Re: Deaths from fossil fuel burning air pollution

[Bernard L Cohen <blc+@pitt.edu>]

On Wed, 22 Nov 2000, Otto G. Raabe wrote:

> At 11:58 AM 11/22/00 -0600, Prof. Cohen wrote:
> >	--In the case of radiation, it is easy to imagine gross errors;
> >for example, there could be a threshold at 20 rem, or even a negative
> >slope below 20 rem. This would still be consistent with the RERF data.
> >	 But for air pollution, it is not credible that the
> >dose-response curve wandering through 151 data points can be so strange
> >that a linear interpolation within the region would give a gross error.
> ***********************************************
> I don't disagree with your statements as given, but there is large
> uncertainty and variability in the air pollution data and the actual levels
> of exposure for the affected people. Do you really think that these data
> can be used to estimate an increase in mortality risk if the concentration
> of airborne particulate matter goes from 30 micrograms per cubic meter to
> 32 micrograms per cubic meter? Is it possible that there is actually no
> increase over this range?
> Do you really think these regression fits are that good?

	Dear Otto:
	If the data show a credible consistent trend of increasing from 20
to 30 to 40 to 50 micrograms per square cm, that would indicate to me that
the dose-response curve is increasing steadily in that region and I would
be surprised if it were not to increase (proportionately less) from 30 to
32 micrograms per square cm. That would imply that it is a step function,
increasing from 20 to 30, staying level from 30 to 32, and then increasing
again from 32 to 40 to 50. It would be difficult (but not impossible) to
explain such behavior, and there is no reason to expect it here.
	Accuracy is another question, but regression analysis gives a
standard deviation for the slope and in most cases (including the data of
Pope et al), it is small enough to be unimportant. There are other
potential problems with regression analysis - books have been written
about them - so nothing is certain, but the increase in response from 30
to 32 in the above example seems plausible to me.
	In the case of the RERF data on radiation induced cancer, the data
can be well fit with a threshold at 20 rem, and there is surely no
difficulty in explaining such behavior. 

************************************************************************
The RADSAFE Frequently Asked Questions list, archives and subscription
information can be accessed at http://www.ehs.uiuc.edu/~rad/radsafe.html