cancer risk in general population

August 22, 2008 12:45 PM

If you take 10,000 random 40 year old US citizens right now, how many of them will be dead from cancer 20 years from now? I don't care about risk factors - the people are just selected randomly. Is there any way to get this number? Thanks.
posted by crapples to health & fitness (8 answers total)

The data and modelling software's freely available here. Looking here [PDF], the risk of a 40 year old being diagnosed in the following 20 years is 8.70%, and the risk of dying eventually is 21.81%. You could presumably get the precise number for dead-in-20-years from the dataset. The big jump in diagnoses (and fatality) comes between 60 and 70.
posted by holgate at 12:58 PM on August 22, 2008 [1 favorite]

What you want is age-specific cumulative death rates. The exact figure you want isn't published, but this PDF provides a way to sort of back-of-the-envelope estimate it.

So let's look at the 2005 age-adjusted rates by age: age 35-44 is 32.8, age 45-54 is 118.3, and age 55-64 is 329.7. Adding them up, you get 480.8 deaths per 100,000; dividing by 10 gives you 48.08 deaths per 10,000 people.

There are a couple problems with this analysis. a) You want future data, not 2005 data. b) You want to know about one cohort of current 40 year olds, not 3 separate cohorts each of whose members' birthdays span a decade. c) Current 40 year olds likely won't have the same number in 2025 as 60 year olds do in 2005; that number will proably be different. Still, for back of the envelope, I'd bet it's pretty close to reality.
posted by ikkyu2 at 1:28 PM on August 22, 2008

Also, there is another problem: the data for the oldest cohort is per 100,000 people who lived long enough to enter the cohort, which means that each of the 3 cohorts I summed over has a slightly different numerator. Only serious statistic geeks would care about the small error this introduces, though; the reason I am pointing it out is because this error can be corrected for and I am too lazy to correct for it in a back-of-the-envelope calculation.

Actually just ignore this post.
posted by ikkyu2 at 1:33 PM on August 22, 2008

Eyeballing that data, a really naive thing you could do would be to take the existing trends and extrapolate them further.

Just noodling, if you assume that the 1978-2005 trends continue (which is a silly thing to assume), adjust the endpoints of each age bracket for cumulative risk reduction and take the average (which is also silly), then the "predicted" number of people dead from cancer in 2028 is 38.1 out of your 10000 people who are now 40.

More realistically, we can't know this very well. Maybe some currently kill-you-always-and-fast cancers will be curable by 2028. Maybe they'll get cryogenics on line by 2028, so the risk of death by cancer approaches zero: if we can't fix it, we freeze you before you're too fucked up.

On the other hand, maybe we have a global thermonuclear war before 2028 and the cancer rates skyrocket among the miserable survivors.
posted by ROU_Xenophobe at 2:14 PM on August 22, 2008

I think you can get pretty close to what you're looking for here, at the National Cancer Institute. Playing around with the parameters, I get:

Selections:
Statistic Type = Probability of Dying of Cancer;
Year = 2003-2005;
Race = All Races;
Sex = Both Sexes;
Site = All In Situ and Malignant Sites;
Starting Age = 40;
Ending Age = 60;

Results:
2.54%


So, out of 10,000 randomly selected 40-year-olds, you'd expect 254 to be dead of cancer by the time they're 60. (I know nothing about the reliability or quality of this data set, nor of the programming, so it might be total crap. But it looks pretty similar to what you're looking for, so I'll leave it to you to poke around the site and determine if it's good enough for your purposes.)
posted by iminurmefi at 3:23 PM on August 22, 2008

No, that's not right, iminurmefi. You want the ending age to be 40, not 60, and you are not interested in the probability of dying of cancer between 2003 and 2005, but instead lifetime.

Unfortunately, that query's not accepted.
posted by ikkyu2 at 4:17 PM on August 22, 2008

No, you don't want 'lifetime'. You would want Year = 2003-2023, more or less. 2008-2028 would be best of course, but you can't really take technological changes and such into account. So probably the most reasonable number that you can get to filter out technological changes (i.e. 'at current technology and treatment methods') would be looking at what % of people at each age die of cancer and then figuring out how to sum that over 20 years with accounting for the changing denominator.
posted by Lady Li at 12:14 AM on August 24, 2008

Right, Lady Li, that's where I got 48 per 10,000 above.
posted by ikkyu2 at 12:03 PM on August 24, 2008