Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology...

46
Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego State University

Transcript of Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology...

Page 1: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Case-Control Studies(retrospective studies)

Sue Lindsay, Ph.D., MSW, MPH

Division of Epidemiology and Biostatistics

Institute for Public Health

San Diego State University

Page 2: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Case Control Study Design

Exposed to risk factor

Not exposed to risk factor

Exposed to risk factor

Not exposed to risk factor

Cases:

With Disease

Controls:

No Disease

Source

Population

select select

Page 3: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Case-Control Study Design

• The hallmark of the case-control study design is that it begins with cases and compares them with non-cases (controls).

SELECT CASES

AND CONTROLS

ASSESS

EXPOSURE

SOMETIME

IN

THE PAST

Start here

Page 4: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Design Considerations

• These investigations are initially oriented to disease status

• The objective is to compare the odds of exposure among persons with the disease to the odds of exposure among persons without the disease

• You need a well-defined source population

• How well can you identify individuals with the disease? Case identification should be as complete as possible within the source population.

• The sample of cases should be representative of all cases.

Page 5: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Design Considerations

• The sample of controls should be representative of the general population.

• Can you accurately detect exposures to your risk factor?

• When possible verify exposures by multiple methods: interview, medical record review, blood test etc.

• How should you select your cases? How should you select your controls?

• Selection of cases or controls should not be influenced by prior exposure to the risk factor.

Page 6: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Steps in Conducting Case-Control Studies

1. Select source population

2. Identify and select cases

3. Identify and select controls• Match cases to controls

• Group matching or individual matching?

4. Measure exposure in cases and controls

5. Compare odds of exposure in diseased to odds of exposure in non-diseased persons.

Page 7: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Sources of Case Selection

• Population-based case-control studies• Surveillance systems

• Patients identified by:• Physician practices

• Clinics

• Registries

• Hospitals

• Hospital-based case-control studies• Cases admitted to a hospital or hospitals

Page 8: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Issues in Case Selection

• Are the cases selected representative of all cases in the community?

• Are there institutional or hospital differences which may affect the study?

• Are there physician practice differences that may affect the study?

• Should you use incident cases or prevalent cases?

Page 9: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Incident or Prevalent Cases?

• Must be able to identify new cases

• Survivorship/risk bias less of a problem

• Early deaths will still be excluded

• More cases available

• May over-represent survivors

• Risk factors may be associated with survivorship

Incident Cases

Prevalent Cases

Page 10: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Characteristics of Controls

• Should be from source population

• Should be representative of general population, or at least the source population

• Should be comparable to cases except on risk factor

• Random selection when possible

• Selected independently of exposure

• Should be from same sampling time frame

• Should be “at risk” for being categorized as a case

Page 11: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Sources of Population-Based Controls

• Random sample of total population

• Random sample from source population

• Neighborhood controls (random households)

• Primary care clinics, private practice offices

• Other diseases – registries

• Friends

Page 12: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Hospital-Based Controls

• Captive population

• Poorly defined reference population

• Not comparable to general community • Possibly older, sicker, risk factor differences

• Use a sample of all other patients admitted?

• Select specific diagnoses for control group?

• What diseases to include and exclude in the control group?

Page 13: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Selection Bias

Disease

No Disease

ExposedNot Exposed

Selection bias stems from an absence of comparability between the two

groups being studied (cases and controls).

Page 14: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Misclassification Bias

Disease

No Disease

ExposedNot Exposed

Incorrect determination of exposure or outcome or both.

Non-differential misclassification bias

Differential misclassification bias

Diagnostic suspicion bias particularly challenging

Page 15: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Case-Control 2 X 2 Table

a

Cases Controls

Exposed (+)

Not Exposed (-)

b

c d

a + c b + d

First Select

Then Classify Exposure

Page 16: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Case-Control Analysis

• In case-control studies we cannot calculate risk or incidence: therefore we cannot calculate relative risk as we can in cohort studies

• Instead, calculate the Odds Ratio (OR). Based on the concept of relative odds of disease

Page 17: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Case-Control Analysis

Odds of case exposure Odds of control exposure

Proportion cases exposed Proportion controls exposed

Proportion cases not exposed Proportion controls not exposed

Page 18: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

The Case-Control 2 X 2 Table

a

Cases Controls

Exposed (+)

Not Exposed (-)

b

d

Proportions Exposed a/a+c b/(b+d)

c

Proportions Not Exposed c/a+c d/(b+d)

Page 19: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

The Odds of Case ExposureThe Odds of Control Exposure

a

Cases Controls

Exposed (+)

Not Exposed (-)

b

c d

Odds of case exposure: a/(a+c) = a

c/(a+c) c

Odds of control exposure: b/(b+d) = b

d/(b+d) d

Page 20: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

The Odds Ratio in a Case Control Study

Odds

Ratio

Odds of case exposure

Odds of control exposure

OR =a/c

b/d

=

=ad

bc

a b

c d= cross-product ratio

Page 21: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Case-Control Study of CHD and Smoking

112

CHD Cases Controls

Smoking (+)

No Smoking (-)

176

88 224

OR = (112 x 224) = 1.62

(88 x 176)

The odds that a patient with CHD was exposed to smoking is 1.62 times greater than a patient without CHD.

Page 22: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Interpretation of Odds Ratio Estimates

• If OR = 1: Risk in Exposed = Risk in Non-exposed (No Association)

• If OR > 1: Risk in Exposed > Risk in Non-exposed (Positive Association)

• If OR < 1: Risk in Exposed < Risk in Non-exposed (Protective Association)

Page 23: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Another way to look at the Odds Ratio

Cases Controls

Exposed (+)

Not

Exposed (-) c

ba

d

The OR can be viewed as the ratio

of the product of the 2 cells that

support the hypothesis, cells a and d,

(diseased people exposed and

non-diseased people unexposed)

to the product of the 2 Cells

that negate the null hypothesis of association,

cells b and c, (exposed non-diseased

people and unexposed diseased people)

Page 24: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Case-Control Odds Ratio: An Estimation of Relative Risk

• Case- control Odds Ratios can be used to estimate Relative Risk if the following conditions are met:

• The controls are representative of the general population

• The cases are representative of all cases

• The frequency of the disease in the population is small

a b

c d

RR= a/(a+b)

c/(c+d)

If a is small in relation to b

If c is small in relation to d=

a/b

c/d=

ad

bc

Exposed

Not Exposed

Cases Controls

Page 25: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

A Rare Disease

45

Cases Controls

Exposed (+)

Not Exposed (-)

4955

29 4971

Relative Risk = (45/5000)/(29/5000) = 1.55

5000

5000

Odds Ratio = (45 x 4971)/(29 x 4955) = 1.56

Page 26: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

A Common Disease

4500

Cases Controls

Exposed (+)

Not Exposed (-)

500

2900 2100

Relative Risk = (4500/5000)/(2900/5000) = 1.55

5000

5000

Odds Ratio = (4500 x 2100)/(2900 x 500) = 6.52

Page 27: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Problems with Selections of Controls: An Example Using Coffee and

Pancreatic Cancer

• MacMahan, 1981, case-control study of pancreatic cancer

• Cases drawn from 11 Boston and Rhode Island hospitals - histologically confirmed pancreatic cancer

• Controls selected from same hospitals, admitted by the same physician as each case

• The association between coffee drinking and pancreatic cancer was not the main hypothesis of the study

Page 28: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Odds Ratio in Men

207

Pancreatic Cancer Controls

Coffee drinking

No coffee

275

9 32

Men

OR = (207 x 32)/(275 x 9) = 2.68

Page 29: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Odds Ratio in Women

140

Pancreatic Cancer Controls

Coffee drinking

No coffee

280

11 56

Women

OR = (140 x 56)/(280 x 11) = 2.55

Page 30: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Biased Control Selection

• Controls were patients hospitalized at the same time by the same physician who hospitalized the cases

• Easier to obtain physician cooperation and control participation

• Most admitting physicians were gastroenterlogists

• Gastroenterologists were more likely to admit control patients with other GI disorders

• Patients with serious GI disorders were less likely to consume coffee

Page 31: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Odds Ratio in Women

140

Pancreatic Cancer Controls

Coffee drinking

No coffee

280

11 56

Women

1. The percent of controls reporting coffee drinking was less than expected

3. Controls were not representative of the general population

2. The percent of controls reporting no coffee drinking was greater than expected

Page 32: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Recall IssuesCan subjects remember exposure accurately?

• Recall Limitations• Subject has incorrect information, forgets, does

not have knowledge

• Recall Bias• Selective recall by cases

• Differential recall between cases and controls

Page 33: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Matching in Case-Control Studies

• Purpose: To control for confounding

• Confounder:

• A known risk factor for your disease of interest

• Also associated with your risk factor

• Distorts the association between your risk factor and disease

• Matching: Selects controls so that they are similar to the cases on confounding variables: age, sex, ses, etc.

• Increases statistical precision of estimates allowing smaller sample size

Page 34: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Types of Matching in Case-Control Studies

• Group Matching

• Individual Matching

• Match by frequency or proportion of a selected characteristic

• Pair-wise matching, each case is paired with a similar control

Page 35: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Examples of Types of Matching

• Group Matching

• Individual Matching

• 25% of cases married, controls selected to be 25% married

• Case is a 45 year old Caucasian woman, control is selected who is also a 45 year old Caucasian woman

Page 36: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Problems With Matching in Case-Control Studies

• Practical• Attempting to match on too many characteristics

• Time consuming

• Cases who are not successfully matched must be discarded from the analysis

• Analytical• When controls are matched to cases on a given characteristic, that

characteristic cannot be studied as an independent risk factor for the disease

• Do not match on a characteristic you are interested in studying!!

Page 37: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Practical Problems with Matching

• Match on age, sex, race, marital status, number of children, zip code

• Can you find a control who is a 35 year old Caucasian male, married, 4 children in zip code 92123?

Page 38: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

General Guidelines for Matching in Case-Control Studies

• Only match on variables that are known risk factors for your disease of interest.

• Do not match on variables whose relationship with the disease needs to be studied

• Beware of unplanned matching and overmatching

Page 39: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Oral Contraceptives and Cancer: An Example of Unplanned Matching

a

CancerBest-Friend

Controls

Contraceptive use

No contraceptive use

b

c d

The % of controls

reporting OC use

Is likely to be greater

than expected

Best friends share lifestyle characteristics with cases which will affect any association that is observed

Page 40: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Analysis of a Case-Control Study with Pair-wise Matching

W

Control

Exposed

Control

Not Exposed

Case exposed

Case not exposed

X

Y Z

W and Z are concordant pairs, X and Y are discordant pairs

OR = X

Y

Page 41: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Example of a Case-Control Study with Pair-wise Matching

• Antunes, 1979, case-control study of endometrial cancer

• Baltimore hospitals: 1973-1977

• Research Question: Is there an association between estrogen use and endometrial cancer?

• Selected cases with Stage 1 tumors

• Pair-wise matched with controls by hospital, race, and age

Page 42: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Estrogen Use and Endometrial Cancer

17

Control

Used Estrogen

Control

No estrogen

Case: used estrogen

Case: no estrogen

76

10 111

OR = 76 = 7.6

10

Page 43: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Use of Multiple Controls

• Controls of the same type

• Controls of different types

• Multiple controls per case will increase the statistical power of your study

• Up to case-control ratios of approximately 1:4

Page 44: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

When to Use Multiple Controls of Different Types?

• A single control group may be biased in some way

• A hospitalized control group is non-representative of the community

• Neighborhood or best-friend controls are overmatched

• Can learn more about the disease process

Page 45: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Multiple Controls of Different Types: Prenatal Radiation and Brain Tumors in Children

Children with

brain tumors

Children

with no cancer

Children with

other types of

cancer

Cases Normal Controls Cancer Controls

Page 46: Case-Control Studies (retrospective studies) Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego.

Radiation and Brain Tumors

0

5

10

15

20

25

BrainTumors

OtherCancer

NormalControls

% Radiation Exposure

• Prenatal radiation is a risk factor specifically for brain tumors (not all cancers)

Is there recall bias?