Analytical Study Analytical Study Designs Designs Dr. Jessievilita Dr. Jessievilita C. Fabian C. Fabian
Nov 07, 2014
Analytical Study Analytical Study DesignsDesigns
Dr. Jessievilita C. Fabian Dr. Jessievilita C. Fabian
ANALYTICAL STUDIESANALYTICAL STUDIES
Case-control Case-control – RetrospectiveRetrospective
CohortCohort– ProspectiveProspective
Usefulness of Study Usefulness of Study DesignsDesigns Hypotheses generation OnlyHypotheses generation Only
– Cross-sectional surveysCross-sectional surveys– Ecologic studiesEcologic studies
Both Hypotheses generation and Both Hypotheses generation and Hypotheses testingHypotheses testing– Case-control studiesCase-control studies– Cohort studiesCohort studies
Hypothesis GenerationHypothesis Generation
The process of developing a list of The process of developing a list of possible candidates for the possible candidates for the “causes” of the disease and “causes” of the disease and obtaining initial evidence that obtaining initial evidence that support one or more of these support one or more of these candidatescandidates
Hypothesis TestingHypothesis Testing
When one or more hypotheses When one or more hypotheses are generated, they must be are generated, they must be tested by making predictions tested by making predictions from the hypotheses and from the hypotheses and examining new data to determine examining new data to determine if the predictions are correct.if the predictions are correct.
Case-Control StudyCase-Control Study
Examines the association between Examines the association between disease and potential risk factorsdisease and potential risk factors
DISEASE DISEASE ≈ RISK FACTORS≈ RISK FACTORS
Done by taking separate samples Done by taking separate samples of diseased cases and of controls of diseased cases and of controls at risk of developing diseaseat risk of developing disease
Information may be collected on Information may be collected on both cases and controls on:both cases and controls on:– GeneticGenetic– Social Social – BehavioralBehavioral– EnvironmentalEnvironmental– Other determinants of disease riskOther determinants of disease risk
Design of Case-Control Design of Case-Control StudiesStudies
Cases: persons/group with a given Cases: persons/group with a given diseasedisease
Controls: persons/group without the Controls: persons/group without the given disease given disease
Ascertain exposure or background of the Ascertain exposure or background of the two groups and compare the proportiontwo groups and compare the proportion
Best suited for study of diseases where Best suited for study of diseases where medical care usually sought, (hip medical care usually sought, (hip fracture, cancer) because this makes it fracture, cancer) because this makes it easier to identify cases easier to identify cases
Early case-control Early case-control studies studies Broders (1920)Broders (1920)
- association between pipe - association between pipe smoking and lip cancersmoking and lip cancer
19501950
- association between - association between cigarette smoking and lung cigarette smoking and lung cancercancer
Selection CriteriaSelection Criteria
Definition of disease or outcome Definition of disease or outcome of interestof interest
Establish strict diagnostic criteriaEstablish strict diagnostic criteria
Selection of CasesSelection of Cases
Ideally, investigator identifies & enrolls all Ideally, investigator identifies & enrolls all incident cases in a defined population in a incident cases in a defined population in a specified time period specified time period
Select cases from registries or hospitals, Select cases from registries or hospitals, clinicsclinics
When all incident cases in a population are When all incident cases in a population are included, the study is representative; included, the study is representative; otherwise there is potential for bias (e.g. otherwise there is potential for bias (e.g. referral bias)referral bias)
Use of prevalent vs incident casesUse of prevalent vs incident cases
Prevalence vs Prevalence vs IncidenceIncidence
Cross-sectional studiesCross-sectional studies PrevalencePrevalence
– Probability that the person was ill at a Probability that the person was ill at a given point in timegiven point in time
For studies of disease etiology:For studies of disease etiology: IncidenceIncidence
– Probability of developing the disease Probability of developing the disease during the study period among subjects during the study period among subjects who are free of disease initiallywho are free of disease initially
Selection of controlsSelection of controls
Critical that the exposure in the Critical that the exposure in the controls is representative of the controls is representative of the exposure in the populationexposure in the population
Ideal controls would have Ideal controls would have same/similar characteristics as same/similar characteristics as the casesthe cases
Matching cases to controlsMatching cases to controls
Purpose of ControlsPurpose of Controls
To determine if the frequency of To determine if the frequency of exposure in the case group is exposure in the case group is different from that which would different from that which would have been expected based on have been expected based on experience of comparable experience of comparable individuals who do not have the individuals who do not have the diseasedisease
Sources of ControlsSources of Controls
Population-based controlsPopulation-based controls Hospital ControlsHospital Controls Neighborhood controlsNeighborhood controls
Population-Based Population-Based ControlsControls
The best control group is a random The best control group is a random sample of individuals from same sample of individuals from same source population (as the cases) source population (as the cases) who have not developed the disease who have not developed the disease
Population-based controls are the Population-based controls are the best way to ensure that the best way to ensure that the distribution of exposure among the distribution of exposure among the controls is representativecontrols is representative
Random digit dialing or canvassing Random digit dialing or canvassing households households
Hospital ControlsHospital Controls
Hospital controls are the most Hospital controls are the most frequently used source frequently used source
Consists of patients at the same Consists of patients at the same hospital (as the case) who have been hospital (as the case) who have been admitted for conditions other than the admitted for conditions other than the disease being studieddisease being studied
The use of other ill persons as controls will The use of other ill persons as controls will provide a valid result only if their illness is provide a valid result only if their illness is unrelated to the exposure in question. unrelated to the exposure in question.
Benefits of Using Benefits of Using Hospital ControlsHospital Controls
ConvenientConvenient CheapCheap NumerousNumerous Avoids non-responseAvoids non-response When a population-based case registry is When a population-based case registry is
not available, hospital controls better not available, hospital controls better represent the subpopulation from which represent the subpopulation from which the cases arosethe cases arose
Disadvantages of Disadvantages of Hospital Control Hospital Control Hospital controls may not be Hospital controls may not be
representative of exposure rates in the representative of exposure rates in the target population target population
Hospitalized controls are also ill and, Hospitalized controls are also ill and, therefore differ from healthy therefore differ from healthy individuals in some ways individuals in some ways
Hospitalized individuals are more likely Hospitalized individuals are more likely to smoke cigarettes or are heavy to smoke cigarettes or are heavy alcoholic drinkersalcoholic drinkers
Neighborhood Neighborhood ControlsControls Usually selected from residence in Usually selected from residence in
the same city, block, or other the same city, block, or other geographic area as the casegeographic area as the case
In an attempt to reduce the In an attempt to reduce the variability of factors such as access to variability of factors such as access to medical care and socio-economic medical care and socio-economic statusstatus
Usually individually matched to cases Usually individually matched to cases by age and genderby age and gender
Advantages of Advantages of Neighborhood ControlsNeighborhood Controls Control selection does not require Control selection does not require
the existence of a roster or use of the existence of a roster or use of a telephonea telephone
Confounding factors associated Confounding factors associated with neighborhood may be with neighborhood may be balanced between cases and balanced between cases and controlscontrols
DisadvantagesDisadvantages
Potential for not satisfying the Potential for not satisfying the study base principle i.e. case and study base principle i.e. case and controls should be representative controls should be representative of the same base experienceof the same base experience
High cost associated with High cost associated with contacting each potential controlcontacting each potential control
Most commonly used Most commonly used ControlsControls Probability samples of the population Probability samples of the population
from which cases come fromfrom which cases come from Persons seeking medical care at the Persons seeking medical care at the
same institutions as the cases for same institutions as the cases for conditions believed to be unrelated to conditions believed to be unrelated to the cases’ diagnosisthe cases’ diagnosis
Neighbors of the casesNeighbors of the cases Friends, associates, co-workers, Friends, associates, co-workers,
siblingssiblings
Use of Multiple Use of Multiple ControlsControls
Case to control ratio used is Case to control ratio used is usually 1:1; if large number and usually 1:1; if large number and cost is the same for both groupscost is the same for both groups
If a study has a small number of If a study has a small number of cases, increasing the number of cases, increasing the number of controls increases power of studycontrols increases power of study
Ascertainment of Ascertainment of Disease and ExposureDisease and Exposure Source of Disease InformationSource of Disease Information
– Death certificatesDeath certificates– Disease registriesDisease registries– Office records of physicians, hospital Office records of physicians, hospital
admissions, or discharge recordsadmissions, or discharge records– Pathology department log booksPathology department log books
Source of Exposure DataSource of Exposure Data– Study subjects themselves Study subjects themselves
(interview or mailed questionnaire)(interview or mailed questionnaire)– Surrogate persons (i.e. spouses of Surrogate persons (i.e. spouses of
participants, mothers of children)participants, mothers of children)– Information recorded in medical Information recorded in medical
recordsrecords
Procedures used to obtain Procedures used to obtain information must be similar as information must be similar as possible for cases and controlspossible for cases and controls
Advantages of Case Advantages of Case Control DesignControl Design Relatively inexpensive, fairly quick and Relatively inexpensive, fairly quick and
easy to performeasy to perform Can study many risk factorsCan study many risk factors Good for diseases with long latencyGood for diseases with long latency Optimal for rare diseasesOptimal for rare diseases Multiple etiologic factors evaluated for Multiple etiologic factors evaluated for
single diseasesingle disease Shorter timeShorter time Smaller sampleSmaller sample
Major PitfallsMajor Pitfalls
Susceptibility of case-control Susceptibility of case-control study to study to – selection or sampling biasselection or sampling bias– Measurement errorMeasurement error– Missing dataMissing data
Especially when exposures are Especially when exposures are assessed by interviewassessed by interview
Limitations of Case Limitations of Case Control DesignControl Design
Identifying controls may be difficultIdentifying controls may be difficult Temporal relationship between exposure & Temporal relationship between exposure &
disease difficult to establishdisease difficult to establish Can obtain only a relative measure of riskCan obtain only a relative measure of risk Prone to bias (Recall) compared with other Prone to bias (Recall) compared with other
study designsstudy designs Information on potential risk factors and Information on potential risk factors and
confounders may not be available from confounders may not be available from records or the subjects’ memoriesrecords or the subjects’ memories
Limitations of Case Limitations of Case Control Design Cont’dControl Design Cont’d
Unless study is population based Unless study is population based can’t measure incidence of diseasecan’t measure incidence of disease
Can study only one disease Can study only one disease outcome at a timeoutcome at a time
Bad for rare exposures (despite a Bad for rare exposures (despite a large number of cases, may still large number of cases, may still end up with few exposed cases)end up with few exposed cases)
Statistical AnalysisStatistical Analysis
Case-control study provides Case-control study provides estimate of estimate of
RELATIVE RISK RELATIVE RISK
(Disease Odds Ratio)(Disease Odds Ratio)
Synonyms: cross-product ratio;Synonyms: cross-product ratio;
relative oddsrelative odds
Data LayoutData Layout
Exposure Exposure StatusStatus
OutcomOutcomee
DiseaseDisease
(cases)(cases)No No DiseaseDisease
(controls)(controls)
TotalTotal
ExposedExposed aa bb n1n1
Not Not ExposedExposed
cc dd nono
TotalTotal m1m1 momo NN
Analysis:Analysis:
Effect Effect
(Disease)(Disease)
Exposure Exposure
++ --
++ aa bb a+ba+b
-- cc dd c+dc+d
a+ca+c b+db+d a+ba+b+c++c+dd
Measure of AssociationMeasure of Association
Odds Ratio (OR) is the ratio of 2 oddsOdds Ratio (OR) is the ratio of 2 odds– The odds of exposure among cases (a/c)The odds of exposure among cases (a/c)– The odds of exposure among controls The odds of exposure among controls
(b/d)(b/d)
OR= OR= a/c a/c = = adad
b/d bcb/d bc
The Odds Ratio (OR) provides a valid The Odds Ratio (OR) provides a valid estimate of the relative risk under estimate of the relative risk under conditions that prevail in most conditions that prevail in most case-control studiescase-control studies
The cases of the disease are newly The cases of the disease are newly diagnoseddiagnosed
Prevalent cases are not included in Prevalent cases are not included in the control groupthe control group
The selection of cases and controls The selection of cases and controls is not based on exposure statusis not based on exposure status
ExampleExampleNumber of Hospitalized Married Female Number of Hospitalized Married Female Patients w/ and w/o Myocardial Infarction under Patients w/ and w/o Myocardial Infarction under 45 years of age by Oral Contraceptive Practice45 years of age by Oral Contraceptive Practice
Oral Contraceptive PracticeOral Contraceptive Practice With MIWith MI W/o MIW/o MI
UsedUsed 2323 3434
Never UsedNever Used 3535 132132
TotalTotal 5858 166166
OR = OR = adad = = 23 x 13223 x 132 = = 3,036 3,036 = 2.55 = 2.55
bc 35 x 34bc 35 x 34 1,1901,190
Interpretation: The rate of myocardial infarction Interpretation: The rate of myocardial infarction is more than twice among oral contraceptive is more than twice among oral contraceptive users than among non-usersusers than among non-users
Confidence Interval Confidence Interval Procedures: Procedures: Taylor Series ProcedureTaylor Series Procedure1. Given:1. Given: Odds Ratio: 2.55Odds Ratio: 2.55
2. Take the natural logarithm of OR2. Take the natural logarithm of OR
ln(OR) 2.55=.9360ln(OR) 2.55=.9360
3.3. Calculate the standard error of Calculate the standard error of ln(OR)ln(OR)
SE(ln OR) = SE(ln OR) = 1/a+1/b+1/c+1/d1/a+1/b+1/c+1/d
3. SE(ln OR) = 3. SE(ln OR) = 1/23+1/34+1/35+1/1321/23+1/34+1/35+1/132
= 0.327= 0.327
4.4. Determine the Z deviate Determine the Z deviate corresponding to your chosen corresponding to your chosen confidence levelconfidence levelConfidenceConfidence
IntervalIntervalTwo-Two-tailedtailed
One-One-tailedtailed
90%90% 1.641.64 1.281.28
95%95% 1.961.96 1.641.64
99%99% 2.572.57 2.322.32
5.5. Construct the 95% CI of the ln Construct the 95% CI of the ln OR. Take the exponential to get OR. Take the exponential to get the 95% confidencethe 95% confidence
1.1. Lower Limit = ln OR – Z deviate Lower Limit = ln OR – Z deviate [[SE(ln SE(ln OR)OR)]]
2.2. Upper Limit = ln OR + Z deviate Upper Limit = ln OR + Z deviate [[SE(ln SE(ln OR)OR)]]
LL= 0.936 – 1.96 (0.327)LL= 0.936 – 1.96 (0.327)
= 0.296= 0.296
Take exponential e0.296= 1.34Take exponential e0.296= 1.34
UL = UL = 0.936 + 1.96 (0.327)0.936 + 1.96 (0.327)
= 1.57= 1.57
e1.57 = 4.80e1.57 = 4.80
Therefore:Therefore:
OR (95% confidence interval) =OR (95% confidence interval) =
2.55 ( 1.34 – 4.80)2.55 ( 1.34 – 4.80)
You are 95% confident that the rate You are 95% confident that the rate of myocardial infarction is more of myocardial infarction is more than twice among oral than twice among oral contraceptive users than non-contraceptive users than non-users.users.
Hypothesis TestingHypothesis Testing
To test whether the observed measure To test whether the observed measure of association (OR=2.55) is of association (OR=2.55) is statistically significantstatistically significant
Note:Note:OR=1 means no association between OR=1 means no association between
exposure and diseaseexposure and diseaseOR>1 indicates associationOR>1 indicates associationOR<1 indicates protective effect of the OR<1 indicates protective effect of the
exposure variable exposure variable → exposure to the factor → exposure to the factor decreases the risk of developing the diseasedecreases the risk of developing the disease
Hypothesis TestingHypothesis Testing
Ho : There is no association Ho : There is no association between the exposure and the between the exposure and the disease (OR=1)disease (OR=1)
H1 : There is an association (OR>1)H1 : There is an association (OR>1)
Chi-square: test for 2 quali Chi-square: test for 2 quali variablesvariables
22 22
X = X = (N-1) (ad –bc)(N-1) (ad –bc) n1 no m1 mon1 no m1 mo 22
= (= (224-1) 224-1) [([(23x132)-(34x35)23x132)-(34x35)]]
(57) (167) (58) (166)(57) (167) (58) (166)
= 8.29= 8.29
2 22 2
Calculated X = 8.29 > XCalculated X = 8.29 > X0.05, df0.05, df1=3.841=3.84
Therefore: We reject the null Therefore: We reject the null hypothesis and we say that there is hypothesis and we say that there is sufficient evidence (statistically sufficient evidence (statistically significant) significant) to say that there is an to say that there is an association between the use of oral association between the use of oral contraceptive among married women contraceptive among married women under 45 years old with MIunder 45 years old with MI
Note: In conclusion, its eitherNote: In conclusion, its either
We reject or do not reject the null We reject or do not reject the null hypothesishypothesis
Estimating Impact Estimating Impact FractionsFractions Attributable Fraction of the Attributable Fraction of the
Exposed (AFE)Exposed (AFE)– Proportion of the Proportion of the exposed cases exposed cases
which may be attributed to the which may be attributed to the exposureexposure
Attributable Fraction (AF)Attributable Fraction (AF)– Proportion of Proportion of all cases all cases that may be that may be
attributed to the exposureattributed to the exposure
Estimating Impact Estimating Impact FractionsFractions Attributable Fraction of the ExposedAttributable Fraction of the Exposed
(AFE)(AFE)
AFE = AFE = OR – 1 OR – 1 = = 2.55-1 2.55-1 = 0.99 = 0.99
OROR 2.55 2.55
Therefore: 99% of MI among OC usersTherefore: 99% of MI among OC users
could be attributed to oral could be attributed to oral
contraceptivescontraceptives
Attributable FractionAttributable Fraction
AF = (a/m1) (AFE)AF = (a/m1) (AFE) = (23/58) (0.99)= (23/58) (0.99)
= 0.39= 0.39
Therefore:Therefore:
39% of MI cases could be attributed to 39% of MI cases could be attributed to oral contraceptive use.oral contraceptive use.
Thank You!Thank You!
COHORT STUDY COHORT STUDY DESIGNDESIGN
Cohort StudyCohort Study
An analytic study design that is a An analytic study design that is a more straightforward method of more straightforward method of drawing causal inferencesdrawing causal inferences
Involves a design in which the Involves a design in which the information about the study factor information about the study factor (Exposure and Non-exposure) is (Exposure and Non-exposure) is known for all subjects at the known for all subjects at the beginning of the follow-up periodbeginning of the follow-up period
COHORTCOHORT
- - a clearly identified a clearly identified group of people to be group of people to be studiedstudied
Cohort StudiesCohort Studies
Assembling 1 or more cohortsAssembling 1 or more cohortsChoosing persons who were (+) exposed and Choosing persons who were (+) exposed and were not (-) exposed to risk factorswere not (-) exposed to risk factorsSubjects are free of the outcome of interestSubjects are free of the outcome of interestSubjects are followed over time Subjects are followed over time
-whether or not they develop the -whether or not they develop the diseasedisease- whether the risk factors that were - whether the risk factors that were measured at the beginning predict the measured at the beginning predict the
diseases that occurdiseases that occur
A Cohort Study involving incident cases of A Cohort Study involving incident cases of a chronic disease is concerned witha chronic disease is concerned with– First occurrence of the diseaseFirst occurrence of the disease
∆ ∆ the candidate population is composed of all the candidate population is composed of all persons without the disease who are eligible persons without the disease who are eligible to become casesto become cases
Synonyms:Synonyms:
Follow-up studyFollow-up study
Prospective studyProspective study
Incidence or mortality studyIncidence or mortality study
Longitudinal studyLongitudinal study
Defining characteristics of Cohort Defining characteristics of Cohort StudiesStudies
Groups are defined on the basis of Groups are defined on the basis of exposure and are followed for outcomesexposure and are followed for outcomes
Exposure Exposure → Outcome→ Outcome
In contrast with Case-control studies:In contrast with Case-control studies:
OutcomeOutcome ← Exposure← Exposure
Compared groups: Exposed and Compared groups: Exposed and UnexposedUnexposed
Maybe selected from separate Maybe selected from separate populationspopulations
Causal inference depends on the Causal inference depends on the assumption that the population assumption that the population are comparable with respect to are comparable with respect to factors other than the exposure of factors other than the exposure of interestinterest
Strength of Cohort Strength of Cohort Study Study The study factor level on each The study factor level on each
subject is observed at the onset subject is observed at the onset of the study period i.e. before the of the study period i.e. before the disease is detecteddisease is detected Example: Obesity Example: Obesity → Glucose → Glucose
tolerancetolerance- Cause preceded the diseaseCause preceded the disease- Relatively free of selection biasRelatively free of selection bias
Advantages of Cohort Advantages of Cohort StudiesStudies Since one starts with individuals Since one starts with individuals
free of the outcome (disease free of the outcome (disease free), the temporal sequence free), the temporal sequence between the exposure and between the exposure and outcome is clearly established outcome is clearly established → → a requirement for the exposure to be a a requirement for the exposure to be a causal factorcausal factor
Advantages of Cohort Advantages of Cohort StudiesStudies Allows for direct calculation of Allows for direct calculation of
exposure specific incidence ratesexposure specific incidence rates With cohort, one can look at With cohort, one can look at
multiple effects of a single multiple effects of a single exposureexposure
Suitable for studies of rare Suitable for studies of rare exposuresexposures
Selection bias is minimizedSelection bias is minimized
Advantages of Cohort StudiesAdvantages of Cohort Studies
Able to determine time sequence Able to determine time sequence between exposure & diseasebetween exposure & disease
Able to study multiple exposures & Able to study multiple exposures & multiple outcomesmultiple outcomes
Advantages of Cohort StudiesAdvantages of Cohort Studies
Can be performed retrospectively or Can be performed retrospectively or prospectivelyprospectively
Can be used to obtain a true Can be used to obtain a true (absolute) measure of risk(absolute) measure of risk
Disadvantages of Cohort Disadvantages of Cohort StudiesStudies
Typically requires a long follow up period Typically requires a long follow up period (although this is shorter with retrospective (although this is shorter with retrospective cohort studies)cohort studies)
Expensive (again, less so with Expensive (again, less so with retrospective)retrospective)
Can study only the risk factors measured Can study only the risk factors measured at the beginning at the beginning
Disadvantages of Cohort StudiesDisadvantages of Cohort Studies
May require large sample sizeMay require large sample size
Potential for loss to follow upPotential for loss to follow up
Inefficient for rare diseases by reason of Inefficient for rare diseases by reason of large sample size requirementlarge sample size requirement
Need adequate records of exposure to Need adequate records of exposure to perform retrospective cohort studyperform retrospective cohort study
2 Types of Cohort Studies 2 Types of Cohort Studies
Prospective Cohort StudyProspective Cohort Study
Retrospective Cohort StudyRetrospective Cohort Study
PROSPECTIVE COHORT STUDYPROSPECTIVE COHORT STUDY
Assemble study (risk) groups in the Assemble study (risk) groups in the present time, collect baseline date, present time, collect baseline date, continue to collect data for a period that continue to collect data for a period that can last many yearscan last many years
Individuals included in the study are free of Individuals included in the study are free of the outcome of interest at the time of the outcome of interest at the time of initiation of the studyinitiation of the study
Prospective Cohort StudyProspective Cohort Study
Exposure level is measured either:Exposure level is measured either:
–At the beginning of the study (fixed At the beginning of the study (fixed cohort)cohort)
–At the time of enrollment (dynamic At the time of enrollment (dynamic cohort)cohort)
Fixed CohortFixed Cohort
Members are all gathered at the beginning Members are all gathered at the beginning of the studyof the study
There are no additions after study startsThere are no additions after study starts
All members are followed-up for the All members are followed-up for the specified period of timespecified period of time
Dynamic CohortDynamic Cohort
Membership is not fixed at the start of the Membership is not fixed at the start of the studystudy
Recruitment process is on-going even Recruitment process is on-going even after some members are already followed-after some members are already followed-upup
This results in non-uniform times of follow-This results in non-uniform times of follow-up for the members of the cohortup for the members of the cohort
In both fixed or dynamic cohort, the In both fixed or dynamic cohort, the investigator exerts efforts to investigator exerts efforts to
follow-upfollow-up these individuals over time (into the future) these individuals over time (into the future)
to determine the incidence of the outcome to determine the incidence of the outcome of interestof interest→ PROSPECTIVE→ PROSPECTIVE
RETROSPECTIVE COHORT RETROSPECTIVE COHORT STUDYSTUDY
Investigator goes back into history to Investigator goes back into history to define a risk group and follow up to the define a risk group and follow up to the present to see what outcomes have present to see what outcomes have occurredoccurred
- e.g. people exposed to Hiroshima atomic - e.g. people exposed to Hiroshima atomic bomb in August 1945bomb in August 1945
RETROSPECTIVE COHORT RETROSPECTIVE COHORT STUDYSTUDY
A group of individuals who are free of the A group of individuals who are free of the outcome of interest at some point in the past are outcome of interest at some point in the past are identified in terms of their exposure levelidentified in terms of their exposure levelCohorts may also be fixed or dynamicCohorts may also be fixed or dynamicThe outcome status of these individuals up to The outcome status of these individuals up to specified time is determined through existing specified time is determined through existing records records (e.g. morbidity/mortality records) (e.g. morbidity/mortality records) The outcome status of these individuals are The outcome status of these individuals are already known (recorded) at the time of studyalready known (recorded) at the time of study
Selection of the Study Selection of the Study CohortsCohorts Selection of individuals to Selection of individuals to
comprise the exposed population comprise the exposed population and unexposed population in a and unexposed population in a cohort can come from a variety of cohort can come from a variety of sourcessources
Choice depends on scientific Choice depends on scientific considerations, feasibility, and considerations, feasibility, and frequency of exposurefrequency of exposure
Identify special Identify special exposure groupsexposure groups Asbestos workerAsbestos worker Radiology techniciansRadiology technicians Individuals living near a suspected Individuals living near a suspected
environmental hazardenvironmental hazard Groups with unusual diet or Groups with unusual diet or
lifestylelifestyle Individuals exposed to a particular Individuals exposed to a particular
incident (Hiroshima atomic bomb)incident (Hiroshima atomic bomb)
Selection of Selection of Comparison GroupComparison Group An external group is chosen for An external group is chosen for
comparisoncomparison
- Comparison group should be similar - Comparison group should be similar in all respects except for the exposure in all respects except for the exposure
Sampling from general populationSampling from general population- Suitable when there is considerable Suitable when there is considerable
heterogeneity of exposure levelsheterogeneity of exposure levels- Ex. Cigarette smokingEx. Cigarette smoking
Coffee drinkingCoffee drinking
Selection of Selection of Comparison GroupComparison Group Special Resource GroupsSpecial Resource Groups
– Used instead of the general Used instead of the general populationpopulation
– Ex: physician association, nurses, Ex: physician association, nurses, people with prepaid plans, college people with prepaid plans, college studentsstudents
OUTCOME OF OUTCOME OF EXPOSUREEXPOSURE The outcome experience of The outcome experience of
members of the cohort which are members of the cohort which are classified as EXPOSED is classified as EXPOSED is compared to the experience of compared to the experience of members of the cohort who are members of the cohort who are NOT EXPOSEDNOT EXPOSED
Sources of DataSources of Data
Interview of cohort membersInterview of cohort members Medical examination of cohort Medical examination of cohort
membersmembers Existing medical records and Existing medical records and
measures of the environmentmeasures of the environment
Ascertainment of Ascertainment of OutcomeOutcome Periodic examination of cohort Periodic examination of cohort
membersmembers Routine surveillance of deaths, Routine surveillance of deaths,
hospitalizations and clinic visitshospitalizations and clinic visits Interview of cohort membersInterview of cohort members
Comparison of Case/Control Comparison of Case/Control & Cohort Studies& Cohort Studies
InefficientInefficient EfficientEfficient EfficientEfficient
CheapCheap ShorterShorter SmallerSmaller Difficult to assessDifficult to assess
EfficientEfficient InefficientInefficient InefficientInefficient
More ExpensiveMore Expensive LongerLonger LargerLarger Good to Good to
determine (RR)determine (RR)
Rare ExposureRare Exposure Rare DiseaseRare Disease Disease with Disease with
long latencylong latency CostCost TimeTime SizeSize Temporal Temporal
RelationshipRelationship
Case/Control Cohort
DATA ANALYSISDATA ANALYSIS
When follow-up of cohort members is When follow-up of cohort members is completecomplete
When all members of the cohort are When all members of the cohort are followed up for the specified risk period followed up for the specified risk period → → Outcome status is knownOutcome status is known
→ → label: 1. Caselabel: 1. Case
2. Non-cases2. Non-cases
Analysis:Analysis:
Effect Effect
(Disease)(Disease)
Exposure Exposure
++ --
++ aa bb a+ba+b
-- cc dd c+dc+d
a+ca+c b+db+d a+ba+b+c++c+dd
In cohort (Fixed In cohort (Fixed Cohort)Cohort) The Cumulative Incidence/Attack Rate The Cumulative Incidence/Attack Rate
of the effect (disease):of the effect (disease):– Among those exposed to the factor (CIe)Among those exposed to the factor (CIe)
= a= a
__________
a+ba+b
– Among those not exposed (CIu)Among those not exposed (CIu)= c= c
________
c+dc+d
Cumulative IncidenceCumulative Incidence
Used as an estimate of the risk of Used as an estimate of the risk of outcome of interestoutcome of interest
There is statistical association There is statistical association between the suspected causal between the suspected causal factor and the alleged effect if:factor and the alleged effect if:
aa /=/= cc
___ ______ ___
a+b c+da+b c+d
Cumulative Incidence Cumulative Incidence RatioRatioRisk Ratio (RR)Risk Ratio (RR)Absolute RiskAbsolute Risk RR = RR = CIe CIe = = a/a+ba/a+b
CIu c/c+dCIu c/c+d
Indicative ValuesIndicative Values (Magnitude and Direction of Association)(Magnitude and Direction of Association)
Ratio Values (RR)Ratio Values (RR)0 - to infinity0 - to infinityWhere:Where:
0 = indicate strong negative 0 = indicate strong negative associationassociation
1 = null value; no association1 = null value; no associationNull hypothesis:Null hypothesis:→ → No association of Exposure and OutcomeNo association of Exposure and Outcome
>1 = strong positive association>1 = strong positive association 0 ---------0 ---------← 1 →← 1 →
Cumulative Incidence Cumulative Incidence DifferenceDifference
Cumulative Incidence Difference/Cumulative Incidence Difference/
Risk DifferenceRisk Difference
CID=RD= CIe- CIuCID=RD= CIe- CIu
Indicative ValuesIndicative Values(Magnitude and Direction of (Magnitude and Direction of Association)Association) Difference ValuesDifference Values
– The bigger the magnitude of the The bigger the magnitude of the difference, the stronger is the difference, the stronger is the associationassociation
Positive (+) difference:Positive (+) difference:→ → the exposure is a risk factorthe exposure is a risk factorNegative (-) difference:Negative (-) difference:→ → indicates a protective factorindicates a protective factorA zero difference = no associationA zero difference = no association
Example:Example:
A group of anesthesiologists wanted to A group of anesthesiologists wanted to investigate the relationship of type of anesthesia investigate the relationship of type of anesthesia and post-op pulmonary infectionand post-op pulmonary infection
Post-op Pulmo Post-op Pulmo InfectionInfection
AnesthesiaAnesthesia With With InfectioInfectionn
W/O W/O InfectionInfection
TotalTotal
GeneralGeneral 143143 1717 160160
SpinalSpinal 33 5252 5555
TotalTotal 146146 6969 215215
General anaesthesia:General anaesthesia: Cumulative IncidenceCumulative Incidence
CIe = 143/160 = 0.894CIe = 143/160 = 0.894CIu = 3/55 = 0.055CIu = 3/55 = 0.055
Cumulative Incidence Ratio or RRCumulative Incidence Ratio or RR
= CIe/CIu = 0.89/0.055 = 16.38= CIe/CIu = 0.89/0.055 = 16.38
→ → Strong positive association between General Strong positive association between General Anesthesia and post-op pulmonary infectionAnesthesia and post-op pulmonary infection
Cumulative Incidence Difference Cumulative Incidence Difference (CID) or Risk Difference(CID) or Risk DifferenceCID = CIe – CIu = 0.894-0.055CID = CIe – CIu = 0.894-0.055
= 0.839 (positive)= 0.839 (positive)≈ ≈ The exposure is a risk factorThe exposure is a risk factor
General anesthesia is a risk factor General anesthesia is a risk factor for post-op pulmonary infectionfor post-op pulmonary infection
Questions: Questions:
1. What are the cumulative incidence of the 1. What are the cumulative incidence of the exposed and unexposed?exposed and unexposed?
CIe = a/a+bCIe = a/a+bCIu = c/c+dCIu = c/c+d
2. What is the risk ratio of the exposed over the 2. What is the risk ratio of the exposed over the unexposed?unexposed?
RR = CIe/CIuRR = CIe/CIu3. What is the risk difference?3. What is the risk difference?
RD = CIe – CIuRD = CIe – CIu4. Interpret risk difference4. Interpret risk difference
Dynamic CohortDynamic Cohort
Analysis should consider the total Analysis should consider the total period of exposure contributed by period of exposure contributed by each member of the cohorteach member of the cohort
The total population time is the The total population time is the sum of the individuals’ length of sum of the individuals’ length of time under which he is considered time under which he is considered at risk of the outcomeat risk of the outcome
Data LayoutData Layout
Number Number of of ExposedExposed
Number Number of of UnexposeUnexposedd
TotalTotal
Number Number of Casesof Cases
aa bb m1m1
PopulatioPopulation-timen-time
LeLe LuLu LL
Where:Where:
a = no. of exposed with the diseasea = no. of exposed with the disease
b = no. of unexposed without the b = no. of unexposed without the dsedse
m1 = total number of casesm1 = total number of cases
Le = population time (length ofLe = population time (length of
observation) among exposedobservation) among exposed
Lu = population time (length ofLu = population time (length of
observation) among observation) among unexposedunexposed
AnalysisAnalysis
Compute for:Compute for:
1. Incidence Density (ID)1. Incidence Density (ID)– Also called Incidence Rate (IR) Also called Incidence Rate (IR)
IDe = a/LeIDe = a/Le
IDu = b/LuIDu = b/Lu
2.2. Incidence Density Ratio (IDR) orIncidence Density Ratio (IDR) or Rate Ratio (RR)Rate Ratio (RR)
IDR=RR= IDe/IDu = IDR=RR= IDe/IDu = a/Lea/Le b/Lub/Lu
3.3. Incidence Density Difference (IDD) or Incidence Density Difference (IDD) or Rate Difference (RD)Rate Difference (RD)IDD=RD= IDe – IDu = IDD=RD= IDe – IDu = aa – – bb
Le LuLe LuThese measures of association These measures of association have have
similar similar interpretations as those of interpretations as those of Cumulative Incidence Ratio and Cumulative Incidence Ratio and Cumulative Incidence DifferenceCumulative Incidence Difference
ExposurExposure Statuse Status
ObeseObese Non-Non-obeseobese
DeathsDeaths
1960-19671960-19673030 3636 6666
Person-Person-years years follow-up,follow-up,
1960-10671960-1067
699699 13991399 20982098
Given: a = 30Given: a = 30 Le= 699Le= 699b = 36b = 36 Lu= 1399Lu= 1399
Computation:Computation:1.1. ID (exposed) = 30/699 = 0.042ID (exposed) = 30/699 = 0.0422.2. ID (unexposed)= 36/1399=0.026ID (unexposed)= 36/1399=0.0263.3. IDR = 0.042/0.026 = 1.61 (>1)IDR = 0.042/0.026 = 1.61 (>1)
= strong positive association= strong positive association4.4. IDD = 0.042 – 0.026 = 0.016 IDD = 0.042 – 0.026 = 0.016
(positive)(positive) = exposure is a risk factor= exposure is a risk factorTherefore: Therefore: Obesity is a risk factor for Obesity is a risk factor for
death in death in 7 years of study.7 years of study.
THANK YOUTHANK YOU