case control designs[1]

8/13/2019 Case Control Designs[1]

1/45

Case Control Study Design

EPBI 490


2/45

Case-Control DesignHistorical Perspective

Unique contribution of epidemiology to the

repertoire of clinical research designs

First case-control study performed in late 1950s

Doll and Hills study of lung cancer and smoking

behavior among physicians

Jerome Cornfields classic description of

Retrospective Studies

New statistical tools were developed to analyze

the study design - logistic regression


3/45

Select Study Design to Match the

Research GoalsObjective Design

Description of disease or spectrum Case series or report

Cross-sectional study

Determine operating characteristics

of a new diagnostic test

Cross-sectional

Describe prognosis Cohort study

Determine cause-effect Cohort study

Case-control study

Compare new interventions Randomized clinical trial

Summarize literature Meta-analysis


4/45

Case-Control Designs

Family of epidemiologic study designs

Traditional case-control design

Case-control studies within cohorts

Nested case-control study design

Case-cohort study design

Case-parent study design

Case-only study design


5/45

Case-Control Study

A case-control study is a design in which

individuals with an event or condition of

interest, CASES, are identified and thencompared with regard to one or more

exposures to individuals without the event

or condition of interest, CONTROLS.


6/45

Cohort Study Design

PAR SStudy

Group

Exposed

Unexposed

Outcome

NoOutcome

T

OutcomePAR = Population at Risk

S = Sampling design

T = Elapsed time


7/45

Case-Control Study Design

Cases

Controls

Population

at RiskExposed

Unexposed

Exposed

Unexposed


8/45

Cases

Case-Control Study

How its done

Develop a case definitionIdentify new cases within a specified time period


9/45

Case-Control Study

Selection of Cases Case definition must be pre-specified

Incident cases preferred over prevalent cases inmost settings

If prevalent cases chosen, then risk factors identified for

disease may be those related more to survival with

disease than disease occurrence. Survivorship bias also true for incident cases, but

minimized


10/45

Case-Control Study

Selection of Cases Source of Cases

Hospital or clinic

Because risk factors may result from referral patterns to

specific hospitals, multiple hospitals/clinics often chosen Referral of more ill patients to hospitals, especially tertiary

care centers

Population-based or community New cases reported to health departments, registries, hospital

record departments, etc.

Cases cannot be selected based on known orunknown association with exposure of interest


11/45

Case-Control Study

How its done

Cases

Controls

Define and identify

appropriate controls


12/45

Case Control Study

Selection of Controls Fundamental questions:

Should control subjects be similar to cases in allrespects except for having the disease of interest?

Should control subjects be drawn from the sameunderlying population as the cases, i.e., share the samerisk factors and exposures?

No simple answer to these questions Controls who are identical to the cases in all respects

except disease may underestimate risk factors

Characterizing the underlying population is rarelypossible


13/45

Case Control Study

Selection of Controls The controls should be selected to represent the

person-time distribution of exposure in the source

population Probability of selection as a control is proportional

to person-time in source population

Risk-set sampling

Risk set is the unique set of individuals in the source

population who are at risk for developing disease at the

moment each case is diagnosed


14/45

Case Control Study

Selection of Controls

Time

So

urcePopulat

ion

RiskSet1

RiskSet2


15/45

Case Control StudySelection of Controls

Sources of controls

Hospital control group

Hospitalized patients, best if chosen from the same hospital as

cases in order to control for unknown reference population

Select from all patients admitted to the hospital

Select from specific diagnosis

Community control group

Probability sample best, but not often practical

Select from school rosters, insurance companies, etc.

Neighbors of cases

Random digit dialing

Best friend


16/45

Case Control Study

Selection of Controls Controls cannot be selected based on known or

unknown association with exposure(s) or riskfactors of interest

Multiple controls

Controls of the same type

May improve precision of the measure of association Precision rarely improved with more than 5 controls per case

Controls of Different Types

Hospital controls and community controls per case


17/45

Case-Control Study

How its done

Cases

Controls

Exposed

Unexposed

Exposed

UnexposedFor both cases and

controls determine

previous exposure


18/45

Case-Control StudyAssessing Exposure

Exposure is determined in a retrospective

manner, that is one must look back in time

to assess exposure status before a person

became a case.

Exposure must be measured in a blinded

manner

Data collectors must be unaware of whether

subject is a case or control

Data collectors should be unaware of the study

hypothesis


19/45

Case-Control Study

Assessing Exposure Cases and controls must be assessed for

exposure in the same way


20/45

Case-Control Study

How its done

Cases

Controls

Population

at RiskExposed

Unexposed

Exposed

Unexposed

Ensure that cases and controls

arise from the same population at risk


21/45

Case-Control Study

How its done

Cases

Controls

PAR 1

Exposed

Unexposed

Exposed

Unexposed

PAR 2

Cases and controls may arise from

different underlying populations with different

exposure patterns.


22/45

Case Control StudySelection of Controls

Brain

Cancer

Other

Cancer

Normal

Result

Study 1

ResultStudy 2


23/45

Case-Control Study

Selection of Controls Tuberculosis and cancer study, 1929

Autopsy-based study concluded that TB

protected against cancerControls without cancer at autopsy were

selected

Tuberculosis over-represented in controls as itwas a common reason for hospitalization and

death


24/45

Case-Control StudySelection of Controls

Coffee and pancreatic cancer, MacMahon B et al. NEJM 1981

Coffee consumption was associated with pancreatic cancer

OR 23

Dose-response relationship

Controls were selected from other patients admitted to the

hospital by the same physician as the case, often

gastroenterologist

This specialist would admit patients with other diseases

(gastritis or esophagitis) for which he or the patient would

reduce coffee intake

Controls intake of coffee may be less than population - not

representative of source population


25/45

Case-Control Study

How its done

Cases

Controls

Population

at RiskExposed - b

Unexposed - d

Exposed - a

Unexposed - c


26/45

Case-Control Study

How its doneCase Control

Exposed

Unexposed

a b

c d

a + c

a + b

c + d

b + d N

Odds Ratio = ad/bc

OR is the odds

of exposure given

disease divided

by the odds ofexposure given

no disease


27/45

Case Control Study

Interpretation The power of the study design lies in the

symmetry of the OR.

Remember that the odds of exposure among

cases compared with controls is the same as

the odds of disease among exposed andunexposed.


28/45

Mo.

Estrogen

Replacement

Cases

Trauma

Controls

Non-trauma

Controls

- - - - - N (%) - - - - -< 6

6

80

14

(85)

(15)

60

20

(75)

(25)

579

213

(73)

(27)

Total 94 (100) 80 (100) 792 (100)

ORtc= 1.90

Ornc = 2.10

Hip fracture among women according to the number

of months of estrogen replacement therapy, 1977 - 1979

Kreiger et al., 1982.


29/45

Case-Control Study

InterpretationCase Control

Exposed

Unexposed

a b

c d

a + c

a + b

c + d

b + d N

OR =

a

cb

d

Any procedure that distorts the ratio of exposure

in either the cases or controls may lead to bias.


30/45

Case-Control Study

Bias Case-control studies are subject to bias and

confounding, both will distort the results of the

study Bias is defined as the deviation of results, or

inferences, from the truth, or processes leading to

such deviation.

There are about 75 different types of bias now

identified in published case-control studies


31/45

Bias

Selection bias

Information bias

Confounding


32/45

Selection Bias

A distortion in the relationship between exposureand outcome that results from selection of study

participants

The relation between exposure and outcome is

different for those who participate and those whodo not participate but would theoretically beeligible for the study.

Examples

Self-selection bias

Diagnostic bias


33/45

Information Bias

A distortion in measuring exposure or

outcome data that results in different quality

(i.e., accuracy or reliability) or frequency ofinformation between comparison groups.

Differential misclassification

Non-differential misclassification


34/45

No Misclassification

X-ray

Exposure

No X-ray

Exposure

Breast Cancer

No Breast Cancer

40

9,960

80

39,920

Total 10,000 40,000

OR = 2.00

Classification of exposure comparable in

cases and controls, perfect accuracy


35/45

Non-differential Misclassification

X-ray

Exposure

No X-ray

Exposure

Breast Cancer

No Breast Cancer

60

19,940

60

29,940

Total 20,000 30,000

OR = 1.50

Tendency to overestimate exposure, but

comparable between cases and controls


36/45

Differential Misclassification

X-ray

Exposure

No X-ray

Exposure

Breast Cancer

No Breast Cancer

40

19,940

80

29,940

Total 19,980 30,020

OR = 0.75

Assessment of exposure different

between cases and controls


37/45

Case-Control Study

Bias Berksons bias

The combination of exposure and disease lead to

increased likelihood of being hospitalized

Cases more likely to be exposed

Recall bias

Differential recall of exposure between cases and

controls in a study

Example

Mothers of children with congenital malformations may

remember details about possible exposures during

pregnancy that mothers without malformations forget.


38/45

Confounding

A situation in which a measure of the effect of anexposure on risk is distorted because of theassociation of exposure with other factor(s) that

influence the outcome under study. Criteria for confounding

Factor is associated with exposure

Factor is associated with disease in the absence of

exposure Factor is not in the causal path between exposure and

outcome


39/45

Case-Control Study

Matching Matching is defined as the process of selecting

controls so that they resemble the cases withregard to certain characteristics

The goal of matching is to create similardistributions between cases and controls withregard to certain characteristics

Matching can be used to

Adjust for potential confounding factors

Increase precision of estimate


40/45

Case-Control Study

Matching Types of matching

Individual level matching

For each case in the study, one or more controls areselected with identical (similar) characteristics as the

case

Frequency, or group, matching

Select controls so that the proportion with a certaincharacteristic is identical to the proportion of cases

with that characteristic


41/45

Case-Control Study

Problems with Matching Difficult and expensive

Cannot evaluate the effect of controlled

variables May limit the ability to control for other

variables

OvermatchingControls resemble cases in terms of known and

unknown characteristics, some of which may beassociated with the disease


42/45

Case-Control Study

Analytic Strategy Assess relationship/association between

Exposure and independent variables

Case/Control status and independent variables

Calculate crude, or unadjusted, OR for

exposure - case associationMatched analysis required for matched studies


43/45

Case-Control StudyMatched Analysis

Case

Control

Exposed

Unexposed

a b

c d

Exposed Unexposed

OR = b/c

Case-control pairs that share the same exposure

status do not contribute to the estimate of risk.


44/45

Case-Control Study

Analytic Strategy Stratified analysis Calculate stratum-specific ORs for exposure-case

relationship

Determine presence of confounding and interaction

Logistic Regression analysis

Regression technique used to adjust for confounding

and interaction

Special logistic model applied in matched studies


45/45

Case-Control Study Design

Strengths and WeaknessesStrengths

Rare disease

Long latency betweenexposure and disease

Explore multiple

hypotheses

Inexpensive

Weaknesses

Prone to bias

Temporal relationshipscannot be established

Inefficient for rare

exposures, unless

exposure often lead to

disease

case control designs[1]

Documents