poly: Longevity

A couple of general comments, inspired by Robin's summary of the
Wilmoth article in Science about longevity.

The approach he's advocating is a cheap way of doing prediction--
he ignores all detail. It looks cost-effective.

If we need an answer enough to invest some work, then it makes
sense to collect the necessary information to do a better estimate.
Wilmoth says "don't bother".

This is a philosophy of ignorance, but the proof of the pudding is
whether extra statistics produce better estimates. (I don't know
the answer, but my prejudice is to favor the dataful approach.)

For example, we've achieved most of the possible gains from
eliminating occupational disease & workplace accidents.
How this factor balances against other causes of mortality isn't
obvious, but it's probably not the same as last year or last decade.

I have two quibbles with the use of "life expectancy" as a measure
of longevity. (a) is that it overweights infant mortality -- my
interest is in the figure for 50 year old male couch potatoes.
(b) is that it treats each year of life as having an independent
risk, so the figures for any given calendar year give undue weight
to health factors for the year: if the flu season is bad, the
LE figure based on the year's stats will include 70x additional
flu risk. Another side of (b) is that it doesn't model one-time
events such as immunization very well. Two centuries ago, if trying
to predict the life-span of a ten-year-old, the most important
piece of information might be "has he had smallpox?".

The other problem with Wilmoth is that it suggests there's no
point in taking any special steps to promote longevity.
We see people telling congress that curing cancer would only add
a year to life expectancy -- why bother?

Another glossed-over point is whether LE is the correct thing to
maximize: Suppose a death in an auto-accident costs 30 years of LE
for an average victim, while a cancer death costs 5?. Then
maximizing LE implies that we should be willing to spend 6x as
much to prevent the auto death. This might be a good social policy,
but it should be brought out in the open, rather than hidden in the
LE statistic.

Rich Schroeppel rcs@cs.arizona.edu
Received on Tue May 5 20:27:19 1998