Case-Control Studies

ERIC at the UNC CH Department of Epidemiology Medical Center

E R I C N O T E B O O K S E R I E S

Case-control studies are used to

determine if there is an association

between an exposure and a specific

health outcome. These studies

proceed from effect (e.g. health

outcome, condition, disease) to

cause (exposure). Case-control

studies assess whether exposure is

disproportionately distributed

between the cases and controls,

which may indicate that the

exposure is a risk factor for the

health outcome under study. Case-

control studies are frequently used

for studying rare health outcomes or

diseases.

Unlike cohort or cross-sectional

studies, subjects in case-control

studies are selected because they

have the health outcome of interest

(cases). Selection is not based on

exposure status. Controls, persons

who are free of the health outcome

under study, are randomly selected

from the population out of which the

cases arose. The case-control study

aims to achieve the same goals

(comparison of exposed and

unexposed) as a cohort study but

does so more efficiently, by the use of

sampling.

After cases and controls have been

identified, the investigator determines

the proportion of cases and the

proportion of controls that have been

Second Edition Authors:

Lorraine K. Alexander, DrPH

Brettania Lopes, MPH

Kristen Ricchetti-Masterson, MSPH

Karin B. Yeatts, PhD, MS

Second Edition

At baseline:

 ·Selection of cases and controls

based on health outcome or

disease status

 Exposure status is unknown

*Exposure at some specified point before disease onset

ERIC at the UNC CH Department of Epidemiology Medical Center

exposed to the exposure of interest. Thus, the

denominators obtained in a case-control study do not

represent the total number of exposed and non-exposed

persons in the source population.

After the investigator determines the exposure, a table can

be formed from the study data.

Measures of incidence in case-control studies

In case-control studies the proportion of cases in the entire

population-at-risk is unknown, therefore one cannot

measure incidence of the health outcome or disease. The

controls are representative of the population-at-risk, but

are only a sample of that population, therefore the

denominator for a risk measure, the population- at-risk, is

unknown. We decide on the number of diseased people

(cases) and non-diseased people (controls) when we

design our study, so the ratios of controls to cases is not

biologically or substantively meaningful. However, we can

obtain a valid estimate of the risk ratio or rate ratio by

using the exposure odds ratio (OR).*

Diseased person-years

RR = (a/n1)/(c/n2)

Case-Control Study

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

If b and d (from the case-control study) are sampled from

the source population, n1 + n2, then b will represent the

n1 component of the cohort and d will represent the n2

component, and (a/n1)/(c/n2) will be estimated by (a/b)/

(c/d).

Interpreting the odds ratio

The odds ratio is interpreted the same way as other ratio

measures (risk ratio, rate ratio, etc.).

For example, investigators conducted a case-control study

to determine if there is an association between colon

cancer and a high fat diet. Cases were all confirmed colon

cancer cases in North Carolina in 2010. Controls were a

sample of North Carolina residents without colon cancer.

The odds ratio was 4.0. This odds ratio tells us that

individuals who consumed a high fat diet have four times

the odds of colon cancer than do individuals who do not

consume a high-fat diet. In another study of colon cancer

and coffee consumption, the OR was 0.60. Thus, the odds

of colon cancer among coffee drinkers is only 0.60 times

the odds among individuals who do not consume coffee.

This OR tells us that coffee consumption seems to be

protective against colon cancer.

Types of case-control studies

Case-control studies can be categorized into different

groups based on when the cases develop the health

outcome and based on how controls are sampled. Some

E R I C N O T E B O O K PA G E 2

Cases Controls

Exposed

a b

Unexposed

c d

Odds of exposure among cases = a/c

Odds of exposure among controls = b/d

Disease No Disease

Exposed

a n

Unexposed

c n

Cases Controls

Exposed

a b

Unexposed

c d

OR = 1 Odds of disease is the same for exposed

and unexposed

OR > 1 Exposure increases odds of disease

OR < 1 Exposure reduces odds of disease

*Note: Under some conditions, the odds ratio approxi-

mates a risk ratio or rate ratio. However, this is not

always the case, and care should be taken to interpret

odds ratios appropriately.

ERIC at the UNC CH Department of Epidemiology Medical Center

remained free of the health outcome at the end of follow-

up then we call the sampling cumulative density sampling

or survivor sampling. Controls cannot ever have the

outcome (become cases) when using this type of

sampling. In these case-control studies, the odds ratio

estimates the rate ratio only if the health outcome is rare,

i.e. if the proportion of those with the health outcome

among each exposure group is less than 10% (requires

the rare disease assumption).

Incidence density sampling or risk set sampling

When cases are incident cases and when controls are

selected from the at-risk source population at the same

time as cases occur (controls must be eligible to become

a case if the health outcome develops in the control at a

later time during the period of observation) then we call

this type of sampling incidence density sampling or risk

set sampling. The control series provides an estimate of

the proportion of the total person-time for exposed and

unexposed cohorts in the source population. In these case

-control studies, the odds ratio estimates the rate ratio of

cohort studies, without assuming that the disease is rare

in the source population.

Note that it is possible, albeit rare, that a control selected

at a later time point could become a case during the

remaining time that the study is running. This differs from

case-control studies that use cumulative density sampling

or survivor sampling, which select their controls after the

conclusion of the study from among those individuals

remaining at risk.

Selecting controls in a risk set sampling or incidence

density sampling manner provides two advantages:

1. A direct estimate of the rate ratio is possible.

2. The estimates are not biased by differential loss to

follow up among the exposed vs. unexposed controls.

For example, if a large number of smokers left the source

population after a certain time point, they would not be

available for selection at the end of the study – when

controls would be selected in a study that uses cumulative

density sampling or survivor sampling. This would give the

case-control studies use prevalent cases while other case-

control studies use incident cases. There are also different

ways that cases can be identified, such as using

population-based cases or hospital-based cases.

Types of cases used in case control studies

Prevalent cases are all persons who were existing cases of

the health outcome or disease during the observation

period. These studies yield a prevalence odds ratio, which

will be influenced by the incidence rate and survival or

migration out of the prevalence pool of cases, and thus

does not estimate the rate ratio. Case control studies can

also use incident cases, which are persons who newly

develop the health outcome or disease during the

observation period. Recall that prevalence is influenced by

both incidence and duration. Researchers that study

causes of disease typically prefer incident cases because

they are usually interested in factors that lead up to the

development of disease rather than factors that affect

duration.

Selecting controls

Selection of controls is usually the most difficult part of

conducting a case-control study. We will discuss 3 possible

ways to select controls:

1. Base or case-base sampling

2. Cumulative density or survivor sampling

3. Incidence density or risk set sampling

Base sampling or case-base sampling

This sampling involves using controls selected from the

source population such that every person has the same

chance of being included as a control. This type of

sampling only works with a previously defined cohort. In

these case-control studies, the odds ratio provides a valid

estimate of the risk ratio without assuming that the

disease is rare in the source population.

Cumulative density sampling or survivor sampling

When controls are sampled from those people who

E R I C N O T E B O O K PA G E 3

ERIC at the UNC CH Department of Epidemiology Medical Center

investigators biased information regarding the level of

exposure among the controls over the course of the study.

Source populations for case-control studies

Source populations can be restricted to a population of

particular interest, e.g. postmenopausal women at risk of

breast cancer. This restriction makes it easier to control for

extraneous confounders in the population. Controls should

represent the restricted source population from which cases

arise, not all non-cases in the total population. The cases in

the study do not have to include all cases in the total

population.

Sources of cases

 Cases diagnosed in a hospital or clinic

 Cases entered into a disease registry, e.g. cancer, birth

defects, deaths

 Cases identified through mass screening, e.g.

hypertensives, diabetics

 Cases identified through a prior cohort study, e.g. lung

cancers in an occupational asbestos cohort

Sources of controls

 Population controls are non-cases sampled from the

source population giving rise to cases. This is the most

desirable method for selecting controls. Sampling

randomly from census block groups, or a registry such as

the Department of Motor Vehicles (of adults who are

able to drive) are examples of ways to find and recruit

population-based controls.

 Neighborhood or friend controls are appropriate for

selection as controls if these individuals would be

included as cases if they developed the health outcome

of interest. It is not appropriate to select neighbors or

friends as controls if they share the exposure of interest.

 Hospital controls - There are certain problems with

hospital controls in that they may not be from the same

source population from which the cases arose. Hospital

controls may not be representative of the exposure

prevalence in the source population of cases, e.g.

there may be a higher prevalence of smokers in

hospitals. Hospital controls also may have diseases

resulting from the exposure of interest, e.g. the

exposure (smoking) is related to the disease of

interest (cancer) and to heart and lung diseases from

which the controls may be suffering.

 Controls with another disease - However if the study is

on lung cancer, for example, it is essential to exclude

cancers known or suspected to be related to the study

exposure of interest. These controls also share some

of the same problems as hospital controls.

Advantages of case-control studies

Case-control studies are the most efficient design for rare

diseases and require a much smaller study sample than

cohort studies. Additionally, investigators can avoid the

logistical challenges of following a large sample over time.

Thus, case-control studies also allow more intensive

evaluation of exposures of cases and controls. Case-

control studies that use incidence density sampling or risk

set sampling yield a valid estimate of the rate ratio derived

from a cohort study if incident cases are studied and

controls are sampled from the risk set of the source

population. If properly performed (i.e. appropriate

sampling), case-control studies provide information that

mirrors what could be learned from a cohort study, usually

at considerably less cost and time.

Disadvantages of case-control studies

Case-control studies do not yield an estimate of rate or

risk, as the denominator of these measures is not defined.

Case-control studies may be subject to recall bias if

exposure is measured by interviews and if recall of

exposure differs between cases and controls. However,

investigators may be able to avoid this problem if historical

records are available to assess exposure. Choosing an

appropriate source population is also difficult and may

contribute to selection bias. Case-control studies are not

an efficient means for studying rare exposures (less than

10% of controls are exposed) because very large numbers

of cases and controls are needed to detect the effects of

rare exposures.

E R I C N O T E B O O K PA G E 4

ERIC at the UNC CH Department of Epidemiology Medical Center

Practice Questions

Answers are located at the end of this notebook

1) Researchers conduct a case-control study of breast

cancer, using incident cases. The researchers find out

that 90% of the cases had taken hormonal contraceptives

in the past. Should the researchers conclude that hormo-

nal contraceptives increase the risk of developing breast

cancer?

2) Researchers conduct a case-control study of pancreatic

cancer. The study included 200 cases and 200 controls.

Of the cases, 80% reported they smoked cigarettes.

Among the controls, 50% reported they smoked cigarettes.

a) Prepare a 2x2 table with these data

b) Calculate the exposure odds ratio

c) Interpret the exposure odds ratio in a sentence

References

Dr. Carl M. Shy, Epidemiology 160/600 Introduction to

Epidemiology for Public Health course lectures, 1994-

2001, The University of North Carolina at Chapel Hill, De-

partment of Epidemiology

Rothman KJ, Greenland S. Modern Epidemiology. Second

Edition. Philadelphia: Lippincott Williams and Wilkins,

1998.

The University of North Carolina at Chapel Hill, Department

of Epidemiology Courses: Epidemiology 710, Fundamen-

tals of Epidemiology course lectures, 2009-2013, and Epi-

demiology 718, Epidemiologic Analysis of Binary Data

course lectures, 2009-.2013.

E R I C N O T E B O O K PA G E 5

Acknowledgement

The authors of the Second Edition of the ERIC Notebook

would like to acknowledge th e authors of the

ERIC Not ebook, First Edition: Mic hel Ibra him,

MD, PhD, Lorrai ne Alexand er, DrPH, Carl Sh y,

MD, DrPH, and Sherry Farr, GRA, Department of

Epidemiology at t he Univers ity of No rt h Caro lina

at Cha pel Hill. The First Edition of the ER IC

Notebook was produced b y t he Educ ational Arm

of the Epidemio logic Research and Informat ion

Center at Durh am, NC. The funding for the ERIC

Notebook First Ed ition was provided by th e

Departm ent of V eterans Affairs (DV A), Vetera ns

Health Administration (VHA), Cooperative

Studi es Program (CSP) to promote the strat egic

growth of the epid emiolog ic capacity of th e DVA.

Terminology

Cohort studies: An observational study in which

subjects are sampled based on the presence (exposed)

or absence(unexposed) of a risk factor of interest.

These subjects are followed over time for the

development of a health outcome of interest.

Cross-sectional studies: An observational study in

which subjects are sampled at one point in time, and

then the associations between the concurrent risk

factors and health outcomes are investigated.

Exposure odds ratio (OR): the odds of a particular

exposure among persons with a specific health

outcome divided by the corresponding odds of

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.

ERIC at the UNC CH Department of Epidemiology Medical Center

Answers to Practice Questions

1. No, The information provided in this question is only for

cases. No information was given in the question about the

control group that was studied. To assess if exposure to

hormonal contraceptives was associated with the risk of

breast cancer, information would be required on the pro-

portion of control subjects who had taken hormonal con-

traceptives and the researchers would need to calculate

the exposure odds ratio.

a) 2x2 table

b) Exposure odds ratio = (a/c) / (b/d) = (a*d)/(c*b) =

(160*100) / (40*100) = 4.0

c) An odds ratio of 4.0 means that the odds of smokers

being a case are 4 times the odds of non-smokers being a

case.

Cases Controls

Exposed 160 (a) 100 (b) 260

Unexposed 40 (c) 100 (d) 140

200 200 400

E R I C N O T E B O O K PA G E 6