Controlling for lifestyle factors in epidemiological studies

[Isaacson]

A glance at any official health bodies advice will show that we are entreated to eat more healthily, take more exercise and reduce stress. Presumably, there is some reasonable evidence that these factors play a relatively significant role in disease formation and response (otherwise, one would presume they would not be within the remit of such organisations).

 

In what little opportunity I've had to read primary research on medical testing, however (I'm a statistician, not a biologist), I've not come across any attempt at stratification for these factors.

 

If one is, say, testing the response to a new medicine, presuming these factors will promote recovery in your test subjects, then the apparent efficacy of the medicine might be effected by the distribution of these "lifestyle" factors within your test and control groups.

Similarly, on the other side of the coin, if you are testing the response to a new health threat by epidemiological study, a quoted Odds Ratio of 1.2 might be quite radically different for individuals of different general health, yet (in the studies I've read) there seems to be simply a presumption that the control group and the test group will have a similar spread of lifestyles.

 

It certainly seems to be a relevant factor in some cases as individuals with a history of smoking are almost always controlled for in such epidemiological studies (especially related to lung conditions), but I have never come across similar treatment of individuals in high stress jobs, eating fewer than 5 fruit and veg per day, taking less than three hours exercise a week etc.

 

For large effects, these factors would probably be insignificant, but where the effect is small (say a slightly efficacious new treatment, or a low risk environmental contaminant), failure to take account of these factors could theoretically lead to the conclusion that the drug is a slight improvement on what we have, or the threat requires some moderate mitigation, when if fact the drug is virtually ineffective in people who are otherwise unhealthy, or the threat is actually zero to a healthy individual, which would be really important conclusions for public health.

 

My limited interaction with epidemiological studies was quite a few years ago now, so my question is, have things moved on to start controlling for these factors now we have a better idea of their impact, and more significantly, if they have, have any studies from the past been revised in the light of such controls? Also, does anyone know of any accessible studies on the impact of these factors on epidemiology in general?

 

Edited April 2, 2017 by Isaacson

[Prometheus]

Lifestyle factors are poorly quantified even when they are the primary focus in clinical research: it would be hard to detect more subtle interactions. My guess is that any differences would be drowned out in statistical noise from numerous interactions: genetics, lifestyle, drug-drug, age, sex and a plethora of exposures we haven't even thought of.

 

I'm a little confused regarding your questions about new medicines (IMPs). Before licence they are assessed with various phases of clinical trials, all of which are RCTs. As such lifestyle factors should also be randomised across control and treatment groups. Also the pharmacodynamics and pharmacokinetics are are well tested, especially during early phase trials, and will hopefully detect interesting drug/body interactions. It's only post-release that a medicine would be assessed in terms of population effects rather than effects on individuals. Is that what you mean?

 

Not looked at it for ages but the Framingham heart study might be interest to you, i know they looked at aspirin in some detail - they may have expended to look at other meds.

[Isaacson]

Thanks for the link it's certainly interesting reading along the same lines and it's linked to a whole load of other studies of a similar type. The trouble is, studies on this topic seem to be of two camps; one studying the effects of lifestyle factors on some given condition, and the other studying the effects of some environmental/genetic factor on said disease. What I'm really interested in is the effect of lifestyle factors on the effect of environmental factors on some given disease.

 

Take smoking for example, unless I'm mistaken studies have shown that smoking causes a risk of lung cancer (say 1:100 for the sake of argument). Studies have also shown that certain lifestyle factors (such as obesity, stress and anti-oxidant levels in the diet) have an impact on cancers (presumably including lung cancer). What it would be interesting to know then, is what the result is from putting them together. I.e. what is the risk of lung cancer to an otherwise ideally healthy person who for some crazy reason decides to smoke 20 a day.

Edited April 2, 2017 by Isaacson

[Prometheus]

Something like an interaction term on one of the coefficients in a regression model for instance? I'm not aware of any studies that explicitly look at this. I suspect it's of more interest to statisticians than medics though.

[Isaacson]

Yes, the model would be slightly different because I'd be using Odds Ratios rather than standard variables, but it would not differ substantially from models attempting to use multiple interaction terms. I did consider doing this, but found that all I was modelling was the potential change in disease risk, in the light of certain lifestyle factors without being able to control for things like the fact that a person who was otherwise fit and healthy probably wouldn't smoke 20 a day, lowering the sample size beyond significance.

 

Of more significance was the lack of comparable data (that I have access to anyway) which is mainly why I posed the question. I'd ideally want data on the risk from some environmental disease causing variable (or drug side effect) and the lifestyle factors in cohorts that are reasonably comparable. Unless someone with the budget to design studies has actually deliberately done this, I don't think it's going to happen.

 

As far as interest is concerned though, I'm quite surprised it's not been done. The significance as I see it would be that if there was a correlation, there could be a threshold above which the toxin is not toxic, or the otherwise useful drug's side effects are insignificant. Say for example (and forgive my basic grasp of biology), that a certain toxin had its effect by production of free oxide radicals, and that a good diet improved one's health partly by containing anti-oxidants. It could well be that if the quantity of oxides in a given toxin were below the quantity of anti-oxidants in a good diet, one would not have simply lowered one's chances of getting said disease, one would be immune entirely to it at that dose. Quite a significant difference in statement.

[John Cuthber]

I think the fundamental idea is that, if you have a big enough sample size, the effects of things like lifestyle differences will cancel out.

That's part of the reason for (double) blind trials with a control group.

[Isaacson]

I think the fundamental idea is that, if you have a big enough sample size, the effects of things like lifestyle differences will cancel out.

That's part of the reason for (double) blind trials with a control group.

 

That's rather the point. Any odds ratios worked out by comparing an exposed group to an unexposed control where the sample size was large enough to include a full (and fairly equal) range of "lifestyle" factors, would (if lifestyle factors have a significant effect) be totally inapplicable to anyone but the "average" person and, depending on the size of the effect, could be whole orders of magnitude out.

 

As we already have some very useful drugs which have to carry a "do not use if pregnant" or "not suitable for under 12s" warning because of the specific vulnerabilities of those groups, there could theoretically be medicines which are useful but might carry a "do not use if you eat less than 5 fruit and veg a day", or "do not use if you do less than 20 minutes exercise a day" warning.

 

Similarly, levels of toxins such as benzene in certain work environments might be so well tolerated by those who eat well and exercise regularly that such a requirement of workers might be a demonstrably useful control measure.

[Prometheus]

Certain medicines do come with lifestyle advice. Warfarin, for instance, involves a number of dietary considerations, though these are based on a knowledge of enzymatic interactions with the drug. I'm not aware of any studies looking for such interactions statistically. You could try the search term 'personalised medicine' with other search terms, but i'd be surprised if you found exactly what you are looking for. Let us know if you do.