on the line are now wrong. What kind of statistically illiterate fool came up with this idea?