A case-control study may be used to investigate a problem related to a cause of disease. Patients with a particular condition (cases) are compared with an identical group of individuals who do not have the condition (controls). Both groups should be identically matched except for the condition under study. Case-control studies are generally retrospective, and accounts of past history and exposure are investigated to ascertain the common lifetime exposures, linking these to possible causation of disease. The benefits of this form of research are that the researcher can study multiple exposures and diseases that have long latent periods, e.g. breast cancer may recur some 15-20 years later, hence the benefits of breast screening should be evaluated over this time. It is also useful in situations where it would be unethical to carry out an experimental study and where little is known about the cause of the disease. The evidence gained from this sort of research generally is considered lower down the ladder of hierarchical research compared with the RCT. There is potential to overlook influencing factors that may have been outside the realms of the study, and thus the results may be less reliable. There may also be difficulties with selection bias, particularly as the researcher must attempt to rule out other possible threats to the validity of the findings, i.e. other exposures. Information bias may also be a potential problem because exposure status is determined after the outcome has occurred and therefore participants' recall of exposure may be blurred.
Example of case-control (Draper et al., 1997)
To test the hypothesis that 'childhood leukaemia and non-Hodgkin's lymphoma can be caused by fathers' exposure to ionising radiation before the conception of the child, and more generally, to investigate whether such radiation exposure of either patient is a cause of childhood cancer'.
The case group (35949 children diagnosed with cancer) were compared with the control group (a group of individuals selected from the birth register for the same area of birth, matched on sex and born within 6 months of the case).
These included (a) parental employment as radiation worker before conception of child, (b) cumulative dose of external ionizing radiation for various of periods of employment before conception of child, and (c) dose during pregnancy.
Although the researchers confirmed that the fathers of the children with leukaemia or non-Hodgkin's lymphoma were significantly more likely than fathers of controls to have been radiation workers, they concluded that this did not relate to a preconceived radiation dose. As such the absence of a relationship between dose and risk led the researchers to believe that these findings may be related to either chance or perhaps some characteristic other than exposure to radiation.
Cohort studies can be prospective or retrospective (historical), depending on the time that the exposure data were measured. The essential feature of all cohort studies is that the exposure is measured before outcome (Greenhalgh, 2001). In a prospective cohort study, the researcher begins with individuals who have not yet had the outcome of interest (e.g. bowel cancer) and follow the group forward in time measuring exposure (e.g. exposure of interest may be diet) to see if the outcome of interest occurs (i.e. bowel cancer diagnosis). In a retrospective cohort study exposure is measured using data collected before the study started (i.e. looking back through case notes at the main exposures during the study period), although recall bias may be an issue with retrospective data. In addition to bias, Brennan and Croft (1994) highlight the importance of confounding which must be considered in any cohort study. Unlike the RCT, in which participants are randomly allocated to either the experimental or the control arm, participants of cohort studies have chosen to be exposed or not and subsequently this in itself may affect the outcomes of both groups. 'As a consequence, if a confounder is not recognised and adjustments made for its effect the exposed and unexposed groups in such studies will not be comparable' (Brennan and Croft, 1994). Figure 11.2 shows a cohort study design.
Figure 11.2 Cohort study design.
Example of a cohort study (Graham et al., 2002)
To confirm the relationship between severely stressful life experiences and relapse of breast cancer found in a previous case-control study.
This was a prospective study recruiting a cohort of women newly diagnosed with breast cancer. The researchers controlled the group for a biological prognostic factor (i.e. other factors that might affect recurrence rates such as lymph node status and tumour grade).
Figure 11.2 Cohort study design.
Outcome measure Recurrence of disease. Data collection
Women were interviewed every 18 months over a period of 5 years, collecting data on stressful experiences and depression (including data on experiences 12 months before diagnosis).
Overall the researchers found no increased risk of recurrence in women who had one or more severely stressful life experiences in the year before diagnosis compared with women who did not, and of those who had stressful experiences since diagnosis the results demonstrated a lower risk of recurrence, hence confirming that stressful events did not lead to increased risk of recurrence.
The interesting thing about this study is that it is a good example of how different designs may affect research outcome. As the researchers themselves point out, their findings differ from those in an earlier study (Ramirez et al., 1989) which used case-controlled methods. The researchers suggest that this may relate to difficulties, in retrospective studies, in recalling stressful experiences and the possibility that patients look back for something to blame for the development of recurrence, so a prospective study enables greater recall and accuracy of data.
Was this article helpful?