Case -Control Studies...National University of Health Sciences EBP@NUHS Essential Evidence Based...

NationalUniversity of Health Sciences

EBP@NUHS Essential Evidence Based Practice Study Guide Critical Appraisal Study Design Case Control © J. Cambron 2008

CaseCase--Control StudiesControl Studies

Jerrilyn Cambron, DC, PhDJerrilyn Cambron, DC, PhD

Professor, Department of ResearchProfessor, Department of Research




Hierarchy of scientific evidence: observationalHierarchy of scientific evidence: observational

Meta-Analysis,

Systematic Reviews

Best Evidence / Evidence Guidelines

Randomized, controlled trials (RCT)

Clinical trials, Cohort Studies,Case Control

Case series

Case study / case report

Animal studies, in vitro studies

Expert opinions, editorials



CaseCase--control studiescontrol studies

�� Start with subjects who are: Start with subjects who are:

�� Cases (with disease) Cases (with disease)

�� Controls (without disease)Controls (without disease)

�� Look backwards in time Look backwards in time (retrospective)(retrospective)

�� Compare past histories of possible risk factors between Compare past histories of possible risk factors between

cases and controlscases and controls



CaseCase--control studiescontrol studies

�� Observational study starting with persons who have a Observational study starting with persons who have a

disease (case) and who do not have the disease (control), disease (case) and who do not have the disease (control),

and comparing histories of exposure to possible risk and comparing histories of exposure to possible risk

factors.factors.

�� The goal is to assess the relationship of potential risk The goal is to assess the relationship of potential risk

factors to a specific disease.factors to a specific disease.

�� Easiest study design to assess Easiest study design to assess rare diseasesrare diseases..



CaseCase--control vs. Cohort studiescontrol vs. Cohort studies

�� Calendar time is NOT the characteristic Calendar time is NOT the characteristic

that distinguishes these two designsthat distinguishes these two designs

�� Rather, Rather, directiondirection of calendar time is of calendar time is importantimportant



Study timeline differencesStudy timeline differences

Risk factor COHORT Disease or not???

Risk factor??? CASE CONTROL Disease or not

(Prospective)

(Retrospective)



Nested case control studyNested case control study

�� Study design where cases and controls are chosen from a Study design where cases and controls are chosen from a

population that was defined before the study beganpopulation that was defined before the study began

�� Still starts with cases and controls and looks back at risk Still starts with cases and controls and looks back at risk

factorsfactors

�� Can be nested in a cohort study or general population as Can be nested in a cohort study or general population as

long as data were already prospectively being collected long as data were already prospectively being collected

within the groupwithin the group



Case selectionCase selection

�� Case definition must be very clearCase definition must be very clear

�� Incident (new) cases are most desirable so case can Incident (new) cases are most desirable so case can

remember past exposure to risk factorsremember past exposure to risk factors

�� Cases may come from hospitals, disease registries, Cases may come from hospitals, disease registries,

doctordoctor’’s offices, insurance records, surveillance, s offices, insurance records, surveillance,

patient groups, etc.patient groups, etc.

�� Ex: Cases might be all new cases of ovarian cancer diagnosed in Ex: Cases might be all new cases of ovarian cancer diagnosed in

Cook County 2000Cook County 2000--2005 in women aged 212005 in women aged 21--79 years. We may 79 years. We may

look at Cook Co. area hospitals or tumor registries.look at Cook Co. area hospitals or tumor registries.



Control selectionControl selection

�� Controls are people at risk of developing the Controls are people at risk of developing the disease but do not have the diseasedisease but do not have the disease

�� Selecting the controls is the key challenge for Selecting the controls is the key challenge for case control studies, and this is the major case control studies, and this is the major determinant of whether the conclusions are determinant of whether the conclusions are validvalid

�� Multiple controls can be chosen for each caseMultiple controls can be chosen for each case

�� Ex: For a study on ovarian cancer diagnosed in Cook County, Ex: For a study on ovarian cancer diagnosed in Cook County, controls might be selected from gynecologistscontrols might be selected from gynecologists’’ offices in Cook Co. offices in Cook Co. To be eligible, women must have healthy ovaries and be 21To be eligible, women must have healthy ovaries and be 21--79 79 years old.years old.



““Sampling controls with Sampling controls with

replacementreplacement””

�� This means that if a chosen This means that if a chosen ‘‘controlcontrol’’

subject is not eligible or interested in subject is not eligible or interested in

participating, the investigators replace participating, the investigators replace

that control with another subject that control with another subject



Selecting controlsSelecting controls1.1. Frequency (group) matchingFrequency (group) matching

�� Distribution of cases and controls are similar on a Distribution of cases and controls are similar on a known confounding factor known confounding factor

�� Ex: If 20% of cases are 60Ex: If 20% of cases are 60--64 years old, then 64 years old, then 20% of controls are 6020% of controls are 60--64 years old64 years old

�� Commonly usedCommonly used

�� AdvantagesAdvantages�� Eliminates that variable as a confounderEliminates that variable as a confounder

�� More efficient use of dataMore efficient use of data

�� DisadvantagesDisadvantages�� Need to find cases before controls Need to find cases before controls

�� Cannot assess association of matched variable with Cannot assess association of matched variable with diseasedisease



Frequency matching: Frequency matching: Age of Age of cases/controlscases/controls

20-29 yo, Cases 20-29 yo, Controls




0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

Cases Controls



Selecting controlsSelecting controls2.2. Pair (individual) matchingPair (individual) matching

�� Matches each case to a control on one or more confounding Matches each case to a control on one or more confounding factorsfactors

�� Ex: If a 45 year old female is a case, then a 45 year old Ex: If a 45 year old female is a case, then a 45 year old female must be chosen as the matching control female must be chosen as the matching control

�� AdvantagesAdvantages

�� Eliminates these variables as confounding factorsEliminates these variables as confounding factors

�� DisadvantagesDisadvantages

�� Can be more expensive and time consuming to find a Can be more expensive and time consuming to find a match for each casematch for each case

�� May not be able to find a match for each case and May not be able to find a match for each case and therefore lose information on that casetherefore lose information on that case



Pair matching: Age and genderPair matching: Age and gender

53 year old male

WITH DISEASE

(case)

65 year old female

WITH DISEASE

(case)

53 year old male

WITHOUT DISEASE

(control)

65 year old female

WITHOUT DISEASE

(control)



Selecting controlsSelecting controls3.3. No matchingNo matching

�� Investigators do not match on any factorsInvestigators do not match on any factors

�� Ex: If a 45 year old female is a case, then anyone can be Ex: If a 45 year old female is a case, then anyone can be chosen as the control such as a 90 year old male or 18 chosen as the control such as a 90 year old male or 18 year old femaleyear old female

�� AdvantagesAdvantages

�� Easy and NOT time consuming to completeEasy and NOT time consuming to complete

�� Can have a control for every caseCan have a control for every case

�� DisadvantagesDisadvantages

�� Groups may be uneven on certain factors leading to Groups may be uneven on certain factors leading to bias and bias and ‘‘garbagegarbage’’ resultsresults



Sources of bias in case control studiesSources of bias in case control studies

1.1. Recall biasRecall bias-- Cases may (try to) remember information Cases may (try to) remember information

better than controlsbetter than controls

This is why prospective cohort studies are a stronger This is why prospective cohort studies are a stronger

study design than retrospective case control.study design than retrospective case control.

Ex: In a study on tongue cancer, cases may Ex: In a study on tongue cancer, cases may

‘‘rememberremember’’ more details about chewing tobacco use in more details about chewing tobacco use in

adolescence than the associated controls.adolescence than the associated controls.




2.2. NonNon--response and refusal biasresponse and refusal bias-- Cases and/or Cases and/or

controls who refuse to participate may somehow be controls who refuse to participate may somehow be

different than those who do participatedifferent than those who do participate

Ex: In a study on HIV/AIDS, cases may not want to Ex: In a study on HIV/AIDS, cases may not want to

participate if they have to admit that they have participate if they have to admit that they have

used used injectableinjectable drugs or had unprotected sex.drugs or had unprotected sex.




3.3. Selection of controlsSelection of controls-- Controls may not match Controls may not match

the exposure within the general population the exposure within the general population

Ex: Pancreatic cancer and coffee drinkingEx: Pancreatic cancer and coffee drinking�� Cases with pancreatic cancer were admitted to a hospitalCases with pancreatic cancer were admitted to a hospital

�� Hospital controls were used, including large number of Hospital controls were used, including large number of

controls admitted for gastrointestinal diseasescontrols admitted for gastrointestinal diseases

�� Persons with GI diseases were less likely to be coffee Persons with GI diseases were less likely to be coffee

drinkers than population at large causing a biased drinkers than population at large causing a biased

comparisoncomparison




4.4. Misclassification of exposure or diseaseMisclassification of exposure or disease-- A mistake in A mistake in

classification of exposure or disease can cause classification of exposure or disease can cause

spurious end results.spurious end results.

5.5. Confounding factorsConfounding factors-- Other measured or unmeasured Other measured or unmeasured

factors may affect the outcome.factors may affect the outcome.



Advantages of case control studyAdvantages of case control study

1.1. Well suited to rare diseases or those with long Well suited to rare diseases or those with long

latencylatency

2.2. Short time required to conductShort time required to conduct

3.3. Relatively inexpensiveRelatively inexpensive

4.4. Requires relatively few subjectsRequires relatively few subjects

5.5. Existing records can be usedExisting records can be used

6.6. No additional risk to subjectsNo additional risk to subjects

7.7. Can look at multiple exposures for diseaseCan look at multiple exposures for disease



Disadvantages of case control studyDisadvantages of case control study

1.1. Recalled information or existing records determine Recalled information or existing records determine

exposure; validation of information is difficultexposure; validation of information is difficult

2.2. Can study only one disease per studyCan study only one disease per study

3.3. Sometimes unable to control for extraneous variablesSometimes unable to control for extraneous variables

4.4. Difficult to select controls; selection biasDifficult to select controls; selection bias

5.5. Difficult to study timing of eventsDifficult to study timing of events

6.6. Cannot calculate direct measure of risk (incidence)Cannot calculate direct measure of risk (incidence)



Analysis of case control studiesAnalysis of case control studies

�� Odds ratio (OR)Odds ratio (OR)-- Very similar to relative riskVery similar to relative risk

�� Interpreted the same as relative riskInterpreted the same as relative risk

�� RR not used because cannot calculate risk (incidence) in RR not used because cannot calculate risk (incidence) in

retrospective studyretrospective study



Multivariate analysesMultivariate analyses

�� An analytic method that allows the simultaneous An analytic method that allows the simultaneous study of two or more dependent variables (such as study of two or more dependent variables (such as risk factors in a case control study)risk factors in a case control study)

�� When multiple variables are included in an analysis at When multiple variables are included in an analysis at the same time, you are essentially the same time, you are essentially ‘‘controllingcontrolling’’ for for those variables or those variables or ‘‘adjustingadjusting’’ the rates based on the rates based on those variablesthose variables



For exampleFor example……



Questions to ask when reading Questions to ask when reading

a case control studya case control study

1.1. Were the criteria for cases adequately described? Were the criteria for cases adequately described?

For controls?For controls?

2.2. Were cases and controls from the same Were cases and controls from the same

population? Were they comparable?population? Were they comparable?

3.3. Were risk factors adequately defined and Were risk factors adequately defined and

reasonably measured?reasonably measured?

4.4. Was there a potential for recall bias?Was there a potential for recall bias?

5.5. Were there any potentially associated factors that Were there any potentially associated factors that

were not measured or accounted for?were not measured or accounted for?



RothwellRothwell DM, DM, BondyBondy SJ, Williams JI, SJ, Williams JI,

BousserBousser MG. Chiropractic Manipulation MG. Chiropractic Manipulation

and Stroke: A Populationand Stroke: A Population--Based CaseBased Case--

Control Study. Stroke 2001;32;1054Control Study. Stroke 2001;32;1054--

1060.1060.



Smith WS, Johnston SC, Smith WS, Johnston SC, SkalabrinSkalabrin EJ, EJ,

Weaver M, Weaver M, AzariAzari P, Albers GW, P, Albers GW, GressGress

DR. Spinal manipulative therapy is an DR. Spinal manipulative therapy is an

independent risk factor for vertebral independent risk factor for vertebral

artery dissection. Neurology artery dissection. Neurology

2003;60:14242003;60:1424––14281428



Cassidy JD, Boyle E, Cote P, He Y, Cassidy JD, Boyle E, Cote P, He Y,

HoggHogg--Johnson S, Silver FL, Johnson S, Silver FL, BondyBondy SJ. SJ.

Risk of Risk of VertebrobasilarVertebrobasilar Stroke and Stroke and

Chiropractic Care: Results of a Chiropractic Care: Results of a

PopulationPopulation--Based CaseBased Case--Control and Control and

CaseCase--Crossover Study. Spine Crossover Study. Spine

2008;33(4S):S1762008;33(4S):S176––S183S183



Hierarchy of scientific evidenceHierarchy of scientific evidence

Meta-Analysis,

Systematic Reviews

Best Evidence / Evidence Guidelines

Randomized, controlled trials (RCT)

Clinical trials, Cohort Studies, Case Control

Case series

Case study / case report

Animal studies, in vitro studies

Expert opinions, editorials



Any questions?Any questions?

Case -Control Studies...National University of Health Sciences EBP@NUHS Essential Evidence Based...

Documents

Transcript of Case -Control Studies...National University of Health Sciences EBP@NUHS Essential Evidence Based...