Introduction to Study Design: Case Control and Cohort Studies

Martin L. Lesser, Ph.D.

Biostatistics Unit

Feinstein Institute for Medical Research North Shore – Long Island Jewish

Health System

Copyright 2000-2002

CME Disclosure Statement• The North Shore LIJ Health System adheres to the ACCME's new

Standards for Commercial Support. Any individuals in a position to control the content of a CME activity, including faculty, planners and managers, are required to disclose all financial relationships with commercial interests. All identified potential conflicts of interest are thoroughly vetted by the North Shore-LIJ for fair balance and scientific objectivity and to ensure appropriateness of patient care recommendations.

• Course Director, Kevin Tracey, has disclosed a commercial interest in Setpoint, Inc. as the cofounder, for stock and consulting support. He has resolved his conflicts by identifying a faculty member to conduct content review of this program who has no conflicts.

• The speaker, Martin L. Lesser, PhD, has no conflicts.

2

Case Control Studies

4

Sequence of Epidemiologic and Clinical Research

Anecdotal Observation

Case-Control Study

and/orCohort Study

Clinical Trial

Copyright 2000-2002

5

The Basic Idea

A Case-Control Study is an observational study that begins with cases of disease and compares them to controls without disease with respect to exposure to a factor. 1. Hypothesize that exposure (E) to a certain factor

increases an individual's risk of developing the disease (D) or condition being studied

2. Choose a sample of people with the disease or condition

being studied (cases) 3. Choose a sample of people without the disease WHO

ARE AS SIMILAR TO THE CASES IN ALL OTHER WAYS (controls)

4. Compute the proportion of cases and controls who were

"exposed" to the hypothesized risk factor 5. Compare the two proportions. (In actuality, we compare

the "odds" of being exposed.)

Copyright 2000-2002

6

Features of Case-Control Studies

1. Observational - no intervention is involved

2. Retrospective - relies on recall and documentation ofevents that occurred prior to selection of cases andcontrols

3. Calculates odds ratio as a measure of risk

Copyright 2000-2002

7

Selecting Cases

1. Definition of cases - homogeneity of disease entity; strictdiagnostic criteria

2. Sources of cases - hospitals, medical practices; diseaseregistries; general population

3. Prevalent vs. incident cases (existing vs. newly diagnosed cases)

Copyright 2000-2002

8

Selecting Controls

Controls must be selected to represent the population of individuals who would have been identified and included as cases had they also developed the disease.

1. Definition of controls - free of the disease in question; similar to

cases in all other respects 2. Sources of controls - hospital, medical practice; community-at-

large, community organizations; workforce; department of motor vehicles; friends, relatives, neighbors

3. Matching cases and controls...

Copyright 2000-2002

9

Confounding

A factor F is called a "confounding variable" if F is associated with the exposure (E) and F independently is associated with the risk of developing the disease (D). ** Important: To be a confounder, F, must satisfy both of the above conditions. Matching cases and controls on F is the solution to confounding.

Copyright 2000-2002

10

Situations in Which Matching on F is Proper

E E

D D

F F

Situation A Situation B

A: Indirect association between E and D E=alcohol use D=lung cancer F=smoking

B: E and F are both risk factors and are associated with each otherE=OC use D=myocardial infarction F=smoking

Reference: Schlesselman, 1982

Copyright 2000-2002

11

E E

D D

F F

Situations in Which Matching on F is Unnecessary

Situation C Situation D

C: F is associated with D independently of E, but F is not associated with EE=blood group O D=venous thromboembolism F=age

D: F is associated with E, but is not a risk factor for DE=regular exercise D=(lack of) MI F=fluid intake

Reference: Schlesselman, 1982

Copyright 2000-2002

12

Logistics of Matching

1. Specification of matching criteria (categorical, “calipers”, etc.)

2. Individual (1:1, 1:n) vs. frequency matching

Copyright 2000-2002

13

Propensity Score Matching

Copyright 2000-2002

• A method of matching subjects in treatment and control groups with similar propensity scores (i.e. the propensity to receive treatment based upon a set of certain covariates).

• Goal: Find subjects with the same a priori likelihood of being in the treatment group.

• Reduces bias in measured characteristics only

)covariates|treatment Pr(receive )(PS Score Propensity

14

Propensity Score Matching

Copyright 2000-2002

1. Use logistic regression to model the probability of being in the treatment group. Include any covariates that may be related to receiving treatment.

2. Predict the probability of being in the treatment group (propensity score).

3. Match subjects with the “same” propensity score (one in the treatment group, one in the control group). ** Don’t have to match exactly

Caliper’s matching

4. Use standard analyses for matched data

5. Calculate standardized differences after matching to check for balance. |Stzd D| > 10% indicates poor balance.

0.02)PS (i.e.

2

/100 22controltreatment

controltreatment

ssxxDStzd

15

Sources of Bias

1. Recall bias (cases are more (or less) likely to recall exposure than controls)

2. Reporting accuracy bias (cases or more (or less) likely to report accurately)

3. Selection bias (cases are not representative of those with disease)

4. Detection bias (exposed subjects are more (or less) likely to be screened for disease)

5. Referral bias (controls are referred to the health care system for an “unrelated”

health problem, that is actually related to the disease under study)

Copyright 2000-2002

16

Cases Controls

Exposure Yes

a b ab

Exposure

No

c d

cd

Total ac bd abcd

Odds of exposure in "D=yes" group = [a/(a+c)] / [c/(a+c)] Odds of exposure in "D=no" group = [b/(b+d)] / [d/(b+d)] Odds ratio (OR) = (ad) / (bc)

Computing the Odds Ratio

Copyright 2000-2002

17

Interpreting The Odds Ratio

OR is always greater than zero (unless there is a zero cell count) OR > 1 means that E increases risk of D OR = 1 means that E is not associated with D OR < 1 means that E is protective against D

Copyright 2000-2002

18

Odds Ratio: Example

Does oral contraceptive use increase the risk of myocardial infarction (heart attack)?

Myocardial Infarction

Yes No

Current Yes 23 304 OC Use No 133 2816 OR = 23x2816 = 1.6

304x133

Women who are current OC users have 1.6 times greater risk of MI than non-users

Copyright 2000-2002

19

Establishing Causality

1. Temporal sequence 2. Consistency of association 3. Strength of association 4. Dose-response relationship 5. Biological plausibility

Copyright 2000-2002

20

Advantages and Disadvantages of Case-Control Designs

Advantages

1. Well suited to the study of rare diseases or those with long latency. 2. Relatively quick to mount and conduct. 3. Relatively inexpensive. 4. Requires comparatively few subjects. 5. Existing records can occasionally be used. 6. Little or no risk to subjects. 7. Allows study of multiple potential causes of a disease.

Disadvantages

1. Relies on recall or records for information on past exposures. 2. Validation of information is difficult or sometimes impossible. 3. Control of extraneous variables may be incomplete. 4. Selection of an appropriate comparison group may be difficult. 5. Rates of disease in exposed and unexposed individuals cannot be determined. 6. Method relatively unfamiliar to medical community and difficult to explain. 7. Detailed study of mechanism is rarely possible. Reference: Schlesselman, 1982.

Copyright 2000-2002

Case-Control vs. Cohort Study

Rare diseases

Diseases with long latency

Data quality

Relatively inexpensive

Estimate incidence rate of disease

Little or no risk to subjects

Requires comparatively fewer subjects

Describe natural history of diseases

Relatively quick to mount and conduct

Allows study of multiple potential causes of a disease

Allows study of multiple potential effects of an exposure

Case Control

Cohort

(prospective)

Copyright 2000-2002

Cohort Studies

Sequence of Epidemiologic and Clinical Research

Anecdotal Observation

Case-Control Study

and/orCohort Study

Clinical Trial

Copyright 2000-2002

The Basic Idea

A Cohort Study is an observational study that begins with subjects exposed and subjects not exposed to a potential risk factor with follow-up over time to determine and compare the incidence of disease in the exposed vs. unexposed.

1. Hypothesize that exposure (E) to a certain factor

increases an individual's risk of developing the disease (D) or condition being studied

2. Choose a sample of people who are or have been

"exposed" to the factor 3. Choose a sample of people who have not been

exposed to the factor, WHO ARE AS SIMILAR TO THE CASES IN ALL OTHER WAYS

4. Compute the incidence of disease in the exposed

and unexposed groups 5. Compare the two incidence rates

Copyright 2000-2002

Features of Cohort Studies

1. Observational - no intervention is involved

2. Prospective or Retrospective - depending on the temporalrelationship between the initiation of the study and theoccurrence of the disease; in both designs, subjects arefollowed through time

3. Calculates relative risk as a measure of risk

Copyright 2000-2002

Selecting Exposed Subjects

1. Definition of exposure - exposure status must be well-defined and, preferably, measurable

2. Sources of exposed subjects - people in certain occupations;

persons living near environmental hazards; persons with certain lifestyles; persons who have utilized a certain drug or chemical; persons who participate in certain activities

3. Cohorts should also be selected for their ability to facilitate the

collection of relevant data. Examples: doctors, nurses, workers in a company, union members, students

Copyright 2000-2002

Selecting Unexposed Subjects

1. Definition of unexposed - free of the exposure in question; similar tothe exposed cohort in all other respects

2. Sources for the unexposed comparison group - same as forexposed; external source

Copyright 2000-2002

Matching

1. Matching is not often carried out in cohort studies due to sample size,expense, and logistics

2. Statistical adjustments can be made for confounding variables

Copyright 2000-2002

Bias and Other Problems

1. Exposure misclassification bias 2. Change in exposure level over time 3. Loss to follow-up 4. Non-participation bias 5. Reporting bias 6. Healthy worker effect

Copyright 2000-2002

Sources of Exposure Information

1. Pre-existing records

2. Interviews, questionnaires

3. Direct physical examination or tests

4. Direct measurement of the environment

5. Daily logs

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

advantages and disadvantages

Copyright 2000-2002

Sources of Outcome Information

1. Death certificates

2. Physician and hospital records

3. Disease registries

4. Self-report

5. Direct physical examination or tests

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

advantages and disadvantages

Copyright 2000-2002

Computing the Relative Risk

Incidence of Disease in Exposed Group = a/(a+b) Incidence of Disease in Unexposed Group = c/(c+d) Relative risk = RR = [a/(a+b)]/[c/(c+d)] = a(c+d) / [c(a+b)]

Disease Yes No

Exposure Yes

a b ab

Exposure

No

c d

cd

Total ac bd abcd

Copyright 2000-2002

Interpreting the Relative Risk

RR > 0

RR > 1 means that E increases risk of D

RR = 1 means that E is not associated with D

RR < 1 means that E is protective against D

Copyright 2000-2002

Relative Risk: Example

Does oral contraceptive use increase the risk of thrombophlebitis?

Thrombophlebitis

Yes No

Yes 30 970 OC

Use No 3 997 RR = .03 = 30x1000 =10

.003 3x1000

Women who use OC have 10 times greater risk of thrombophlebitis than non-users

1000

1000

Copyright 2000-2002

Advantages and Disadvantages of the Cohort Method

Advantages

1. In principle, provides a complete description of experience subsequent to exposure, including rates of progression, staging of the disease, and natural history.

2. Allows study of multiple potential effects of a given exposure, thereby obtaining information on potential benefits as well as risks.

3. Allows for the calculation of rates of the disease in exposed and unexposed individuals.

4. Permits flexibility in choosing variables to be systematically recorded. 5. Allows for thorough quality control in measurement of study variables.

Disadvantages

1. Large numbers of subjects are required to study rare diseases. 2. Potentially long duration for follow-up. 3. Current practice, usage, or exposure to study factors may change, making

findings irrelevant. 4. Relatively expensive to conduct. 5. Maintaining follow-up is difficult. 6. Control of extraneous variables may be incomplete. 7. Detailed study of mechanism is rarely possible. Reference: Schlesselman, 1982.

Copyright 2000-2002

Cohort vs. Case-Control Study

Rare diseases

Diseases with long latency

Data quality

Relatively inexpensive

Estimate incidence rate of disease

Little or no risk to subjects

Requires comparatively fewer subjects

Describe natural history of diseases

Relatively quick to mount and conduct

Allows study of multiple potential causes of a disease

Allows study of multiple potential effects of an exposure

Cohort

(prospective)

Case-Control