1. Case- Control Studies Dr Sanjay Jaju Presentation compiled by
Head Specialist
Directorate of Research & Studies
Directorate General of Planning
MOH (HQ)

2. Main Sources of this presentation
Case-Control Studies : Design, Conduct, Analysis
by James J. Schlesselman
Essential Epidemiology: Principles and Applications
by William Oleckno

3. Hierarchy of evidence Meta analysis / Systemic Reviews
Randomized Controlled double blind trials
Non-randomized trials
Prospective cohort studies
Retrospective cohort studies
Case Control studies
Cross sectional studies
Case series
Case reports

4. Important discoveries in case-control studies
Cigarette smoking and lung cancer
Post-menopausal estrogens and
endometrial cancer
Relation of Toxic Shock Syndrome to the use of tampons
Association between use of diethylstilbesterol during the 1st trimester of pregnancy and the development, fifteen to twenty years later of vaginal cancer in the daughters born of these pregnancies

5. Aspirin and Reyes syndrome
AIDS and sexual practices
Vaccine effectiveness
Diet and cancer
Reduced risk of fractures with the use of estrogens
Maternal smoking and congenital malformations
Herpes Simplex virus and Bells palsy
Low dose radiation and leukemia etcetera

7. Definition A type of observational analytical epidemiological
investigation in which the subjects are selected on the
basis of whether they do (cases) or do not (controls)
have a particular disease under study.
The groups are compared with respect to the proportion having a history of an exposure or characteristic of interest.
Case-control study design can be Unmatched or Matched

8. Design of a Case-Control study
Population
Sample **
Outcome present
(Cases)
Outcome absent
(Controls)
Time 1
Time 2
Exposure status?
Exposure status?
** Frequently in case control studies cases and controls are selected separately

9. Determining Exposure Disease onset Initial symptoms Clinical Diagnosis
Study exposure is unlikely to be altered at this stage because of the disease onset.
Therefore, exposure assessment during this period is more likely to reflect the exposure that preceded disease onset.
Study exposure is more likely to be altered at this stage because of the disease symptoms.
Therefore, exposure assessment during this period is less likely to reflect the exposure that preceded disease onset.
Diet & stomach cancer

10. Establishing proper temporal sequence between exposure and outcome
To determine causal association exposure must precede outcome
Study must use incident (new) cases and then assess prior exposure
If one uses prevalent cases
exposure preceded outcome?
NOT SURE WHETHER
outcome preceded exposure?
exposure is related to
exposure and outcome occurred simultaneously?
? Disease development OR
? Disease prognosis

11. Determining incident (new) cases is not always easy
It is not clear when the new cases began !!!!!!
For practical purposes-
Incident cases are defined as newly diagnosed cases (date is taken as proxy)
Acute cases: Date of outcome approximates the date of diagnosis
e.g: automobile related injuries, food poisoning etc
Chronic cases: Date of diagnosis may be years after disease has developed
e.g: multiple sclerosis, emphysema, depression etc

12. Case selection Criteria (reliable and widely acceptable)
Definition of a case must be clear and precise with
strict inclusion / exclusion criteria
Evidence from clinical examination & appropriate diagnostic tests
to prevent dilution with non-cases and hence spurious associations
eg: different diseases having similar signs and symptoms but
unrelated causes.

13. General Population based Specific Population based
Include all Incident Cases in a defined population over specific time period
General Population based
less common due to time and expenses, but feasible in case of population based registries.
(assures representativeness and external validity)
Hospital based
certain hospitals, clinics, nursing homes, rehabilitation centers
Specific Population based
pre-paid health plan, schools,
places of employment, military service etc
affects external validity
but not internal validity

14. Retrospective case selection
All leukemia cases diagnosed for the first time between 2000 and 2012
at a particular medical center.
Advantage
increases power of study if time period is further increased .
Disadvantages
Requires accurate database
Detailed information on exposure
Diagnostic methods may have changed over time
Disease classifications may have changed over time
Prospective case selection
All newly diagnosed leukemia cases from Jan 2014 onwards till Dec 2015
at a particular medical center.
Disadvantages
Expenses
May not accumulate the required number of cases to assure power of the study
Diagnostic methods may change over time
Disease classifications may change over time

15. Selection of Controls
Most difficult and most controversial aspect of study design
The control group provides
a basis for comparison by
representing what
is normal or expected.
The control series is intended to provide an estimate of the
exposure rate that would be expected to occur in the cases
if there was no association between the study disease and exposure.

16. Selection of Controls (contd)
Should come from the same source population and time period that has given rise to cases.
Must be free of the disease under study
Comparable to the cases with respect to confounding factors like age, sex, racial/ethnic background, socioeconomic level, occupational status and certain lifestyle behaviors (this is achieved by Restriction, Matching or Statistical adjustment during analysis)

17. Selection of Controls (contd)
Comparable to the cases regarding to the possibility of having past exposure during the time period of risk. This avoids selection bias.
Restrictions/exclusions for cases must also be applied to controls.

18. Confounding Confounder
Is the distortion in the degree of association between an exposure and an outcome
due to mixing of effects between an exposure and an incidental factor (confounder)
Exposure
Outcome
Confounder
For confounding to occur:
The exposure and incidental factor must be associated
The incidental factor must be a risk factor for the outcome
The incidental factor cannot be an intermediate step between the exposure and
outcome
The incidental factor must be present to a greater or lesser degree in the study
group versus the comparison group.
e.g.: Positive association has been shown between cigarette smoking and motor vehicle injuries. What is the confounder?

19. Methods of Selecting Controls
Population controls
A) Probability sample of General Population from which cases come.
Problems-
Selection bias due to low rates of participation
Measurement bias possibly due to poor recall
B) Probability sample of Specific Population from which cases come
eg: school rosters, selective service lists, insurance company lists, neighbors of cases, friends, schoolmates, siblings, fellow-workers.
Overcomes the above two bias. But, may not represent the true range of exposure that exists in the source population due to too much similarity with cases as regards exposure status.

20. Hospital controls
Persons seeking medical care at the same institution as cases for condition
not related to cases’ disease.
Convenient
Better comparability (except for referral/ tertiary care hospitals)
Better participation
Less recall bias
Problems
May not represent the exposure distribution in the source population as they represent ill people who may be more likely to have an unfavorable risk profile than healthy people in the source population.
Caveat
Hospital controls from diagnostic categories known to be associated with the study exposure must be excluded.
eg: In testing hypothesis that cigarette smoking causes lung cancer, cases with chr. bronchitis, emphysema must not be included as controls. Therefore choose controls from a variety of other diagnosis.

21. Ascertainment of Exposure
· Personal interviews
· Existing records
· Physical measurements and lab tests
Define Exposure
· Yes / No
· Intensity
· Length of exposure
eg: cigarette smoking

22. Accurately determining exposure status
is essential to avoid measurement bias
Interviewer bias: unconscious tendency to search more thoroughly for exposure among cases than controls.
Same assessment for cases and controls is a must.
Verification from other sources: lab records, official records, information from spouses, relatives, colleagues etc
Recall bias: cases often remember past exposures better than controls due to natural tendency to know why the disease occurred.

23. Sample size issues
The results of the case control study will be inconclusive if there is insufficient study power ( which is related to sample size)
Using multiple controls per case will lower the sample size
requirement for the case group but increase the total sample size needed.
Up to 4 controls per case assures adequate study power

24. Advantages of Case-Control Study
Weaknesses
Incidence rates in exposed and unexposed subjects ordinarily cannot be determined
Not appropriate for studying rare exposures
Information on prior exposure or potential confounders may not be readily available, accurate or of the same quality between cases and controls (eg due to differential recall)
It may be difficult to identify comparable case and control groups, and thus there is greater potential for selection bias.
Strengths
Relatively quick and inexpensive to conduct
Appropriate for studying rare outcomes
Require moderate number of subjects
Multiple potential risk factors can be examined in the same study
If properly executed, then cause-effect relationship can be established

25. a b c d Analysis of unmatched case control studies
Controls are NOT pair matched with cases during selection process
Outcome status
Cases
Controls a b c d
Exposed
a + b
Exposure
status
Unexposed
c + d
N = a+b+c+d
a + c
b + d
Odds that cases were exposed a/(a+c)
c/(a+c)
Odds that controls were exposed b/(b+d)
d/(b+d)

26. Odds Ratio Odds ratio = ratio of 2 odds ad / bc
Ratio of odds in favor of exposure among cases to odds in favor of exposure among controls.
a/(a+c)
c/(a+c)
____________
a/c
b/d
ad / bc = =
b/(b+d)
d/(b+d
Significance of OR is given by Chi-square Χ2 = n(ad –bc)2
(a+b)(b+c)(a+c)(b+d)
The 95% confidence interval for the OR must be stated.

27. Case-control Study - Example:
To study the association between smoking and lung cancer
People with lung cancer are enrolled to form the case group, and people without lung cancer are identified as controls.
Researchers then look back in time to ascertain each person’s exposure status (smoking history)

28. Variables Used in Study of Smoking and Lung Cancer
Subject Selection
Control selection
• Age matched
• Healthy individuals
• Patients hospitalized for other cancers
• Patients hospitalized for other diseases
• Deaths from other causes than cancer
• Sampling of general population
• Males and/or females
• Occupational groups
• Hospitalized cases
• Autopsy series
• Total lung cancer deaths in an area
• National sampling lung cancer deaths
Other Variables Concurrently Studied
• Geographic distribution
• Occupation
• Marital status
• Coffee and alcohol consumption
• Other nutritional factors
• Parity
• War gas exposure
• Other pathologic conditions
• Hereditary factors
• Air pollution
• Previous respiratory conditions
Methods of Interviewing
• Mailed questionnaires
• Personal interviews subjects/relatives
• Personal interviews controls
Tobacco-Use History
• Type of smoking
• Amount and type
• Duration
• Inhalation practices

29. Association between smoking and lung cancer
Cases Controls
(with lung cancer) (without lung cancer)
Smokers 33 (a) (b)
Non-smokers 2 (c) (d)
Odds Ratio = ad / bc
= 33*27 /55*2
= 8.1
Interpretation:
Lung cancer is 8 times more likely in smokers compared to non-smokers

30. Matched case control studies
Advantages
Comparability on selected variables
Control selection easy – avoids random sampling
Useful in small studies where obtaining cases and controls that are similar on potentially confounding factors may be difficult
Assures adequate number of cases and controls for statistical comparison
Disadvantages
Difficult to find controls if many variables are being matched
Cannot examine the effects of matched variables
Overmatching due to nearly similar groups
Loss of data as pair has to be eliminated if one subject is non responsive

31. Analysis of matched case control studies
Exposed
Unexposed
Exposed a* b* c* d*
a+b
Case
Unexposed
c+d
a+c
b+d
n* *
Odds ratio = b / c
Significance of OR is given by Chi-square Χ2 = (b – c)2
(b+c)
* represents a pair
** n= ½ the total number of cases and controls in the study

32. Thank you