Is it ok to use volunteers for prospective cohort studies

Question

Prospective cohorts are ones where the exposure is known but not the disease outcome.

Volunteer bias is where volunteers tend to be healthier.

I've been told because of this volunteer bias, we should not use volunteers for our epidemiology experiments.

I can see why this is true in the case of prevalence studies. Volunteers tend to smoke less and thus we will underestimate the prevalence of smoking if we only get volunteers.

But I don't see why this must be true for prospective cohort studies.

Let's say there's been a nuclear accident and we rush to attract volunteers from those who've been exposed to radiation. All my volunteers, exposed and unexposed, are very fit and don't smoke.

Their general better health could mean they recover better from radiation. One could say this skews results. However, given that my other unexposed volunteers are equally healthy - wouldn't this be a fair comparison? Is there still volunteer bias here?

Answer 1

First and foremost we must be clear, that experiments and studies are never perfect and bias and skewed results are the norm, not the exception. There is just so many things you simply cannot blind people for. If you study high-dose cortisone against placebo, those with the red faces, the high blood pressure and the sleeping problems are the cortinsone group. Thus a sentence like

we should not use volunteers for our epidemiology experiments.

should be carefully worded and understood more as "if you have better alternatives avoid volunteer groups and if you need to investigate volunteers consider possible influences on the result carefully".

So maybe in your nuclear plant you have so many volunteers that are very eager to be open about any body complaints and report them. Maybe you ask yes/no-questions like "did you have headaches or muscle complaints within the last 5 days" and all your body-complaint-centered volunteers report to have had headaches or muscle complaints - you will hit a ceiling effect in that question and will not find how exposure to radiation lowers the frequency and intensity of headaches. Maybe all those health-centered volunteers do not take illegal drugs or do not come to work tired. You will not find any radiation effect that affects only those taking that modern illegal drug or the accidents that only happen if a worker is tired and radiated.

I know these are not very convincing examples and appear to be taken out of thin air -- that is because they are. However, they are not totally impossible and thus you will have to think (and reason) about possible biases if you investigate volunteers, but you will have to do that with any study group you choose and you better start that in the planning of a study then in the final analysis.

Answer 2

Your question:

“However, given that my other unexposed volunteers are equally healthy

• wouldn't this be a fair comparison? Is there still volunteer bias here?”

The comparison of exposure versus non-exposure among the volunteers yields an unbiased estimate of the effect of exposure among people who are like the volunteers (internal validity) but the findings may not be generalizable to people who not like the volunteers--lacks external validity. For example, if volunteers are more healthy than the general population (they usually are) and the cohort study comparing exposed with non-exposed among volunteers finds no effect of exposure, one cannot conclude that there is no effect of exposure in the general (less healthy) population—there is (potential) bias.

EXPLANATION

An epidemiologic cohort study is a non-experimental (observational) study design which assesses the association between one or more exposures and the development of (or mortality due to) one or more diseases. The goal is to draw causal inferences about the effect of exposure and the development of (or mortality from) the disease.

As explained in this excellent 2010 review article by Song and Chung (and elsewhere):

“The hallmark of a cohort study is defining the selected group of subjects by exposure status at the start of the investigation. A critical characteristic of subject selection is to have both the exposed and unexposed groups be selected from the same source population.” [bolded for emphasis]

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2998589/

Song JW, Chung KC. Observational studies: cohort and case-control studies. Plast Reconstr Surg. 2010;126:2234-2242. doi:10.1097/PRS.0b013e3181f44abc

People in cohort studies are seldom completely representative samples of the population for many reasons. It is difficult to identify representative samples. Some people identified as part of a representative sample don’t want to be in the cohort study. Thus, in practice, people in a cohort study where the study interacts with the people in the study are always volunteers.

NOTE: Some cohort studies involve only computer record-linkage and this can be done (with approval by an Institutional Review Board) without interaction of the study with the people whose records are used in the study. Thus, there is no issue related to volunteerism.

Famous cohort studies such as the British Doctor’s study, the Framingham Study, and three iterations of Nurses’ Health Studies all enrolled only volunteers.

These links are to articles about the history of these landmark cohort studies and how the people in the study entered the study as volunteers.

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5298160/

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4159698/

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4981810/

There are numerous other examples.

Use of a non-representative sample, for example, volunteers, in a cohort study affects the external generalizability (also called the external validity) of the results. For example, nurses who volunteered to be in the Nurses’ Health (cohort) study cohort were preponderantly white and their educational attainment was higher than that of all women in the United States. The findings of the study about the association of an exposure with a disease association might not be generalizable (applicable) to non-white women or women with educational attainment different from nurses.

See Bao et al. (cited above) for discussion of generalizability in the Nurses' Health Study I.

An important source of bias in cohort studies arises because of loss-to-follow-up where there is differential loss-to-follow-up comparing the exposed and unexposed.
See this classic (pay-walled) paper.

Greenland S. Response and follow-up bias in cohort studies. Am J Epidemiol. 1977 Sep;106:184-7. doi: 10.1093/oxfordjournals.aje.a112451

A benefit of enrolling volunteers in a cohort study is that the volunteers are (presumably) interested in the study and thus are less likely to be lost-to-follow-up.