Escolar Documentos
Profissional Documentos
Cultura Documentos
EPIDEMIOLOGICSTUDY
A.Observational
STUDYDESIGNS:
Crosssectional
Casecontrol
Beforeafter,Validation
Prof.JudithM.Sison,MD,MPH,FPOGS
2008
Crosssectional
B.Experimental
NOTrandomized
Communitytrials
Randomized
Clinicaltrials
Casecontol
Samplingw/regardtoDisease/
Effect
ExposureofcharacteristicsATtime
ofstudy
Samplingw/regardtoDisease/
Effect
Retrospectivehistoryofexposure
ofcharacteristicsPRIORtotimeof
study
Cohort
Samplingw/regardtoExposureor
Cause
Prospective/Historical
OBSERVATIONALSTUDIES
CROSSSECTIONALSTUDY
Crosssectional(Survey,Prevalence)
Diseasedescription
Diagnosis&staging
Diseaseprocesses&mechanisms
Casecontrol
Causesofdiseases
Identificationofriskfactors
CHARACTERISTICS:
Singledefinedpopulation
PIOwelldefined
Concurrentcontrol(fromsamepopulation)
CohortProspective(Incidence)
Causesofdiseases
Identificationofriskfactors
Naturalhistory,Prognosis
CohortRetrospectiveHistorical
11/29/08
CROSSSECTIONALSTUDY
Bothdescriptive&analytical
Donefordiseasesofslowonset&long
duration
Generatehypothesis&preliminarytesting
Identifypotentialcause/sofanoutcome
CROSSSECTIONALSTUDY
Evidenceforcausalityisonlysuggestive
Nocriteriainselectionofsubjectsconfounders
NocontroloverexposureNoblinding
Nostandardizedoutcomeassessment
Nosimilar/comparableE&Egrps
Servesasaninitialstageofacohortstudy
CROSSSECTIONALSTUDY
USES:
Determine&verifyanevent(Exploratory
study)
Establishbaselinedatatoquantifyhealthstatusof
thecommunity
Evaluatehealthservicedelivery
Determinemagnitudeofadisease
CROSSSECTIONALSTUDY
Advantages
Generalizability
Lesscostly:nofollowup
Lesstimeinvolved
Disadvantages
Notappropriate
Rare
Shortdurationdiseases
Diseasesw/periodsofexacerbations&remissions
11/29/08
CROSSSECTIONALSTUDY
CROSSSECTIONALSTUDY
DESIGN:
EO
EO
NNe
EO
EO
TABLE:
Exposed
Not
exposed
Fromthetargetpopulation,subjectsareselectedthen,exposureand
outcomearemeasuredatthesametime
with
outcome
w/o
outcome
CROSSSECTIONALSTUDY
STATISTICS:
Oddsofbeingexposed
aO
Odds(E)=OROdds(E)=
bO
Oddsofbeingunexposed
cO
Odds(E)=OROdds(E)=
dO
14
14 28
10
12 22
50
odds(E)
OddsRatio=
odds(E)
e.g.
Odds(E)=14/14=1
Odds(E)=10/12=.833
OR=1/.833=1.2
OddsofhavingOinEis1.2
higherthantheE
11/29/08
CROSSSECTIONALSTUDY
TodetermineiftheresanassociationbetE&O,we
have2choices:
CalculateprevalenceofOinEsubjects
PREVALENCE
Measureoffrequency
Prevrate=
andcompareittoprevalenceofOinE
CalculateprevalenceofEinsubjectsw/O
andcompareittoprevalenceofEinsubjO
INCIDENCE
MeasuresthenumberofNEWcasesduringa
specificperiod
Incidencerate=
Interviewquestions
Typeofmeasure
Doyoucurrentlyhaveasthma?
Pointprevalence
Haveyouhadasthmaduringthelast(n)years?
Periodprevalence
Haveyoueverhadasthma?
Cumulativeincidence
BIASESinPREVALENCESTUDIES:
TEMPORALCONSEQUENCES
Possible causes
Incidence Study
Population free of
disease but exposed/
not exposed to
possible causes
Disease or Outcome
Development of cases
overtime
Prevalence Study
Past or present
exposure to possible
causes
Population of
existing cases &
noncases
11/29/08
USESofPREVALENCERATE
Decidingtheneedforhealthcare
Planningofhealthservices
Investigatingpotentiallycausalrelationshipb/
wriskfactors&adiseaseorprognosticfactors
&anoutcome
Assigningaprobabilitythatapatienthasthe
condition
PREVALENCEvsINCIDENCE
LOWPREVALENCEDISEASES:
Shortlived
Rapidlycured
Fatalatanearlystage
HIGHPREVALENCEDISEASES:
Lowmortality
Lowcurerate
Highdisablingeffect
CHARACTERISTICSof
PREVALENCE&INCIDENCE
PREVALENCEvsINCIDENCE
HIGHINCIDENCEDISEASES:
Common
LOWINCIDENCEDISEASES
Characteristics Prevalence
Incidence
Numerator
Denominator
Time
Single pt
Duration of the pd
How measured
Prevalence (cross-sectional)
study
Cohort
Rare
11/29/08
CASECONTROLSTUDY
ExplorestherelationshipbetpriorEtoa
specificriskfactor&thelikelihoodof
developingaparticulardisease.
Retrospectivestudythatcomparesagrpof
cases&1grpofcontrolsw/respecttoa
currentorpreviousexposure
CASECONTROLSTUDY
Populationbasedsampling
Incidentcases
Lessexpensive
Maybecompletedmorerapidly
CASECONTROLSTUDY
CHARACTERISTICS:
observationalwelldefinedpopulation,specific
outcomeandfactor/s
Comparativethecasesandthecontrols
Retrospectivelookbacktofindwhatmaycause
theeffect
Analyticrelationshipofoutcometoprobable
cause/s
CASECONTROLSTUDY
Cases:hospitalrecords,surveillancesystems,
deathcertificates
Controls:sampledfrompopulationthatgave
risetothecases
Hospitalbased
11/29/08
SELECTIONCRITERIAforHOSPITAL
BASEDCONTROLS
TODO:
Variousdiagnostic
groupsnospecificRF
isoverrepresented
Patientsw/acute
conditionsearlierE
couldnothadbeen
influencedbythe
condition
TOAVOID:
Patientswhohave
multipleconcurrent
conditions
Patientsw/dxknownto
berelatedtotheRisk
Factorofinterest
CASECONTROLSTUDY
Weakness
Recallbias
Selectionbias
Validityissue
Matching
confoundingormixingtheeffectsofEtotherisk
factor
statisticalprecisionofestimatesallowingsmaller
samplesize
Timeconsuming
CASECONTROLSTUDY
USES/APPLICATIONS:
Exploringforprobablecauses
Determiningharmfulagentsthatcouldhave
causedapresentcondition
Appropriatefordiseases:
Rare
Chronicorwithlonglatency
CASECONTROLSTUDY
DESIGN:
EO
O
EO
N
EO
O
EO
Fromthesamepopulation,cases&controlareselectedthen,look
backintimetofindprobablefactor/sthatcouldhaveaneffectto
theoutcome
11/29/08
CASECONTROLSTUDY
TABLE:
Exposed Unexposed
Case
Control
CaseExposureProblty
=Exposedcases(a)
Allcases(a+b)
OddsofCaseExposure
=Exposedcases(a)
Allcases(a+b)
Unexposedcases(b)
Allcases(a+b)
=a/b
OddsofControlExposure
=c/d
NESTEDCASECONTROLSTUDY
Population
DevdiseaseDoNOTdevdisease
CASECONTROLSTUDY
STATISTICS:
OddsofhavingtheEinCases&Control
ac
Odds(O)=Odds(O)=
bd
odds(O)a/baxd
OddsRatio==
odds(O)c/dbxc
OR:associationbetweenE&Ounlikelytohave
occurredbychancealone
NESTEDCASECONTROLSTUDY
Advantages:
Dataobtainedbeforeanydiseasehas
developedrecallbias
Riskfactorsarerepresentedthanearly
subclinicaldisease
Moreeconomicaltoconductthancohort
study
CasesControl
11/29/08
BEFOREAFTERSTUDY
BEFOREAFTERSTUDY
Transitiondesign
Retrospectivemeasureoutcomethenassess
exposure
Prospectiveapplyexposurethenmeasure
outcomefollowingexposure
Comparative2groupsonly
MeasurementofoutcomescanNOTbeconcurrent
Same/differentstudypopulation
BEFOREAFTERSTUDY
BEFOREAFTERSTUDY
CHARACTERISTICS:
Studysubjectsideallymustbethesamebeforeand
after(issues:comparable,selectionbiasanddrop
outs)but,maybedifferent
Interventionideallybeundercontrolofinvestigator
(standardized)butisnot,speciallyinHARMstudies
Outcomeisassumednotpresentatthestartofstudy
(before)anddevelopedafterduetoexposure
Differentstudypopulation:
WhenNe1isdonesimultaneouslyw/Ne2convertedtoa
COHORT
USES/APPLICATIONS:
HARMstudiesEhappensatthestart&
beyondthecontrolofinvestigator
Evaluatingeffect/sofnewpoliciesand
strategies(Mgt,BZ,marketing&sales)
11/29/08
BEFOREAFTERstudy
DESIGN:
BEFOREAFTERstudy
TABLE:
Samepopulation
OO
NNeEE
OO
beforeafter
Twodifferentsubjectpopulations,identifyN1&
Ne1forbeforeandN2,Ne2forafter
BEFOREAFTERSTUDY
STATISTICS:
Riskofdevelopingoutcomeduetoexposure
ac
Risk(E)=Risk(E)=
a+bc+d
RiskRatio:R(E)
RR=
R(E)
OUTCOME
presentabsent
Exposed a b
Not
Exposedc d
VALIDATIONOFADIAGNOSTICTEST
Aspecialresearchstudytodetermine
howacertaindiagnostictestcompared
toareferencestandard
PurposeofthisstudydesignisNOTfindingthe
BESTdxtestbutgeneratingdatausefulfor
clinicaldecisionmaking
RR=riskofdevelopingOinEis>/<thanthe
riskofdevelopingOinunexposed
10
11/29/08
ELEMENTSOFAVALIDATIONDIAGNOSTIC
TESTSTUDY
Diagnostictest
Reference(gold)standard
Reproducibility
Blindandindependentcomparison
Studysubjectssuspectstohavedisease
Dataonaccuracyofthetest
VALIDATION
REFERENCE(GOLD)STANDARD
TRUTH
Globallyacceptablecriteriathatadiseaseis
presentorabsent(100%certainty)
Couldbeanyoffollowing:
Histopathology
Anydiagnostictestresult
Consensusofexperts
LEVELSofEVIDENCEforRATING
ACCURACYofDIAGNOSTICTESTS
Level1
Indptinterpretationofdxtestresult
Indptinterpretationofgold/referncestdresult
Subjectsarepatientssuspectedtohavedisease
Reproducibilityofbothdx&stdtests
Samplesize:atleast50patientsw/disease&
50patientsw/outdisease
Level24/5criteriaaremet
Level33/5
Level42/5
Level51/5
Level6NONE
Validation
Diagnostictestunderstudy
Examination(radiography,laboratory,etc.)use
todeterminethepresenceofadisease
Onediagnosticoptionwhendoinggold
standardisnotpossible
Performanceislowerthangoldstandard
(<100%)
11
11/29/08
VALIDATION
STUDYSUBJECTS
Atthestartofstudy:
Noonehasthedisease
Allaresuspects(withsignsandsymptoms)to
havethedisease
Ideally,allundergogoldstandardand
diagnostictests
VALIDATION
INDEPENDENCE&BLINDING
Validation
Reproducibility(method&results)
Boththegoldstandardanddiagnostictests
aredescribedindetailtoallowthosenotpart
ofthestudyduplicatetheprocedure
Generating,readingandinterpretingresults
canberepeatedanywhere
VALIDATION
COEFFICIENTOFAGREEMENT(VARIABILITY)
Thedoerandreaderofthediagnostictest
isnotinfluencedbyanddoesnotknowthe
referencestandardresultandviceversa
Referencestandardanddiagnostictestsare
doneseparatelyandpreferably
concurrently
Intraobservervariabilityinreadings/
interpretationdonebyonepersonat
differenttimes
Interobservervariabilityinreadings/
interpretationofanumberofpersons
12
11/29/08
KAPPASTATISTIC
Expressestheextenttow/ctheobserved
agreementexceedsthatw/cwouldbe
expectedbychancealone(numerator)
relativetothemostthattheobservers
couldhopetoimprovetheiragreement
(denominator)
Kappa=observedagreement(OA)
potentialagreement(PA)
OA=observedchanceagreement
PA=100%chanceagreement
Percentagreement
=41+27/75x100
=90.7%
The2pathologists
agreedon90.7%of
thereadings
Grading
Pathologist
A
Gr II Gr
III
Grding Gr II 41
3
44
Pth B
GrIII 4
27 31
45
30
75
Grading
Pathologist A
%chanceagreement
=26.4+12.4x100
75
Grding
=51.7%
Pth B
The2pathologists
expectedtoagreeby
chanceon51.7%of
thereadings
Totals
by A
Gr2
Gr3
Totals by B
Gr 26.4 17.6 44
II
58.7%
Gr 18.6 12.4 31
III
41.3%
45
30
75
OA=observedchanceagreement
PA=100%chanceagreement
Kappa=90.7%51.7%=39%=0.81
100%51.7%48.3%
Kappa=observedagreement(OA)
Thereisasubstantialagreementbetween
potentialagreement(PA)
the2pathologists
60% 40%
13
11/29/08
Validation
BasicTable
STRATIFIEDQUALITATIVEEQUIVALENTofKAPPA
NUMERICALVALUES
00.2Slight
0.20.4Fair
0.40.6Moderate
0.60.8Substantial
0.81Almostperfect
REFERENCE (GOLD)STANDARD
DIAGNOSTIC
TEST
Validation
BasicStatistics
Sensitivity
a
sn=
a+c
Specificity
d
sp=
b+d
a+d
Accuracy
positive
negative
positive
True (+)
a
False (+)
b
negative
False (-)
c
True (-)
d
SENSITIVITY(Sn)
Sn
ThegreatertheSn,themorelikelythetestwill
detectpersonsw/disease
UsefulclinicallytoR/Othepresenceofadisease
(SnOUT)
Iftestishighlysensitive,anegativeresultwould
excludethepossibilitythatthepatienthasthe
disease
Percentageoffalse()errorislow
a+b+c+d
14
11/29/08
SPECIFICITY(Sp)
RELATIONSHIPBET.PREVALENCE&
PREDICTIVEVALUE
ThegreatertheSp,themorelikelythetestwill
detectpersonsw/odisease
UsefulclinicallytoR/Ithepresenceofadisease
(SpIN)
Iftestishighlyspecific,apositiveresultwould
includeorstronglysuggestthatthepatienthas
thedisease
Percentageoffalse(+)errorislow
Thehighertheprevalence,thehigherthePV
Screeningprogramismostefficientifdirected
toahighrisktargetpopulation
VALIDATION
DATAFORCLINICALAPPLICATION
RELATIONSHIPBET.SPECIFICITY&
PREDICTIVEVALUE
Sp
Predictivevalues
ad
Ppv=Npv=
a+bc+d
Anincreaseinspecificityresultsinamuch
greaterincreaseinpredictivevaluethandid
thesameincreaseinsensitivity
LikelihoodRatios
Forpositiveresultfornegativeresult
ac
a+ca+c
LR(+)==Sn/1SpLR()==1Sn/Sp
bd
b+db+d
15
11/29/08
LIKELIHOODRATIO
ThebiggertheLR+,thestrongerthe
associationbetweenhavinga+testresult&
havingthedisease
ThesmallertheLR,thestrongerthe
associationbetweenhavingatestresult&
NOThavingthedisease
DoNOTvaryaccordingtoprevalenceof
disease
RECEIVEROPERATING
CHARACTERISTIC(ROC)CURVE
RECEIVEROPERATING
CHARACTERISTIC(ROC)CURVE
Ateverypt.alongthisdashed
line,theLR+is1&a+testresultis
equallylikelyforpersonsw/&w/
odisease
Aclinicallyusefuldxtesthasan
ROCcurvethatisfarfromthis
dashedline
Towardthelefthandportion,
thereissteepriseinSnw/only
modestreductioninSp
Towardtherighthandportion,
theslopeofROCisflatlittle
improvementinSnfora
substantialincreaseinSp
RECEIVEROPERATING
CHARACTERISTIC(ROC)CURVE
Summaryindexofoveralltestperformance
calculatedastheareaundertheROCcurve
Highestpossiblevaluefortheareaunderthe
ROCcurveis1almostperfecttest
Thegreaterthearea,thebetterthetest
performance
16
11/29/08
THANKYOU
Formultilevelorcontinuousoutcometest
results,thechoiceofacutoffpointvalue
threshold+willSp(fewerfalse+)butalossof
Sn(>falseresultsormissedcases)
threshold+willSn(fewerfalse)butalossof
Sp(>false+results)
17