Case-control studies

Question 1

Case-control study - design

Answer 1

Cases and non-cases are enrolled and the frequency of exposure to a factor is measured retrospectively and contrasted between groups. Usually performed retrospectively (outcome has already occurred when the study begins). Preferable to select incident cases since risk factors for prevalence reflect both incidence and duration.

Question 2

Case-control study - defining the source population (study base)

Answer 2

1. Primary study base (population-based): representative sample of all cases drawn from a clearly delineated population, with controls drawn from same population [e.g. using disease registries, farm records] - avoids selection bias but more difficult to implement
2. Secondary study base: one or more steps removed from the source population [e.g. referral clinic, laboratory]

In latter situation, it is harder to identify the source population that gave rise to the cases. Thus it is more difficult to select appropriate controls.

Question 3

Case selection in case-control study - considerations (3)

Answer 3

Purposive samplling. Ideally, cases are randomly sampled from animals in the source population who have the disease of interest. All cases may be included (e.g. rare outcomes) or a representative sample selected. Sampling method must be independent of exposure status.

1. Identify the source population - primary or secondary base
2. Well-defined case definition - clinical and diagnostic criteria (note that detailed criteria could mean cases are different to individuals in source population e.g. high cost of diagnosis)
3. Determine whether only incident cases (best) or incident and prevalent cases will be selected

Question 4

??? Control selection in case-control study - considerations ???

Answer 4

Purposive sampling. Ideally, controls are randomly sampled from animals in the source population who do not have the disease of interest (can be difficult if cases are selected from a secondary-base as source population is difficult to define). In general, controls should be representative of the exposure experience in the population which gave rise to the cases, i.e. they would have been cases if the outcome occurred. Sampling should be independent of the exposure of interest.

1. Risk-based (cumulative incidence) sampling: controls selected from among those that did not become cases by the end of the study period; design is appropriate if the population is closed and is most informative if the risk period is oer before subject selection begins (e.g. point-source outbreak of disease with short incubation)
2. Rate-based (incidence-density) sampling: controls selected

Question 5

Case-control studies - analysis

Answer 5

For dichotamous outcome/exposure: report proportion of cases and controls who were exposed (chi square test). Crude and adjusted ORs (logistic regression).

Question 6

Case-control studies - advantages (4), disadvantages (4)

Answer 6

Advantages Useful for rare diseases and those which develop over a long time Useful for preliminary investigation of causal hypotheses Relatively quick Relatively inexpensive Disadvantages Only allow study of single outcome Do not provide information on disease frequency in the population Not suitable for rare exposures Difficult to ensure unbiased selection of control group

Question 7

Case-control study - bias

Answer 7

1. Selection bias: a) can arise if cases/controls are not representative of the distribution of exposures in cases/controls in the source population (e.g. sampling occurs in a way that is dependent on the exposure), b) self-selection bias (cat owners may be more likely to participate in study on bartonellosis if “study on cat scratch disease” is announced), c) detection bias (people who have been scratched by a cat may be more likely to seek medical assistance and get tested for cat scratch fever)
2. Information bias: a) recall bias - if exposure is self-reported (since outcome has already occurred)[not an issue if exposure status is determined using records generated before outcome occurred] e.g. person with cat scratch might be more likely to recall being scratched than someone without diagnosis.
3. Confounding bias

Question 8

Case-control study - example

Answer 8

Investigation of risk factors for zoonotic transmission of bartonellosis (cat scratch disease)

1. Case selection from infectious disease clinic (secondary base), with controls selected from same clinic
2. If advertized as a study on “cat scratch disease” might enroll a greater proportion of cat owners (selection bias)
3. Cat owners may have greater awareness of the disease (given namesake) and so may be more likely to seek medical attention and get diagnosed compared to dog owners (detection bias - dog owners more likely to be non-cases purely due to lower rates of testing)
4. If exposures are self reported, cases might be more likely to recall being scratched or bitten by a cat compared to non-cases