Case Control Studies

Case-control studies
The simplest and most commonly used analytical strategy in epidemiology involves the case-control study. It is designed primarily to establish the causes of diseases by investigating associations between exposure to a risk factor and the occurrence of disease. The design is relatively simple, except that it is backward-looking (retrospective) based on the exposure histories of cases and controls. With this type of study, one investigates an association by contrasting the exposure of a series of cases of the specified disease with the exposure pattern of carefully selected control groups free from that particular disease (Figure 2.1). Data are analysed to determine whether exposure was different for cases and for controls. The risk factor is something that happened or began in the past, presumably before disease onset, e.g. smoking, or a previous infection or medication. Information about the exposure is obtained by taking a history and/or from records.
Occasionally, the suspected factor or attribute is a permanent one, such as blood group, which can be ascertained by clinical or laboratory investigation. A higher frequency of the attribute or risk factor among cases than among controls is indicative of its association with the disease/condition – an association that may be of etiological significance. In other words, if a greater proportion of cases than controls give a history of exposure, or have records or indications of exposure in the past, the factor or attribute can be suspected of being a causative factor.
Selection of cases
What constitutes a case in the study should be clearly defined with regard to the histological type and other specific characteristics of the disease, such as date of diagnosis, geographical location, etc. Cases that do not fit these criteria should be excluded from the study. This design is particularly efficient for rare diseases, because all cases that fit the study criteria in a particular setting within a specific period are usually included. This allows for a reasonable number of cases to be included in the study without waiting for the occurrence of new cases of the disease, which might take a long time. For reasons of convenience and completeness of case records, the cases identified for case-control studies are often those from a hospital setting, from physicians’ private practices, or from disease registries. Newly diagnosed cases within a specific period (incident cases) are preferred to prevalent cases, since such a choice may eliminate the possibility that long-term survivors of a disease were exposed to the investigated risk factor after the onset of the disease. The selection of cases should be such that the study results are reliable and valid. For these reasons, the following guidelines should be used when selecting cases in a case-control study:
a. The criteria for inclusion in the study (what constitutes a case) and criteria for exclusion from the study must be clearly specified; this will improve the validity of the results;
b. The sources of cases may be:
• all cases admitted to or discharged from a hospital, clinic, or private practice within a specified period;
• all cases reported or diagnosed during a survey or surveillance programme within a specified period;
• incident or newly diagnosed cases;
• incident cases in an ongoing cohort study or in an occupational cohort (sometimes called a nested casecontrol study);
• case units with a prescribed health outcome;
c. If the number of cases is too large, a probability sample may be used;
d. Cases selected for the study should be representative of all cases of the disease under consideration.
Selection of controls
It is crucial to set up one or more control groups of people who do not have the specified disease or condition in order to obtain estimates of the frequency of the attribute or risk factor for comparison with its frequency among cases. This is the most important aspect of the case-control study, as biases in the selection of controls may invalidate the study results, and bias in the selection of controls is often the greatest cause for concern when analysing data from case-control studies.
a. The sources of comparison groups may be:
• a probability sample of a defined population, if the cases are drawn from that defined population;
• a sample of patients admitted to, or attending the same institution as the cases;
• a sample of relatives or associates of the cases (neighbourhood controls);
• a group of persons selected from the same source population as the cases, and matched with the cases for potentially confounding variables;
• on other risk factors (other than the one under consideration);
b. The selection of controls may involve matching on other risk factors:
• Matching means that controls are selected such that cases and controls have the same (or very similar) characteristics other than the disease and the risk factor being investigated. The characteristics are those that would confound the effect of the putative risk factor, i.e. these characteristics are known to have an association with the disease, and may be associated with the risk factor being studied. The purpose of the matching is to ensure comparability of these characteristics for the two groups, so that any observed association between the putative risk factor and the disease is not affected by differential distribution of these other characteristics. It is common to match for age, sex, race and socioeconomic status in case-control studies on diseases, as we know all of these factors affect the incidence of most of the diseases. Matching may be done on an individual basis (one-to-one matching) or on a group basis (frequency matching). Individual matching is preferable, because of the ease of analysis accounting for matching. The disadvantages of matching include a loss of precision and overmatching. Also, once a matched design is used, the matching variable is eliminated from consideration, and therefore it cannot be investigated for etiological association with the disease. For example, if we matched for marital status in a study of breast cancer, we would not know whether single or married women had different risks for breast cancer. Many epidemiologists prefer to conduct studies without matching, and use statistical methods to adjust for possible confounding during analysis, because of the increased precision and the ability to investigate any possible interaction effects. The use of unmatched controls, obtained through random sampling, allows greater flexibility in studying various interactions. What is most important is that information on potential confounding factors should be collected in the study, so that these can be adjusted in the analysis.
c. The number of control groups may vary. It is sometimes desirable to have more than one control group, representing a variety of disease conditions other than that under study and/or non-hospitalized groups. Use of multiple controls confers three advantages:
• If the frequency of the attribute or risk factor does not differ from one control group to another, but is consistently lower than that among the cases, this increases the internal consistency of the association;
• If a control group is taken of patients with another disease, which is independently associated with the risk factor, the difference in the frequency of the factor between cases and controls may well be masked. In such a case, the use of another control group will save the research project;
• Multiple controls provide a check on bias.
The impact of poorly chosen controls on the conclusions of a case-control study is commonly exemplified by Pearl’s study in 1929. Pearl compared 816 malignancies identified among 7500 autopsied cases at the Johns Hopkins Hospital in Baltimore, Maryland, USA with 816 non-malignant autopsied cases matched at death for age, sex, race and date of death. Lesions of active tuberculosis were found in 6.6% of cases and in 16.3% of controls, which led to the conclusion that there was antagonism between tuberculosis and cancer. This finding could not be corroborated in animal experiments. One explanation for Pearl’s findings is that his control group inadvertently included many individuals who had died of tuberculosis, because tuberculosis patients were more frequently autopsied than were patients with other causes of death, and were thus unrepresentative of the general population of deaths.
Collection of data on exposure and other factors
Often data are collected through interviews, questionnaires and/ or examination of records. Occasionally, clinical and laboratory examinations are carried out, but often this is not possible, especially if the ‘cases’ include past cases which may also include some deaths. The following precautions should be taken when deciding on the datacollection strategy:
• observation should be objective, or, if obtained by survey methods, well standardized;
• the investigator or interviewer should not know whether a subject is in the case or control group (blinding);
• the same procedures, e.g. interview and setting, should be used for all groups.

Multifactorial case-control studies
The common form of case-control study addresses one main factor or attribute at a time. It is possible, however, to investigate several exposure factors in the same study. For example, in a study in three states in the USA with a population of 13 million, all mothers of leukaemic children of 1-4 years old (diagnosed in 1959-67) were interviewed. As controls, a sample of 13 000 other women was taken. Four factors were considered, two preconceptional (preconceptional radiation and previous reproductive wastage) and two post-conceptional (in utero irradiation and viral infection during pregnancy). Analysis showed that each factor was related to leukaemia in their children (Gibson et al.,1968). Further analysis was conducted for combinations of factors, where the estimated relative risk in the absence of any of the four factors was made equal to 1.0, It is apparent that the effect was the greatest among women with all four factors, and that there is synergism between the factors.
Advantages of case-control studies
The following are examples of the advantages of case-control studies:
• feasible when the disease being studied occurs only rarely, e.g. cancer of a specific organ;
• relatively efficient, requiring a smaller sample than a cohort study;
• little problem with attrition, as when follow-up requires periodic investigations and some subjects refuse to continue to cooperate;
• sometimes they are the earliest practical observational strategy for determining an association (e.g. use of diethylstilbesterol and clear-cell adenocarcinoma of the vagina in daughters).
Enhancement of the validity of case-control studies
Ways in which one can increase the validity of a study include ensuring that:
• the cases are representative of all cases in a particular setting;
• the controls are similar to cases with respect to risk factors other than the study factor;
• multiple controls are used with consistent results;
• cases and controls are truly selected independently of exposure status;
• the sources of bias are mitigated, or at least shown not to have affected the results. (A common example is the British study of smoking and lung cancer by Doll and Hill (1952). After the cases and controls had been interviewed, it was discovered that some of the cases had been wrongly diagnosed as cancer. Reanalysis showed the persistence of the association and indicated that, in the study, the fact of being told that they had lung cancer did not bias the respondents with regard to the history they gave of smoking);
• repeated studies in different settings and by different investigators confirm each other (for example, the association between smoking and lung cancer has been reported by over 25 investigators from ten countries);
• it is possible to demonstrate a dose-response or gradient relationship (for example, several case-control studies showed that the number of cigarettes smoked per day was related to the risk of lung cancer);
• a hybrid design of case-control study nested in a cohort study with a defined population is used; this is a most powerful strategy.
Disadvantages and biases of case-control studies
The following are some of the problems associated with casecontrol studies:
• the absence of epidemiological denominators (population at risk) makes the calculation of incidence rates, and hence of attributable risks, impossible;
• temporality is a serious problem in many case-control studies where it is not possible to determine whether the attribute led to the disease/condition, or vice versa;
• there is a great risk of bias in the selection of cases and controls. This is particularly serious when a single control group is related to the risk factor under investigation;
• it may be very difficult or impossible to obtain information on exposure if the recall period is long;
• selective survival, which operates in case-control studies, may bias the comparison; there is no way of ascertaining whether the exposure was the same for those who died and those who survived;
• because most case-control studies are performed in hospitals, they are liable to Berkson’s fallacy, or the effect of differing admission policies and rates;
• measurement bias may exist, including selective recall and misclassification (putting cases in the control group, or vice versa); there is also the possibility of the Hawthorne effect: with repeated interviews, respondents may be influenced by being under study;
• case-control studies are incapable of disclosing other conditions related to the risk factor: for example, in a study of the side-effects of oral contraceptives, one has to know their side-effects before a case-control design can be set up.