Case Control Flashcards
Utilisation of case control design
In the past, case-control studies were typically used for chronic diseases (e.g. cancer). Nowadays, they are being used more and more in a wider range of conditions and hypotheses e.g. infectious diseases and injuries and evaluation of screening or vaccines
Incident versus prevalent cases
An important issue to be considered is whether to include prevalent as well as incident cases in the study. Incident cases are new cases appearing within a fixed period of time (e.g. a year) whereas prevalent cases are all those with the disease at one point (or short period) in time. Prevalent cases will include patients who may have had the disease for some time, and may be different in terms of exposure compared to recent incident cases. Inclusion of prevalent cases may also miss people with more severe disease who may die early and whose exposure levels may differ from prevalent cases with mild disease and who are still alive. These problems are reduced by taking incident cases.
An important point to consider is that both prevalent and incident cases may have changed their habits (or “exposures”) because of the disease – either from knowledge of the diagnosis or from symptoms/consequences of the disease. For example, lung cancer cases may have given up smoking because they had developed a cough. Obtaining accurate recall of exposures in the past (e.g. smoking) before the onset of disease is required for both incident and prevalent cases but recall may be more difficult for prevalent cases who might have been diagnosed long ago.
Source of cases
The study might be ‘hospital-based’ and the cases taken from all patients fulfilling the case definition criteria attending a certain hospital, e.g. all cases of stillbirth delivered in the maternity department of Basingstoke District Hospital 2004-8. Alternatively, the study might be ‘population-based’ and cases taken from a defined population over a fixed period of time, e.g. all cases of salmonella food poisoning reported to North East Thames Regional Health Authority in one year. In general, population-based studies are more readily generalizable to the population but are also more logistically difficult to conduct.
It is important to consider whether the cases chosen are representative of all cases in the population.
Selection bias
It is important to consider whether the cases chosen are representative of all cases in the population. There is usually some degree of selection, and the role of selection bias in casecontrol studies may be extremely important. Issues that need to be considered are patient survival, referrals to specialist hospitals and refusals. For example in a study of depression, hospital based cases may be different from those in the community (they may be more severe, have other co-morbidity, or social problems), cases attending general practitioners for help with depressive symptoms may be different from cases of depression who do not seek help. The selection of cases (and controls) must be made independently of their exposure status.
Source of controls
The key point for recruiting control participants is that controls should represent the prevalence of exposures (and confounders) in the same population from which the cases arise. Identifying the controls is usually straightforward when the cases come from a clearly defined geographic locality and controls may then be selected from the same locality. However if hospital-based cases come from a widely dispersed locality, then identifying controls may be more problematic. Many studies use “neighbourhood controls” so that for each case a control is identified from the same local population register, or same general practice age sex register as the case. Hospital-based controls have often been used in case-control studies because they are easy to identify, and these patients are often happy to participate. The major weakness of hospital-based controls is that they may not be representative of the prevalence of exposures in the population. For example, relative to population-based controls, hospitalbased controls are often affected by other health conditions, and are more likely to be smokers or drinkers or have more unhealthy lifestyles. Population-based controls are nearly always preferable. However, it is often challenging to recruit population-based controls. Lower participation rates are associated with increased probability of selection bias, particularly with respect to the level of exposure.
What is the most difficult part of designs a case control trial?
Choice of a suitable control group is the most difficult part of designing a case-control study. Some studies use more than one type of control group, but this can lead to problems: the two control groups may give different results, a situation which may be difficult to interpret, although it may also bring insights into selection biases.
Why would studies choose more than one control per case?
Studies may select more than one control per case as this increases the statistical power to detect a difference in exposure levels between cases and controls. It is common to see 2 or 3 (or more!) controls selected for every 1 case in a case-control study.
Matching
Matching refers to the recruitment procedure whereby controls are selected on the basis of similarity for certain characteristics to each case who was selected into the study. Common matching characteristics are age and sex, but others characteristics might be place of residence, socio-economic status, or parity. The characteristics chosen for matching are those that are thought to be potential confounders.
Confounding
Confounding is the alteration of the disease/exposure relationship brought about by the association of other factors (confounders) with both the disease and the exposure)
Matching is done to increase the statistical power of the study (as you may learn in Statistics courses) but it is essential that matching is restricted to potential confounding factors and is not performed for the exposure under investigation. ‘
What is a possible issue which could arise when attempting to match?
Overmatching’ may also be a problem; if we make the cases and controls too similar to each other, we may mask the association between the exposure and outcome. For example in a study on smoking and lung cancer if we match on alcohol consumption (and in this example alcohol is not a risk factor for lung cancer and so is not a confounder) then because smoking and alcohol are highly associated we would fail to see an effect of smoking. If controls are matched on an exposure then that exposure cannot be studied.
Nested case-control studies
It is possible to ‘nest’ a case-control study into a cohort study. In a nested case-control study design, a cohort of participants has their exposure data collected at baseline. As the cohort is followed up over time, some of the participants develop the disease outcome. These participants become ‘cases’. When these incident cases are identified, another member of the cohort is selected to become a ‘control’. The selection of controls can either be a purely random selection from all participants, or can be limited to a random selection from participants who have matched on specific characteristics. For each case-control pair, the baseline exposure status is compared. Using data from all case-control pairs, it is possible to assess whether there exists an association between the exposure and outcome.
Advantages of nests case control trials
The advantage of a nested case-control study design is two-fold. First, it allows for collection of exposure data in advance of development of the disease outcome. Second, the exposure status of all cohort participants does not need to be assessed at baseline. The exposure status is only measured for cases and the selected controls. This study design is advantageous, for example, when a blood sample can be collected on everyone in the cohort at baseline, but an expensive blood test on the samples only needs to be run on those who are selected as cases and controls
Information (or measurement) bias
It is inevitable that there will be some inaccuracies in the information reported by respondents for some of the exposures. The critical question is whether inaccuracies in exposure measurements are different between cases and control. These inaccuracies lead to what is known as information (or measurement) bias (see Lecture 6: Interpretation of epidemiological studies I). Observer and responder bias are the two main types of information bias.
Observer bias
Observer bias occurs when the process of gathering exposure data by the investigator is different for cases than for controls. Ideally, the investigator or interviewer should be unaware, or ‘blind’/masked to the hypothesis under study and to who is a case and who is a control. While this level of masking is very difficult to achieve perfectly, investigators must be trained in the unbiased collection of data. Information must be collected in an objective and consistent way (e.g. the same forms and questionnaires used for both cases and controls).
Responder or recall bias
Responder or recall bias occurs when the way in which study subjects supply information about exposure differs between cases and controls. Cases, in particular, may be influenced in their answers by awareness of being a case. Responder/recall bias can be minimised by keeping the study members unaware of the hypotheses under study and, where possible, ensuring that both cases and controls have similar incentives to remember past events, or using information that was recorded before disease status was known.
