In this type of study, subjects who have the disease of interest (cases) are compared with subjects who do not have the disease but are otherwise similar to the cases (controls). Cases should be representative of all those with the disease, excluding those too ill to participate. The criteria for diagnosis needs to be defined. Cases can then be identified, say all those passing through a particular clinic. Choosing suitable controls is more problematic. Controls could be selected either from hospital patients or from the community. If hospital patients are chosen as controls, it is important to select from a wide range of conditions. This reduces the chance of detecting a difference between cases and controls that is merely an artefact of a condition suffered by the controls. If community controls are chosen, these should be of similar age, sex and geographical area to the cases. One way to achieve this would be, for each case, to select from their GP's records the next person alphabetically who is of the same age and sex, but without the disease.

Case-control studies have a number of advantages. They are easy to carry out and economical in terms of the resources required. Sometimes they are the only way of investigating a particular hypothesis, especially if the disease is rare. However, they also have several disadvantages. Trying to obtain information about the past means there is a problem with accuracy and completeness of information. Memory may be biased by subsequent development of disease. Also, the study may not distinguish between the causal factor and other factors associated with it. Some of these weaknesses can be partially overcome by replicating the study in different areas and in different groups of patients - it is unlikely that the same biases will occur every time.

Past diet is particularly difficult to assess; memory of this tends to be strongly influenced by current diet. Diet may have altered after the onset of the illness, either because appetite has been affected by the disease or because of dietary advice. One way around this problem is to select as cases subjects who have evidence of the disease but no symptoms. For example, in a study of coronary heart disease, subjects with evidence of ischaemia on ECG (electrocardiograph), but no history of myocardial infarction or angina, could be selected as cases.

Ideally the interviewer should not know which subjects are cases and which are controls, as this may introduce bias to the study. This may not always be possible, e.g. if community controls are used they are likely to be at home, whereas the cases may be in hospital. One way around this problem is to delay interviewing the cases until they have been discharged from hospital, if possible. To reduce the potential for interviewer bias, the interviewer should be unaware of the hypothesis being tested. Alternatively, self-administered questionnaires could be used.

Controls are likely to be less compliant than the cases, since the study has no direct relevance to them. It is important that both cases and controls should be pursued equally, as differing response rates are a source of bias.

Another source of bias is that the cases may be aware of various theories as to the cause of their illness. In a case-control study of infants with eczema, for example, mothers of the cases will try harder to recall any adverse reactions to foods in their offspring when solids were introduced. To avoid drawing attention to the foodstuffs of particular interest, it is advisable to ask questions about a wide range of foodstuffs.