From: Robin Hanson (hanson@econ.berkeley.edu)
Date: Mon Apr 05 1999 - 13:24:15 MDT
At 05:19 PM 4/3/99 -0500, you wrote:
>>>It's implausible that the medicine we get (as opposed to additional
>>>treatment we don't normally get) doesn't help. People get things like
>>>appendicitis, pneumonia in the young, and gangrene which were major risks
>>>in the past but very rarely die of them.
>>
>>Sure we are lots healthier now than in the past. The question is how much
>>credit medicine deserves for that. Lots of other things have changed
>>besides medicine.
>
>That's why I picked those. Gangrene and appendicitis still have nearly
>100% mortality if not treated. Pneumonia is generally survivable by the
>young but still often lethal if not treated.
The question here is about the marginal health consequences of medicine,
*averaging* over all the things medicine does. Some things may help, but
when averaged in with other things that hurt, the average could be zip.
>>>>A recent analysis of 5 million Medicare patients, using regional
>>>spending variations of a factor of two (controlling for lots of stuff),
>>>>found that any mortality benefit of spending in the last six months of
>>>>life is less than a one part in a thousand.
>>>
>>>That sounds like a biased sample. People who die within six months are
>>>people for whom treatment has failed. If medical treatment works, they
>>>won't show up in the sample. Am I missing something?
>>
>>It is a random sample of all Medicare patients. It looks at 5000 hospital
>>regions in the country, and predicts total mortality in each region from
>>a long list of features, one of which is how much is spent there in the
>>last six months of life.
>
>I wouldn't expect survival from a specific cause to be related to
>last-six-months spending; treated survivors are excluded while patients with
>other disease are included. ... My concern with excluding survivors remains.
This study looks at total mortality in each region, and everyone in a
region eventually dies. Everyone also has some spending level in their
last six months. I don't see a selection bias here.
>Statistical significance is a real problem with medical studies;
>some require hundreds of thousands of patients.
The figure I gave you, of less than one part in a thousand mortality
benefit, was a 95% confidence level figure. 5 million people is enough
to see real effects. Consider also that people have seen effects
of fat, excersize, social status, etc. with far less data.
Robin Hanson
hanson@econ.berkeley.edu http://hanson.berkeley.edu/
RWJF Health Policy Scholar FAX: 510-643-8614
140 Warren Hall, UC Berkeley, CA 94720-7360 510-643-1884
after 8/99: Assist. Prof. Economics, George Mason Univ.
This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 15:03:29 MST