Instructor Resource

Chapter 11

Copyright © Scott B. Patten, 2015.

Permission granted for classroom use with Epidemiology for Canadian Students: Principles, Methods & Critical Appraisal (Edmonton: Brush Education Inc. www.brusheducation.ca).

Chapter 11. Case-control studies

Objectives

• Define case-control studies.• Explain how to interpret measures of association calculated

from case-control data.• Describe recall bias.• Describe the rare disease assumption.• Distinguish between induction and latency periods for

disease, and describe the dynamic of component causes of disease.• Define primary, secondary, and tertiary prevention.• List strengths and weaknesses of case-control studies.

What is a case-control study?• It is a type of analytic study: it usually has analytic

goals.• Participants are selected based on their disease

status.• Participants include:• cases (who have the disease under investigation)• controls (who do not)

• Exposure is assessed retrospectively.

2 x 2 table for case-control studies

  Cases Controls

Exposed a b

Nonexposed c d

Total ncases ncontrols

2 x 2 table for case-control studies(continued)

• You can’t calculate prevalence from this 2 x 2 table because there is no expectation that the estimated proportion with disease would reflect the population prevalence.

A measure of association for case-control studies

Odds R atio=

𝑎𝑐𝑏𝑑

=𝑎𝑑𝑏𝑐

Odds ratios

• Recall that a prevalence odds ratio could be calculated from cross-sectional data.• The formula was the same.• However, as the case-control design usually has

analytical goals, incidence is of more interest.• Case-control studies rarely use prevalent cases.• When the cases are incident cases, the estimated

odds ratio is an incidence odds ratio.

Advantages of case-control studies• The main advantage of the case-control design is its

efficiency. • The design is efficient because selection of cases

can be accomplished flexibly and the control group is usually only a small subset of the analogous group in the general population.• Therefore, a relatively small investment of research

resources can produce a precise estimate.

Case-control studies

• Links to examples:• http://www.ncbi.nlm.nih.gov/pubmed/8059766• http://www.ncbi.nlm.nih.gov/pubmed/15364185• http://www.ncbi.nlm.nih.gov/pubmed/23164221• http://www.ncbi.nlm.nih.gov/pubmed/25208536• http://www.ncbi.nlm.nih.gov/pubmed/21033532

Selection bias in case-control studies• Selection bias is a major methodological concern

for case-control studies.• Selection bias results if selection of cases and

controls depends on exposure in some way that is not equivalent between the 2 groups.• It is easiest to be confident that selection bias has

not occurred if the selection of cases and controls does not depend in any way on exposure.

Berkson’s bias

• This is bias that arises when selection of cases and controls occurs in a hospital setting.• The probability of hospitalization is often related to

exposure in a way that differs in cases and controls.

Misclassification bias in case-control studies• Since the target of estimation is a more complex

parameter (an odds ratio rather than simply odds, a frequency, or a rate), this is a more difficult issue than it was in prevalence or incidence studies.• The problem becomes easier if you consider the

numerator (odds of exposure in cases) and the denominator (odds of exposure in controls) separately.• If the numerator is inflated, the estimate of the odds

ratio will be too high. • If the denominator is inflated, the estimate of the

odds ratio will be too low.

Nondifferential misclassification of exposure• Exposure assessment is prone to error in case-

control studies because it is retrospective.• The measure of exposure can have low sensitivity

and/or low specificity.• Low sensitivity produces false negatives.• Low specificity produces false positives.• If the error rates are the same in cases and

controls, the misclassification is nondifferential.

Differential misclassification of exposure• If the error rates in classification of exposure are

different in the cases and controls, the misclassification is differential.• A classical type of differential misclassification bias

in case-control studies is recall bias.

Recall bias

• Imagine a study of drug exposures during pregnancy (exposure) as a risk factor for birth defects (outcome).• Since mothers of babies with birth defects are likely

to reflect more deeply about their pregnancy, they may have more accurate recall of past drug exposures. • What effect would this have on the odds ratio (OR)?

Recall bias (continued)

• The sensitivity for exposure in the controls is lower than in the cases.• Therefore, the odds of exposure in the controls

(denominator of the OR) will be too small.• This will lead to an overestimation of the OR.

Nondifferential misclassification bias• What if the sensitivity for assessment of exposure

was the same (nondifferential) in cases and controls.• The direction of bias is towards the null value for

the OR (which is 1).

The rare disease assumption• Imagine the classic 2 x 2 table.• Note that when a disease is rare, a and b are small

and the ratio of the 2 odds approximates the ratio of the analogous 2 proportions.

  Disease No Disease

Exposed a b

Nonexposed c d

The rare disease assumption (continued)• When the disease is rare….

abis approximately

aa+b

  Disease No Disease

Exposed a b

Nonexposed c d

The rare disease assumption (continued)• (and) When the disease is rare….

cdis approximately

cc+d

  Disease No Disease

Exposed a b

Nonexposed c d

Induction period

• This is the period of time between exposure to a risk factor and the initiation of a disease mechanism.• An advantage of the case-control design is that it

can assess associations characterized by long induction periods.

Latency period

• Latent means “hidden.”• A latency period is the time between the initiation

of a disease and its emergence as a clinically evident entity.

Primary, secondary, and tertiary prevention

• Primary prevention: preventing the occurrence of disease by eliminating a risk factor or its effects (occurs in the induction phase).• Secondary prevention: improving outcomes

through earlier detection (in the latency phase).• Tertiary prevention: diminishing the impact of a

disease (after the disease has become a clinical entity).

Strengths of the case-control study design• efficiency• fit with rare diseases• fit with long induction and latency periods• ability to examine multiple exposures

Weaknesses of the case-control study design• vulnerable to selection and misclassification bias• not a good fit for rare exposures• limited to estimating odds ratios (cannot, for

example, estimate incidence.• temporality issues: in studies of prevalent cases,

the timing of exposure and disease may be unclear

End