Aim

Obtaining a representative sample appropriate to the research question.

Description

Epidemiological research implies conducting measurements on people and deriving estimates from this. All kinds of errors prior, during and after the study can be responsible for bias in the final results. This bias can be caused by measurement errors, as well as through poorly chosen source and study populations. Bias can also be introduced during the sampling procedure (for instance through selection bias).
Given that epidemiological research is generally carried out with a limited number of people/patients, the generalisability of the results could be limited by this type of bias. A good sample from a well-described study population is therefore of utmost importance.

Three steps need to be completed in specifying a study population.

The delineation of the domain is the first step. The description of the study domain comprises which individuals can participate in the study and therefore also which group of people the study results will apply to at a later stage. Specification of the domain and the research questions are inextricably linked together. For instance: For a study into the effect of a treatment on the prognosis of a myocardial infarction, the domain could consist of men older than 55 admitted due to a myocardial infarction. The research question and study results will only be of relevance to the group of male patients older than 55.

The second step is the identification of the source population. The source population comprises individuals who, on the basis of the previously defined domain, could be involved in the study. In the above example the source population could consist of the patient population of a hospital. The study population now consists of male patients who are older than 55 and admitted with a myocardial infarction.

Finally, the third step is to draw a sample from this study population. The observations carried out on the sample will provide information to answer the research questions.
The method of sampling depends on the type of research. For instance, the entire study population will be selected in a cohort study. In other cases a random sample will be drawn from the study population. In a patient-control study both the patients and controls need to be selected from the same source population.

The sample size is based on power calculations, which should take into account the expected non-response and dropouts.

N.b.: The data are collected in a totally different way in qualitative research. Refer to the guide on qualitative research.

Once the study population has been established, individuals from the sample will be invited to participate in the research. An important principle in medical research is that participation should always be voluntary and that potential respondents should only be “bothered” a limited number of times with invitations to participate. In terms of the subsequent analysis the number of refusals and reasons for refusal should be recorded. This “non-response” is always required (in full) for every study. It is advisable to draw up a plan before recruiting trial participants for collecting non-response data (which data are required (at least), how to approach non-responders, etc.).

V1.1:  1 Jan 2010: Translation into English.
V1.0:  21 april 2004.

• Have the domain, source population and study population been documented?
• Have the choices made in terms of arriving at the final study population been documented?
• How was sampling undertaken?
• How many people refused to participate? Have the reasons for refusal been recorded?