EPI-546 Block I

Lecture 9: Case-Control Studies

Mathew Reeves, BVSc, PhD

Credit to Michael Collins, MD, MS
and Michael Brown, MD, MS

1
Observational Studies
= Investigator has no control over exposure
• Descriptive
• Case reports & case series (Clinical)
• Cross-sectional (Epidemiological)

• Analytical
• Cohort
• Case-control
• Ecological

2
Objectives - Concepts
• Define and identify case reports and case series
• Define, understand and identify (CCS)
• Distinguish CCS from other designs (esp. retrospective cohort)
• Understand the principles of selecting cases and controls
• Understand the analysis of CCS
• Calculation and interpretation of the OR
• Understand the concept of matching
• Understand the origin and consequence of recall bias
• Example of measurement bias
• Advantages and disadvantages of CCS

3
 4
Grimes DA and Schulz KF 2002. An overview of clinical research. Lancet 359:57-61.
Case Report and Case Series
• Profile of a clinical case or case series which should:
• illustrate a new finding,
• emphasize a clinical principle, or
• generate new hypotheses

• Not a measure of disease occurrence!

• Usually cannot identify risk factors or the cause (no

control or comparison group)
• Exception: 12 cases with salmonella infection, 10 had eaten
cantaloupe

5
Occasionally case reports or case
series become very important…
• Famous Examples:
• A report of 8 cases of GRID, LA County (MMWR
1981)

• A novel progressive spongiform encephalopathy in

Cattle (Vet Record, October 1987)
– Clinical and pathologic findings of 6 cases reported

• Twenty five cases of ARDS due to Hanta-virus,

Four Corners, US (NEJM, 1993)
6
Case-Control Studies (CCS)
• An alternative observational design to identify risk
factors for a disease/outcome.

• Question:
• How do diseased cases differ from non-diseased (controls)
with respect to prior exposure history?

• Compare frequency of exposure among cases and controls

• Effect cause.

• Cannot calculate disease incidence rates because the CCS

does not follow a disease free- population over time

7
 Case-control Study – Design
Select subjects on the basis of disease status

Disease
+ -

Exp + a b

Exp - c d

d1 d0

8
Schulz KF and Grimes DA 2002. Case-control studies. Lancet 359:431-34. 9
Example CCS - Smoking and Myocardial Infarction
Study: Desert island, population = 2,000 people, prevalence of
smoking = 50% [but this is unknown], identify all MI cases that
occurred over last year (N=40), obtain a random sample of N=40
controls (no MI). What is the association between smoking and MI?
•
MI
+ -

Smk + 30 20

Smk - 10 20

40 40

OR = a . d = 30 . 20 = 3.0 (same as the RR!)

c.b 10 . 20 10
Examples of CCS
• Outbreak investigations
• What dish caused people at the church picnic to get sick?
• What is causing young women to die of toxic shock?
• Birth defects
• Drug exposures and heart tetralogy
• New (unrecognized) disease
• DES and vaginal cancer in adolescents
• Is smoking the reason for the increase in lung CA? (1940’s)
– Four CCS implicating smoking and lung cancer appeared in
1950, establishing the CCS method in epidemiology

11
Essential features of CCS design

• Directionality
• Outcome to exposure
• Timing
• Retrospective for exposure, but case ascertainment can be either
retrospective or prospective.
• Rare or new disease
• Design of choice if disease is rare or if a quick “answer” is needed
(cohort design not useful)
• Challenging
• The most difficult type of study to design and execute
• Design options
• Population-based vs. hospital-based

12
Selection of Cases

• Requires case-definition:
• Need for standard diagnostic criteria e.g., AMI
• Consider severity of disease? e.g., asthma
• Consider duration of disease
– prevalent or incident case?

• Requires eligibility criteria

• Area of residence, age, gender, etc

13
Sources of Cases
• Population-based
– identify and enroll all incident cases from a defined population
– e.g., disease registry, defined geographical area, vital records

• Hospital-based
• identify cases where you can find them
– e.g., hospitals, clinics.
• But……
– issue of representativeness?
– prevalent vs incident cases?

14
Selection of Controls
• Controls reveal the ‘normal’ or ‘expected’ level of
exposure in the population that gave rise to the
cases.

• Issue of comparability to cases – concept of the

“study base”
• Controls should be from the same underlying population or
study base that gave rise to the cases?
• Need to determine if the control had developed disease
would he or she be included as a case in the study?
– If no then do not include

• Controls should have the same eligibility criteria as

the cases
15
Sources of Controls

• Population-based Controls
– ideal, represents exposure distribution in the general
population, e.g.,
• driver’s license lists (16+)
• Medicare recipients (65+)
• Tax lists
• Voting lists
• Telephone RDD survey

• But if low participation rate = response bias

(selection bias)

16
Sources of Controls
• Hospital-based Controls
• Hospital-based case control studies used when population-based
studies not feasible
• More susceptible to bias

• Advantages
– similar to cases? (hospital use means similar SES, location)
– more likely to participate (they are sick)
– efficient (interview in hospital)
• Disadvantages
– they have disease?
• Don’t select if risk factor for their disease is similar to the disease
under study e.g., COPD and Lung CA
– are they representative of the study base?

17
Other Sources of Controls
• Relatives, Neighbors, Friends of Cases
• Advantages
– similar to cases wrt SES/ education/ neighborhood
– more willing to co-operate

• Disadvantages
– more time consuming
– cases may not be willing to give information?
– may have similar risk factors (e.g., smoke, alcohol, golf)

18
• Odds of exposure among cases = a / c
• Odds of exposure among controls = b / d

Disease
case control

Exp + a b

Exp - c d

d1 d0

19
Analysis of CCS
The OR as a measure of association

• The only valid measure of association for the CCS is the

Odds Ratio (OR)

• Under reasonable assumptions (– the rare disease

assumption) the OR approximates the RR.

• OR = Odds of exposure among cases (disease)

Odds of exposure among controls (non-dis)

– Odds of exposure among cases = a / c

– Odds of exposure among controls = b / d
– Odds ratio = a/c = a.d [= cross-product ratio]
b/d b.c

20
Example CCS - Smoking and Myocardial Infarction
Study: Desert island, population = 2,000 people, prevalence of
smoking = 50% [but this is unknown], identify all MI cases that
occurred over last year (N=40), obtain a random sample of N=40
controls (no MI). What is the association between smoking and MI?
•
MI
+ -

Smk + 30 20

Smk - 10 20

40 40

OR = a . d = 30 . 20 = 3.0 (same as the RR!)

c.b 10 . 20 21
Odds Ratio (OR)
• Similar interpretation as the Relative Risk

• OR = 1.0 (implies equal odds of exposure - no effect)

• ORs provide the exact same information as the RR if:

• controls represent the target population
• cases represent all cases
• rare disease assumption holds (or if case-control study is undertaken with
population-based sampling)

• Remember:
• OR can be calculated for any design but RR can only be calculated in RCT
and cohort studies
• The OR is the only valid measure for CCS
• Publications will occasionally mis-label OR as RR (or vice versa)

22
Controlling extraneous variables
(confounding)
• Exposure of interest may be confounded by a
factor that is associated with the exposure
and the disease i.e., is an independent risk
factor for the disease

A B

C
23
How to control for confounding
• At the design phase
– Randomization
– Restriction
– Matching

• At the analysis phase

– Age-adjustment
– Stratification
– Multivariable adjustment (logistic regression modeling,
Cox regression modeling)

24
Matching is commonly used in CCS

• Control an extraneous variable by matching

controls to cases on a factor you know is an
important risk factor or marker for disease
• Example:
– Age (within 5 years)
– Sex
– Neighbourhood

• If factor is fixed to be the same in the cases

and controls then it can’t confound

25
Matching
• Analysis of matched CCS needs to account
for the matched case-control pairs
• Only pairs that are discordant with respect to
exposure provide useful information
• McNemar’s OR = b/c
• Conditional logistic regression

• Can increase power by matching more than 1

control per case e.g., 4:1
• Useful if few cases are available

26
 Matched CCS - Discordant pairs
Match 40 controls to 40 cases of AMI so they have the same age and
sex. Then classify according to smoking status.

•
Controls
+ -

+ 32 20
Cases
- 10 18

80

McNemar’s OR = b = 20 = 2.0
c 10 27
Over-matching
• Matching can result in controls being so
similar to cases that all of the exposures are
the same

• Example:
• 8 cases of GRID, LA County, 1981
• All cases are gay men so match with other gay
men who did not have signs of GRID
• Use 4:1 matching ration i.e. 32 controls
• No differences found in sexual or other lifestyle
habits
28
Recall Bias
• Form of measurement bias.
• Presence of disease may affect ability to recall or
report the exposure.
• Example – exposure to OTC drugs during pregnancy
use by moms of normal and congenitally abnormal
babies.
• To lessen potential:
• Blind participants to study hypothesis
• Blind study personnel to hypothesis
• Use explicit definitions for exposure
• Use controls with an unrelated but similar disease
– E.g., heart tetralogy (cases), hypospadia (controls)

29
Other issues in interpretation of CCS

• Beware of reverse causation

• The disease or sub-clinical manifestations of it
results in a change in behaviour (exposure)

• Example:
– Obese children found to be less physical active than
non-obese children.
– Multiple sclerosis patients found to use more multi-
vitamins and supplements

30
CCS - Advantages

• Quick and cheap (relatively)

• so ideal for outbreaks
(http://www.cdc.gov/eis/casestudies/casestudies.htm)

• Can study rare diseases (or new)

• Can evaluate multiple exposures (fishing

trips)

31
Case-control Studies - Disadvantages

• uncertain of E D relationship (esp.

timing)
• cannot estimate disease rates
• worry about representativeness of controls
• inefficient if exposures are rare
• Bias:
• Selection
• Confounding
• Measurement (especially recall bias)

32