Thursday, January 05, 2006

BoE Risk Assessment - bird 'flu

Effect Measure on human bird flu in Turkey

I blog a bit, somewhat excessively probably, on the bird 'flu; but such is life - have "Plagues and Peoples" by McNeill and "Plague's Progress" by Karlen neatly together on my "at hand, but not work related" shelf in the office.

It is interesting, because the influenza viral family is endemic in humans, has an active animal reservoir, is well characterised and is changing (probably driven by recombination more than mutation, though journalists will always talk about "mutation").

So with the H5N1 variant, we can see it change and become more probable to be human-to-human contagious.
It is currently highly lethal, though how so is poorly constrained - the clinical cases spotted have 30-60% mortality, but those are people who got acutely sick and rushed to high tech medical facilities where testing could be done. The incidence of weak or asymptomatic cases is essentially unknown. So true mortality could be much lower.

So, it is likely, but not certain, that H5N1 will go human-to-human in the next few years.
Most probable scenario is an "error" during replication in a victim infected with both H5N1 and a "regular" 'flu virus, increasing the new hybridised H5N1s ability to directly infect humans. Likely such a change would change its clinical effects, including mortality rate, since it is already highly lethal, it could well become (much) less lethal. Virii are not well served by being excessivel lethal; but evolution is not purposeful, H5N1 could well do a side-branch that is not optimal in the long run, for us or them.

So, how bad could it be?
Well, I'm a theorist, dammit, so we will estimate, using our best Ansatzes and Back-of-the-Envelope guesstimates, driven by crude historical analogy and order of magnitude arithmetic.
One of these exercises that is essentially wrong in every detail, but will still likely give a "right" answer!

Major epidemics happen about once per century, with mortality rates of order 1%.
In recent historical times, there have one-to-few epidemics per millennia with more like 10% mortality rates...
On the other hand medical competence is now high and care of patients and interruption of transmission much better.
On the third hand, with even 1% incidence most modern medical systems will break down through overload and attrition and a large fraction of patients will receive no care, or ignorant amateur care in the home.

So, say in the next decade we have 50% probability of a pandemic (this is Bayesian - it will either happen or not; sophists will object to the prior choice of time frame; I say phbt! it is my theory).
We have 10% probability of 1% mortality and 1% probability of 10% mortality! Handy that.

So expected casualties: 0.5*(0.1*0.01 + 0.01*0.1 + 0*0)*Pop = Population/1000
(I am assuming that (1 - 0.1 - 0.01) probability of "negligible" deaths, which is actually the highest time weighted risk since regular 'flu is a killer and is endemic.

So you can, incorrectly, interpret that in several ways:

Expected worldwide number of deaths from pandemic in next decade ~ 6 million
Expected deaths in US from pandemic ~ 300,000
Your personal probability of dying from a pandemic ~ 1/1000 in next decade or ~ 10^-4 per year.

The US mortality rate is 847.3/100,000 or about 1/120 for a random person, per year.
So epidemic mortality risk is ~ 1% of your integrated mortality risk over the next decade (yeah, you can age adjust that, but who cares about details).

Interestingly, looking at the CDC mortality tables this estimate is a bit higher current risk of death from AIDS in the US, which is an ongoing pandemic, albeit a slow one, and about 1/2 of the endemic risk of dying from influenza or pneumonia (although that is very age weighted).

So, estimate is probably reasonable and sane.

For comparison, accidental death rate is almost several times higher and death rate by firearms about the same. Death rate from car accidents is almost twice as large.

So, why worry?
As always, because of the low probability, high risk scenario.
Taken at face value, there is a ~ 1/00 risk of a pandemic killing ~ 1 billion people worldwide and tens of millions in the US; this no higher than the "mean risk" of a pandemic of ~ 1/1000 per year, even though I take the prior that we know of a probable impending pandemic with H5N1. So I'd call it a conservative estimate, with the true risk maybe twice as large!
Of course it is the virus we don't know that will probably get us...

What is the insurance strategy?
If you take actuarial risk, then the US has a ~ 1/100 risk of losing > 10 trillion dollars in economic actitivy over a decade; so a rational abatement strategy would be to spend < $50 billion over a decade to avert this scenario.
It would be rational for the US to devote O($ 1 billion) per year to head off a 'flu pandemic, or have means to sharply reduce its impact.

That's in the range of things it is rational to worry about.
'Flu Wiki or related sites offer very low cost for individual risk abatement, effective at the 30-90% level of lowering risk for an individual or family, maybe.
Worth a little bit of time and effort, but not too much. Rationally.

Final, horrid thought - it is 'flu season; what happens if H5N1 gets into an immunocompromised population?


Anonymous Colin Purrington said...

I've done some related, but terribly inaccurate, calculations. If you're interested, here they are:

5:19 PM  

Post a Comment

<< Home