NURSING RESEARCH,

STEP BY STEP By Courtney Keeler, PhD, and Alexa Colgrove Curtis, PhD, MPH, FNP, PMHNP
A series coordinated by the Heilbrunn Family Center for Research Nursing at Rockefeller University

Case–Control Studies
Appropriate uses of and approaches to this form of
observational design.

Editor’s note: This is the eighth article in a series on clinical research by nurses. The series is designed to
give nurses the knowledge and skills they need to participate in research, step by step. Each column will
present the concepts that underpin evidence-based practice—from research design to data interpreta-
tion. The articles will be accompanied by a podcast offering more insight and context from the authors. To
see all the articles in the series, go to https://links.lww.com/AJN/A204.

L
ike cohort studies, case–control studies con-
nect a health outcome with a specific expo-
sure, allowing researchers to compare the
association between the two. Here, we compare
cohort and case–control studies, discuss the selec-
tion of cases and controls, describe how one might
evaluate whether an association exists between
them, and outline potential sources of bias. We
end by discussing when a case–control study might
be appropriate.
COHORT VS. CASE–CONTROL STUDIES
Both case–control and cohort studies compare a
given exposure with a specific health outcome. Like
cohort studies, case–control studies are observa-
tional, meaning that exposure is “observed” rather
than assigned by the researcher. For instance, sup-
pose a researcher is interested in exploring the rela-
tionship between smoking status and heart disease.
In an observational study, the researcher would
“observe” smoking status within a sample, rather
than assign participants to smoking and nonsmok-
ing groups. Indeed, this example highlights the
necessity of an observational study design in some
contexts, since a researcher cannot ethically assign
subjects to dangerous exposure groups like smoking.
As with all observational study designs, causal-
ity is difficult to establish in case–control and cohort
studies. Instead, in interpreting study findings, the
researcher assesses whether an association exists
between an exposure and a given health outcome. The
researcher cannot determine cause and effect because,
unlike in an experimental study such as a randomized
controlled trial, exposure is not randomly assigned,
and subsequent health outcomes are not tracked.
In determining whether any association exists
between an exposure and an outcome, cohort stud-
ies begin with an exposure and track subsequent
outcome manifestations.1, 2 This process can be
either prospective (tracking future outcome mani-
festations) or retrospective (reflecting on past expe-
rience). Unlike cohort studies, case–control studies
begin with a specific disease or other health out-
come and retrospectively track exposure. (Figure 1
details this distinction in study design.) More specif-
ically, case–control studies begin with two groups:
those with a specific disease or health outcome
(cases) and those without the specific disease or
health outcome (controls). Next, researchers com-
pare exposure between those with and those with-
out the disease or health outcome. In summary, in a
cohort study one compares rates of disease, in a
case–control study one compares rates of exposure.
Let’s consider an example. As part of a larger
cohort study, D’Souza and colleagues used a case–
control framework to investigate the association
between human papillomavirus (HPV) and the patho-
genesis of oropharyngeal squamous-cell carcinoma.3
The investigators recruited 300 participants: 100
patients with a newly confirmed diagnosis of oropha-
ryngeal squamous-cell carcinoma and 200 controls.
The control group met four important criteria; con-
trols had no history of cancer (1) and were treated for
benign conditions (2) in the same clinic as the cases (3)
during the same time frame (4). The research team
then assessed the presence of HPV in each group, col-
lecting specimens from the cases before treatment and
from the controls at enrollment.3
CASE–CONTROL STUDIES
Identifying cases. Case–control studies begin
with the identification of cases. To identify cases,
researchers need to create a working definition of
the disease or health outcome of interest. They can

ajn@wolterskluwer.com AJN ▼ February 2022 ▼ Vol. 122, No. 2 51

 NURSING RESEARCH,
STEP BY STEP
Figure 1. Comparison of Cohort and Case–Control Study Designs
Design of a Cohort Study
POPULATION
NOT
EXPOSED EXPOSED
DISEASE DISEASE COMPARE
RATES OF
NO NO DISEASE
DISEASE DISEASE
rely on specific diagnostic criteria or use validated
survey instruments to assess the presence of disease
or other health outcome. Before the study begins
(or a priori), researchers need to identify who will
and will not be included in the study sample, estab-
lishing inclusion and exclusion criteria. Importantly,
researchers will need to ensure that cases are drawn
from the target population of interest, so the study
findings are generalizable to this group.
For instance, a researcher may be interested in
understanding what exposures predisposed clini-
cal nurses toward severe cases of COVID-19 during
the pandemic, such as length of a typical shift; type
of personal protective equipment used; clinical unit;
or underlying health characteristics such as age, body
mass index, or presence of pulmonary disease. Like
many health outcomes, COVID-19 severity could
be measured in several ways, including by hospital-
ization, duration of disease, and oxygen utilization,
among other factors. Suppose the researcher defines
severe cases of COVID-19 as those requiring hospital-
ization with use of oxygen. With this case definition
in mind, the researcher can select participants who
were hospitalized with COVID-19 over a fixed period
(such as March 1 through December 31, 2020).
In a related example, Cai and colleagues investi-
gated the association between frontline medical
work in China during the pandemic and mental
health outcomes.4 Medical workers were defined as
“physicians, nurses, and other healthcare workers
[such as] medical technicians, respiratory therapists,
or emergency room attendants.” Cases were identi-
fied based on responses to a series of questionnaires
sent to individuals working in hospitals in China
between February 11 and 26, 2020, who met the
above definition of a “medical worker.”
Identifying controls. Criteria for selecting con-
trols. Controls provide a comparison with those
52 AJN ▼ February 2022 ▼ Vol. 122, No. 2 ajnonline.com
Design of a Case–Control Study
POPULATION
NO
DISEASE
DISEASE
EXPOSED EXPOSED COMPARE
RATES OF
NOT NOT EXPOSURE
EXPOSED EXPOSED
who manifest target outcomes (cases). In case–
control studies, controls must meet two important
requirements. First, controls must be drawn from
the same “at-risk” target population as cases, ensur-
ing that both cases and controls are exposed to sim-
ilar observable and unobservable factors that influ-
ence the association of interest.5 Second, exposure
status must not influence their selection. Further,
that exposure should ideally be measured with com-
parable precision across case and control groups.
Consider a researcher exploring the relationship
between cigarette smoking and stroke among women
ages 18 and older. The researcher should ensure that
both cases (those who experienced a stroke) and con-
trols (those who did not experience a stroke) were
drawn from the same at-risk population. Addition-
ally, in assessing smoking behaviors, the researcher
should gauge exposure with equal accuracy. Admit-
tedly, in this example, it would be difficult to collect
equally accurate information from cases and controls
if case patients were unable to self-report smoking
status. In this case, because the researcher might need
to rely on secondary accounts from a partner, rela-
tive, friend, or if available, medical records, the sec-
ond criterion would be hard to meet.
Identifying populations from which to draw con-
trols. A researcher has several options when select-
ing controls: population-based controls and hospi-
tal- or clinic-based controls.1, 2 In some cases,
researchers utilize both options (that is, they have
two control populations).
Researchers seek to minimize the influence of con-
founders by optimizing the matching of cases with
controls. When selecting controls, the researcher tries
to mirror cases in as many observable dimensions as
possible (for example, age, weight, or gender). One
way to accomplish this is through individual or fre-
quency matching with population- and clinical-based
controls. In individual matching, each case subject
is matched with a control subject with similar char-
acteristics. In frequency matching, the proportion of
potential confounding factors is matched within each
comparison group.
Population-based controls are selected from the
target population. Researchers can use school ros-
ters, selective service lists, random digit dialing, voter
registration lists, or similar resources for recruiting.
While in theory all members of this population have
an equal probability of being selected for the sample,
this outcome is unlikely in practice. Although this
scheme works best with a well-defined target popula-
tion, even then, recruiting population-based controls
can be difficult and time consuming.1
Suppose a researcher is exploring the relationship
between obesity and severe pneumonia (defined as
pneumonia requiring hospitalization). The research-
ers might use random digit dialing within the hos-
pital catchment area, inquiring about individuals’
weight, height, and previous experience with pneu-
monia. If the literature suggests that age may be a
confounding factor in the association of obesity and
pneumonia, cases and controls could be matched
by age or age range through individual or frequency
matching. The researcher could also address con-
founding by limiting the sample to those within a
narrow age range, say ages 50 to 65.
Hospital- or clinic-based controls represent both
an alternative and a complement to population-
based controls. Here, researchers can select controls
from the hospital or clinic, typically patients admit-
ted or being treated for diseases or health outcomes
other than the one of interest in the study2; the ill-
ness of hospital- or clinic-based controls should also
be unrelated to the exposure explored in the study.
While hospital- and clinic-based controls are often
less costly than population-based controls, they
are also less generalizable to the target population
and far from random.1 Regarding our example of
the relationship between obesity and severe pneu-
monia, ideally control patients should be admitted
with conditions that are independent of exposure—
in this case, obesity.
Is the health outcome of interest related to
exposure? As Gordis notes, “We anticipate that if
the exposure . . . is in fact related to the disease . . . ,
the prevalence of history of exposure among the
cases . . . will be greater than that among the con-
trols.”2 Similarly, if a negative association exists,
one might expect the opposite to be true—namely,
that the incidence of exposure will be lower among
the cases than the controls. Beyond this compar-
ison of raw incidence values, the researcher will
likely want to address the relative magnitude of the
ajn@wolterskluwer.com
effect or the statistical significance of the relation-
ship. Researchers can calculate the disease odds
ratio (defined as the odds of being a case among
the exposed divided by the odds of being a case
among the unexposed) or the exposure odds ratio
(defined as the odds of being exposed among the
cases divided by the odds of being exposed among
the controls).1 While the disease odds ratio and the
exposure odds ratio are conceptually distinct, they
are mathematically equivalent; both can be calcu-
lated as follows1:
Odds ratio = Number of exposed cases × Number of unexposed controls
       Number of unexposed cases × Number of exposed controls
Odds ratios less than 1 indicate a negative associ-
ation between exposure and a given health outcome,
odds ratios equal to 1 suggest no association, and odds
ratios greater than 1 suggest a positive association.
Consider Table 1, which provides data relating
to a hypothetical disease and exposure. Looking at
the table, we can state the following:
• number of exposed cases = 60
• number of unexposed controls = 50
• number of unexposed cases = 40
• number of exposed controls = 50
Using the above equation, we can calculate the
odds ratio as follows:
Odds ratio = (60 × 50) / (40 × 50)
= 3,000 / 2,000
= 1.5
Given that the odds ratio is greater than 1, the
data in this example indicate that disease and expo-
sure are positively associated. Suppose these data
were collected during the COVID-19 pandemic.
Further suppose that the health outcome explored
was “depression” as assessed using the 10-item Kes-
sler Psychological Distress Scale and exposure was
defined as “frontline medical work.” In interpreting
the odds ratio of 1.5, one might infer that frontline
medical work during the pandemic was associated
with increased risk of depression.
Table 1. Two-by-Two Table of Hypothetical Disease and Exposure
Status
Disease No Disease
(Cases) (Controls) Total
Exposed 60 50 110
Unexposed 40 50 90
Total 100 100 200
AJN ▼ February 2022 ▼ Vol. 122, No. 2 53
 NURSING RESEARCH,
STEP BY STEP
While the odds ratio provides some indication
of the relative magnitude of the effect, the above
calculation provides no information on statistical sig-
nificance. More sophisticated statistical methodology
should be considered in assessing whether an asso-
ciation exists (however, such approaches are beyond
the scope of this article). Researchers should consult
biostatisticians or other experts in developing their
evaluation strategy. For instance, in the above exam-
ple, the researcher should interpret the odds ratio
with caution and not assume a positive association
between frontline medical work during the pandemic
and increased risk of depression without incorporat-
ing additional controls and statistical analysis.
Sources of bias. In designing a study, researchers
want to ensure the external and internal validity of
their work. Bias limits the validity of study results.
Selection bias limits the external validity of the
study by limiting the researcher’s ability to general-
ize findings to the target population of interest.
Selection bias occurs when the population of
interest is systematically different from the target
population.6 If the study participants are not rep-
resentative of the target population, then research-
ers will not be able to generalize their findings back
to the target population. Returning to the exam-
ple of severe pneumonia and obesity, suppose the
researcher selected cases and controls from adults
admitted to the ICU. Further suppose that both
cases and controls had a disproportionately higher
number of comorbidities relative to the general adult
population. As a result, the researcher might find it
difficult to generalize the findings of an association
between severe pneumonia and obesity to a popula-
tion of adults without numerous comorbidities.
Information bias poses a simple question: are
you measuring what you think you are measur-
ing? Information bias results from the systematic
mismeasurement or misreporting of data6 and its
presence can limit the internal validity of the study.
While information bias can take many forms, its
major sources in case–control studies concern lim-
itations in recall (that is, participants may not
remember or may misremember an event).2 In the
extreme, recall/rumination bias—systematic misre-
porting or differences in other recollections within
or across groups—may be present. Discrepan-
cies in recall across groups is referred to as differ-
ential recall.2 Suppose a research team is interested
in exploring exposures related to placental abrup-
tion. Women who have experienced a placental
abruption (cases) are far more likely to be aware
of potential exposures that may have triggered this
adverse health event than women who had a rela-
tively standard pregnancy (controls).
54 AJN ▼ February 2022 ▼ Vol. 122, No. 2 ajnonline.com
Broader consideration of confounders. Con-
founding variables distort the observed association
between the exposure and the outcome of inter-
est. To the extent possible, researchers should con-
trol for all known confounders. For instance, in
the example of placental abruption, suppose the
researchers are exploring the relationship between
that event and workplace stress. They believe that
age is correlated with both workplace stress and the
likelihood of placental abruption. To the extent that
age remains unaddressed in the study design, it will
confound (or distort) the researchers’ findings. One
solution might be for the researchers to stratify the
study sample into age groups, such as less than 35
and 35 and older (by definition, a “geriatric preg-
nancy” occurs among women 35 and older). More
broadly, the matching process will help control for
confounding variables, as discussed above.
WHEN IS A CASE–CONTROL STUDY APPROPRIATE?
Case–control studies allow researchers to investi-
gate a range of exposures and can be a useful start-
ing point in analyzing an association between expo-
sure and health outcomes or when limited evidence
on the association is available. For instance, Park
and colleagues argued that the limited literature
regarding medication use and falls among nursing
home residents in Japan made it important to con-
duct their case–control study on the topic.7 While
their findings may not be generalizable to all nurs-
ing home residents, they provided initial evidence
linking medication use and patient falls in Japan.
Case–control studies are also useful when a health
event is rare, a disease has a long latent period, or a
population is difficult to follow.1 For example, given
the relative rarity of the events under investigation,
Wiebe used a case–control study design to explore
whether having a gun in the home was associated
with the incidence of suicide and homicide.8 Wiebe
concluded that having a gun in the home was indeed
a risk factor for both outcomes.
Finally, compared with other study designs,
case–control studies are relatively inexpensive, in
terms of both finances and time. For Hansen and
Stevens, a case–control study offered a convenient
opportunity to explore the association of nursing
night shift work and the incidence of breast can-
cer within the context of an ongoing cohort study
(this is referred to as a nested case–control study).9
Table 2 provides an expanded discussion of these
study examples.7-10
At the opposite end of the spectrum, Lewallen
and Courtright catalog some of the disadvantages
of case–control study designs, including the retro-
spective nature of the data, the potential for bias,
Table 2. Case–Control Study Examples
Article Sample and Time Frame
Hansen, This case–control study was
Stevens (2012)9 part of a larger cohort study of
explored the nurses (a nested case–control
association study).
between shift
work and The data were drawn from
breast cancer interviews with “91,140 female
risk among members of the Danish Nurses
Danish nurses. Association, covering over
95% of nurses in Denmark.”
The study sample was lim-
ited to those women “who
were alive on 1st July 2001,
free of breast cancer, and born
between 1933 and 1970.”
The sample was therefore
younger than 70 when the
interviews took place between
March 2002 and July 2005.
Park and col- The study sample was drawn
leagues (2019)7 from the records of 58 Japa-
explored the nese nursing homes. Only resi-
association dents ages 65+ were included
between medi- in the study. Records were
cation use and drawn from January 1 to
falls among December 31, 2012.
Japanese nurs-
ing home resi-
dents.
Wiebe (2003)8 Data were drawn from sev-
explored the eral large, nationally repre-
association sentative U.S. secondary data
between homi- sources. The study sample
cide, suicide, included adults ages 18+. The
and firearms in author conducted two case–
the home. control analyses, one relating
to homicide and one relating
to suicide.
ajn@wolterskluwer.com
Cases Controls Exposure
Primary case of breast The authors used an Shift work
cancer as identified by incidence density
the national Danish approach “to select The authors focus on
Cancer Registry randomly four live any lifetime shift work
breast cancer–free (≥ 1 year)
controls per case
from the restricted
cohort, who were
matched on year of
birth . . . and inter-
viewed in the same
period as the cases.”
Richardson (2004)10
outlines the inci-
dence density
method in detail.
Residents who experi- Age-, sex-, and Medication use
enced a fall. The authors facility-matched
defined a fall as “uninten- control group The authors defined
tionally coming to rest medication use as fol-
on the ground, floor, or lows: “usage periods were
other lower level other calculated from the dis-
than as a consequence pensing date, and study
of sudden onset of paral- participants were users of
ysis, epileptic seizure, or a medication when the
overwhelming external index date fell within a
force.” dispensing episode.”
Cases were drawn from Controls were drawn Presence of firearms in
the 1993 National Mor- from the 1994 the home
tality Followback Survey National Health Inter-
(NMFS), which drew on a view Survey (NHIS), Specifically, the 1993
sample of 22,000+ death which included infor- NMFS asks: “At any time
certificates from 1993 mation on firearm during the last year
(via the Current Mortalityownership. “Control of life, were there any
Sample). subjects were drawn firearms kept in or
randomly without around [the decedent’s]
Using these data, the replacement from the home? Include those
author identified cases NHIS data and fre- kept in a garage, outdoor
based on the cause of quency matched to storage area, truck or car.”
death listed on the death the case subjects by
certificate (homicide in sex, race . . . , and age The 1994 NHIS asks: “Are
the first case–control group.” any firearms now kept in
study and suicide in the or around your home?
second). Include those kept in a
garage, outdoor storage
area, truck or car.”
AJN ▼ February 2022 ▼ Vol. 122, No. 2 55
 NURSING RESEARCH,
STEP BY STEP
the difficulty selecting strong controls, and potential
issues with record keeping.11
CONCLUSION
Observational research designs, including cross-
sectional, cohort, and case–control studies, are indis-
pensable tools for the nursing researcher. Observa-
tional study designs help nursing researchers further
understand and describe the environment around
them. Among other things, case–control studies pro-
vide the nursing researcher with the opportunity to
retrospectively examine disparities in health out-
comes observed in clinical practice. Results from
observational studies, such as case–control studies,
inform and strengthen nursing leadership through
the generation and dissemination of data-based
knowledge supporting patient care. ▼
Courtney Keeler is an associate professor and Alexa Colgrove Curtis
is assistant dean of graduate nursing and director of the MPH–DNP
dual degree program, both at the University of San Francisco School
of Nursing and Health Professions. Contact author: Courtney Kee-
ler, ckeeler@usfca.edu. Bernadette Capili, PhD, NP-C, is the column
coordinator: bcapili@rockefeller.edu. This manuscript was supported
in part by grant No. UL1TR001866 from the National Institutes of
Health’s National Center for Advancing Translational Sciences Clin-
ical and Translational Science Awards Program. The authors have
disclosed no potential conflicts of interest, financial or otherwise. A
podcast with the authors is available at www.ajnonline.com.
56 AJN ▼ February 2022 ▼ Vol. 122, No. 2 ajnonline.com
REFERENCES
1. Aschengrau A, Seagre GR, III. Case-control studies. In:
Essentials of epidemiology in public health. 4th ed. Burlington,
MA: Jones and Bartlett Learning; 2020. p. 237-66.
2. Gordis L. Case-control studies and other study designs. In:
Epidemiology. 4th ed. Philadelphia: Elsevier Saunders; 2008.
p. 177-200.
3. D’Souza G, et al. Case-control study of human papillomavirus
and oropharyngeal cancer. N Engl J Med 2007;356(19):
1944-56.
4. Cai Q, et al. The mental health of frontline and non-frontline
medical workers during the coronavirus disease 2019
(COVID-19) outbreak in China: a case-control study.
J Affect Disord 2020;275:210-5.
5. Coggon D, et al. Quantifying disease in populations. In:
Coggon D, et al., editors. Epidemiology for the uninitiated.
5th ed. Newark, NJ: John Wiley and Sons; 2003.
6. Aschengrau A, Seagre GR, III. Critical reviews of epidemiological
studies. In: Essentials of epidemiology in public health. 4th ed.
Burlington, MA: Jones and Bartlett Learning; 2020. p. 367-92.
7. Park H, et al. Medications and fall risk: a case-control study
in nursing home residents in Japan. Aging Clin Exp Res
2020;32(5):885-92.
8. Wiebe DJ. Homicide and suicide risks associated with fire-
arms in the home: a national case-control study. Ann Emerg
Med 2003;41(6):771-82.
9. Hansen J, Stevens RG. Case-control study of shift-work and
breast cancer risk in Danish nurses: impact of shift systems.
Eur J Cancer 2012;48(11):1722-9.
10. Richardson DB. An incidence density sampling program for nested
case-control analyses. Occup Environ Med 2004;61(12):e59.
11. Lewallen S, Courtright P. Epidemiology in practice: case-
control studies. Community Eye Health 1998;11(28):57-8.