"following tens of thousands of people over 30 years"
That's called a cohort, it's a well studied method in statistics and depending on the size of the sample and the number of studies you can build correlations out of these.
As your sample gets larger the population for the proportion of heavy meat eaters and the rest of your sample tend to normalize in almost every variable[1], that means the median of every other observable variable will be close to the general population's value for both groups, and you can assume that the quantity of meat is the sole variable worthy of analysis.
Mind you that the use of statistics in health and social sciences is because these fields are not physics, we can't find general solutions based on the present state of affairs putting some number on a equation. How many people will get cancer or whatever.
I doubt this was the type of study that are used in this case because a cohort study is really expensive to conduct, generally this type of study is conducted more using the Case-control method which makes than more affordable.
The fact that if you conduct such a study and find that heavy meat eaters get more cancer or heavy coffee drinkers live more doesn't mean that every coffee drinker will live more or that there will be much more heavy meat eaters with cancer than the rest of the population, it will only means that some epidemiological indices are higher in a group than in the other.
If you really wish to test both hypothesis you must do the same study as many times as you can and try to use the median of these indices obtained in the same study for both populations, the median is a robust measure and see if they are too distant numerically one from the other.
Of course, this can still proves nothing and only find that correlations are in fact established. It can simply be that rich people live longer and get more cancer than the rest and both drinking coffee and eating too much meat be associated with income.
[1]: That's not exactly true you can have confounding variables in your data, which sadly sometimes are not included in the data, when they are you could read the methods of controlling this developed by Mantel and Haenszel.
That's called a cohort, it's a well studied method in statistics and depending on the size of the sample and the number of studies you can build correlations out of these.
As your sample gets larger the population for the proportion of heavy meat eaters and the rest of your sample tend to normalize in almost every variable[1], that means the median of every other observable variable will be close to the general population's value for both groups, and you can assume that the quantity of meat is the sole variable worthy of analysis.
Mind you that the use of statistics in health and social sciences is because these fields are not physics, we can't find general solutions based on the present state of affairs putting some number on a equation. How many people will get cancer or whatever.
I doubt this was the type of study that are used in this case because a cohort study is really expensive to conduct, generally this type of study is conducted more using the Case-control method which makes than more affordable.
The fact that if you conduct such a study and find that heavy meat eaters get more cancer or heavy coffee drinkers live more doesn't mean that every coffee drinker will live more or that there will be much more heavy meat eaters with cancer than the rest of the population, it will only means that some epidemiological indices are higher in a group than in the other.
If you really wish to test both hypothesis you must do the same study as many times as you can and try to use the median of these indices obtained in the same study for both populations, the median is a robust measure and see if they are too distant numerically one from the other.
Of course, this can still proves nothing and only find that correlations are in fact established. It can simply be that rich people live longer and get more cancer than the rest and both drinking coffee and eating too much meat be associated with income.
[1]: That's not exactly true you can have confounding variables in your data, which sadly sometimes are not included in the data, when they are you could read the methods of controlling this developed by Mantel and Haenszel.