ERIC Notebook

ERIC NOTEBOOK SERIES

Second Edition

Case-Control Studies

Case-control studies are used to Selection is not based on exposure

Second Edition Authors:
determine if there is an association status. Controls, persons who are free
Lorraine K. Alexander, DrPH between an exposure and a
At baseline:
Brettania Lopes, MPH specific health outcome. These
 ·Selection of cases and controls
studies proceed from effect (e.g.
based on health outcome or
Kristen Ricchetti-Masterson, MSPH health outcome, condition, disease)
disease status
to cause (exposure). Case-control
Karin B. Yeatts, PhD, MS
studies assess whether exposure is  Exposure status is unknown
disproportionately distributed
of the health outcome under study,
between the cases and controls,
are randomly selected from the
which may indicate that the
population out of which the cases
exposure is a risk factor for the
arose. The case-control study aims to
health outcome under study. Case-
achieve the same goals (comparison
control studies are frequently used
of exposed and unexposed) as a
for studying rare health outcomes or
cohort study but does so more
diseases.
efficiently, by the use of sampling.
Unlike cohort or cross-sectional
After cases and controls have been
studies, subjects in case-control
identified, the investigator determines
studies are selected because they
the proportion of cases and the
have the health outcome of interest
proportion of controls that have been
(cases).

*Exposure at some specified point before disease onset

ERIC at the UNC CH Department of Epidemiology Medical Center

 ERIC NOTEBOOK PA G E 2

exposed to the exposure of interest. Thus, the Case-Control Study

denominators obtained in a case-control study do not
Cases Controls
represent the total number of exposed and non-exposed
persons in the source population. Exposed a b
After the investigator determines the exposure, a table can
Unexposed c d
be formed from the study data.

OR = (a/c)/(b/d) = (a/b)/(c/d) = (axd)/(cxb)

Cases Controls
If b and d (from the case-control study) are sampled from
Exposed a b the source population, n1 + n2, then b will represent the
n1 component of the cohort and d will represent the n2
Unexposed c d
component, and (a/n1)/(c/n2) will be estimated by (a/b)/
(c/d).
Measures of incidence in case-control studies
Interpreting the odds ratio
In case-control studies the proportion of cases in the entire
The odds ratio is interpreted the same way as other ratio
population-at-risk is unknown, therefore one cannot
measures (risk ratio, rate ratio, etc.).
measure incidence of the health outcome or disease. The
controls are representative of the population-at-risk, but OR = 1 Odds of disease is the same for exposed
are only a sample of that population, therefore the and unexposed
denominator for a risk measure, the population- at-risk, is
OR > 1 Exposure increases odds of disease
unknown. We decide on the number of diseased people
(cases) and non-diseased people (controls) when we OR < 1 Exposure reduces odds of disease
design our study, so the ratios of controls to cases is not
For example, investigators conducted a case-control study
biologically or substantively meaningful. However, we can
to determine if there is an association between colon
obtain a valid estimate of the risk ratio or rate ratio by
cancer and a high fat diet. Cases were all confirmed colon
using the exposure odds ratio (OR).*
cancer cases in North Carolina in 2010. Controls were a
Odds of exposure among cases = a/c sample of North Carolina residents without colon cancer.
The odds ratio was 4.0. This odds ratio tells us that
Odds of exposure among controls = b/d
individuals who consumed a high fat diet have four times
Diseased person-years the odds of colon cancer than do individuals who do not
consume a high-fat diet. In another study of colon cancer
Disease No Disease and coffee consumption, the OR was 0.60. Thus, the odds
of colon cancer among coffee drinkers is only 0.60 times
Exposed a n1
the odds among individuals who do not consume coffee.
Unexposed c n2 This OR tells us that coffee consumption seems to be
protective against colon cancer.
RR = (a/n1)/(c/n2)
Types of case-control studies
*Note: Under some conditions, the odds ratio approxi-
Case-control studies can be categorized into different
mates a risk ratio or rate ratio. However, this is not
groups based on when the cases develop the health
always the case, and care should be taken to interpret
outcome and based on how controls are sampled. Some
odds ratios appropriately.

ERIC at the UNC CH Department of Epidemiology Medical Center

 ERIC NOTEBOOK PA G E 3

case-control studies use prevalent cases while other case- remained free of the health outcome at the end of follow-
control studies use incident cases. There are also different up then we call the sampling cumulative density sampling
ways that cases can be identified, such as using or survivor sampling. Controls cannot ever have the
population-based cases or hospital-based cases. outcome (become cases) when using this type of
sampling. In these case-control studies, the odds ratio
Types of cases used in case control studies
estimates the rate ratio only if the health outcome is rare,
Prevalent cases are all persons who were existing cases of i.e. if the proportion of those with the health outcome
the health outcome or disease during the observation among each exposure group is less than 10% (requires
period. These studies yield a prevalence odds ratio, which the rare disease assumption).
will be influenced by the incidence rate and survival or
Incidence density sampling or risk set sampling
migration out of the prevalence pool of cases, and thus
does not estimate the rate ratio. Case control studies can When cases are incident cases and when controls are
also use incident cases, which are persons who newly selected from the at-risk source population at the same
develop the health outcome or disease during the time as cases occur (controls must be eligible to become
observation period. Recall that prevalence is influenced by a case if the health outcome develops in the control at a
both incidence and duration. Researchers that study later time during the period of observation) then we call
causes of disease typically prefer incident cases because this type of sampling incidence density sampling or risk
they are usually interested in factors that lead up to the set sampling. The control series provides an estimate of
development of disease rather than factors that affect the proportion of the total person-time for exposed and
duration. unexposed cohorts in the source population. In these case
-control studies, the odds ratio estimates the rate ratio of
Selecting controls
cohort studies, without assuming that the disease is rare
Selection of controls is usually the most difficult part of in the source population.
conducting a case-control study. We will discuss 3 possible
Note that it is possible, albeit rare, that a control selected
ways to select controls:
at a later time point could become a case during the
1. Base or case-base sampling remaining time that the study is running. This differs from
2. Cumulative density or survivor sampling case-control studies that use cumulative density sampling
or survivor sampling, which select their controls after the
3. Incidence density or risk set sampling
conclusion of the study from among those individuals
remaining at risk.
Base sampling or case-base sampling
Selecting controls in a risk set sampling or incidence
This sampling involves using controls selected from the density sampling manner provides two advantages:
source population such that every person has the same
1. A direct estimate of the rate ratio is possible.
chance of being included as a control. This type of
sampling only works with a previously defined cohort. In 2. The estimates are not biased by differential loss to
these case-control studies, the odds ratio provides a valid follow up among the exposed vs. unexposed controls.
estimate of the risk ratio without assuming that the For example, if a large number of smokers left the source
disease is rare in the source population. population after a certain time point, they would not be
Cumulative density sampling or survivor sampling available for selection at the end of the study – when
controls would be selected in a study that uses cumulative
When controls are sampled from those people who
density sampling or survivor sampling. This would give the

ERIC at the UNC CH Department of Epidemiology Medical Center

 ERIC NOTEBOOK PA G E 4

investigators biased information regarding the level of prevalence in the source population of cases, e.g.
exposure among the controls over the course of the study. there may be a higher prevalence of smokers in
hospitals. Hospital controls also may have diseases
Source populations for case-control studies
resulting from the exposure of interest, e.g. the
Source populations can be restricted to a population of exposure (smoking) is related to the disease of
particular interest, e.g. postmenopausal women at risk of interest (cancer) and to heart and lung diseases from
breast cancer. This restriction makes it easier to control for which the controls may be suffering.
extraneous confounders in the population. Controls should
represent the restricted source population from which cases  Controls with another disease - However if the study is
arise, not all non-cases in the total population. The cases in on lung cancer, for example, it is essential to exclude
the study do not have to include all cases in the total cancers known or suspected to be related to the study
population. exposure of interest. These controls also share some
of the same problems as hospital controls.
Sources of cases
Advantages of case-control studies
 Cases diagnosed in a hospital or clinic
Case-control studies are the most efficient design for rare
 Cases entered into a disease registry, e.g. cancer, birth diseases and require a much smaller study sample than
defects, deaths cohort studies. Additionally, investigators can avoid the
 Cases identified through mass screening, e.g. logistical challenges of following a large sample over time.
hypertensives, diabetics Thus, case-control studies also allow more intensive
evaluation of exposures of cases and controls. Case-
 Cases identified through a prior cohort study, e.g. lung control studies that use incidence density sampling or risk
cancers in an occupational asbestos cohort set sampling yield a valid estimate of the rate ratio derived
Sources of controls from a cohort study if incident cases are studied and
controls are sampled from the risk set of the source
 Population controls are non-cases sampled from the
population. If properly performed (i.e. appropriate
source population giving rise to cases. This is the most
sampling), case-control studies provide information that
desirable method for selecting controls. Sampling
mirrors what could be learned from a cohort study, usually
randomly from census block groups, or a registry such as
at considerably less cost and time.
the Department of Motor Vehicles (of adults who are
able to drive) are examples of ways to find and recruit Disadvantages of case-control studies
population-based controls. Case-control studies do not yield an estimate of rate or
 Neighborhood or friend controls are appropriate for risk, as the denominator of these measures is not defined.
selection as controls if these individuals would be Case-control studies may be subject to recall bias if
included as cases if they developed the health outcome exposure is measured by interviews and if recall of
of interest. It is not appropriate to select neighbors or exposure differs between cases and controls. However,
friends as controls if they share the exposure of interest. investigators may be able to avoid this problem if historical
records are available to assess exposure. Choosing an
 Hospital controls - There are certain problems with appropriate source population is also difficult and may
hospital controls in that they may not be from the same contribute to selection bias. Case-control studies are not
source population from which the cases arose. Hospital an efficient means for studying rare exposures (less than
controls may not be representative of the exposure 10% of controls are exposed) because very large numbers
of cases and controls are needed to detect the effects of
rare exposures.
ERIC at the UNC CH Department of Epidemiology Medical Center
 ERIC NOTEBOOK PA G E 5

Terminology References
Cohort studies: An observational study in which Dr. Carl M. Shy, Epidemiology 160/600 Introduction to
subjects are sampled based on the presence (exposed) Epidemiology for Public Health course lectures, 1994-
or absence(unexposed) of a risk factor of interest. 2001, The University of North Carolina at Chapel Hill, De-
These subjects are followed over time for the partment of Epidemiology
development of a health outcome of interest.
Rothman KJ, Greenland S. Modern Epidemiology. Second
Cross-sectional studies: An observational study in Edition. Philadelphia: Lippincott Williams and Wilkins,
which subjects are sampled at one point in time, and 1998.
then the associations between the concurrent risk
The University of North Carolina at Chapel Hill, Department
factors and health outcomes are investigated.
of Epidemiology Courses: Epidemiology 710, Fundamen-
Exposure odds ratio (OR): the odds of a particular tals of Epidemiology course lectures, 2009-2013, and Epi-
exposure among persons with a specific health demiology 718, Epidemiologic Analysis of Binary Data
outcome divided by the corresponding odds of course lectures, 2009-.2013.
exposure among persons without the health outcome
of interest. Yields a valid estimate of the incidence rate
ratio or risk ratio derived from a cohort study,
depending on control sampling.
Incident case: a person who is newly diagnosed as a
case.
Prevalent case: a person who has a health outcome of
interest that was diagnosed in the past.
Risk ratio (RR): the likelihood of a particular health
outcome occurrence among persons exposed to a
given risk factor divided by the corresponding
likelihood among unexposed persons.
Source population: the population out of which the
cases arose.
From: Medical Epidemiology, R.S. Greenberg, 1993,
1996.

Acknowledgement
The authors of the Second Edition of the ERIC Notebook would like t o acknowledge t he aut hors of the ERIC
N ot eb ook, Firs t Edit ion: Michel Ib rahim , MD, PhD, Lorraine Alexander, DrPH, Carl Shy, MD, DrPH,
and Sherry Farr, GRA, Departm ent of Epidem iology at the Univers it y of N ort h Carolina at Chapel
Hill. The First Edition of t he ERIC Notebook was produced b y the Educat ional Arm of t he
Epidem iologic Research and Information Cent er at Durham, N C. The funding for t he ERIC N ot ebook
First Edit ion was provided b y t he Departm ent of V et erans Affairs (DV A), V et erans Health
Administ ration (V HA), Cooperat ive St udies Program (CSP) t o promote the s trat egic growth of the
epidem iologic capacit y of the DV A.

ERIC at the UNC CH Department of Epidemiology Medical Center