[maxander]

Happily, these kinds of studies aren't attempting to create a comprehensive statistical model of the human body, in any reasonable sense- they're just asking, "if we change X in modern human lifestyle, and leave every other factor within the range of typical variation, how does it change Y?" Thus, in almost every case, the millions of other factors influencing human wellbeing can be completely ignored for the purposes of the study, as long as they're reasonably well represented in the study population.

In terms of caffeine studies, yes- there are definitely thousands of other foods and so-forth that effect the risk of heart disease, cancer, and whatever else they're looking at. There's even things which interact with caffeine specifically to modulate or plausibly even reverse it's effects. All of these things work in egregiously complicated ways. But when I read the study and try to think what it tells me about my own situation, this doesn't matter since I can usually assume that my own exposure to these other factors will be typical compared to that of the study population- thus, I'll tend to respond similarly to caffeine as they did.

This approach is obviously has an abundance of limitations- it only tells me how I'm more likely to respond to caffeine or whatever else, but doesn't give any guarantees. And for someone who is different than the study population- a black woman living in Uganda versus a study done on white male freshmen at Yale, say- the results rapidly become less meaningful. But they do work.