Clinical Epidemiology (Midterm 3)

Front 1

Sensitivity

Back 1

= a / (a+c)

= true positive/ all actual positive

Front 2

Specificity

Back 2

= d / (b+d)

= true negative/ all actual negative

Front 3

Positive Predictive Value

Back 3

= a / (a+b)

= true positive/ all tested positive

Front 4

Negative Predictive Value

Back 4

= d / (c+d)

= true negative/ all tested negative

Front 5

Diagnostic accuracy

Back 5

= Percent agreement = [(a+d) / (a+b+c+d)] * 100%

= percentage of true results out of all results

Front 6

Chamberlain's percent positive agreement

Back 6

= (a+d) / (a+b+c)

• In the context of reliability, useful when most subjects are negative, which would artificially inflate diagnostic accuracy
  
• In the context of accuracy, useful when the population has many true negatives, which would artificially inflate diagnostic accuracy
  

Front 7

OARP

Back 7

= a/b

= true positive/ false positive

Odds of being Affected Given a Positive Result

Front 8

DOR

Back 8

= ad/bc

= true results/ false results

Diagnostic Odds Ratio: summary measure of test effectiveness

Front 9

Youden's J

Back 9

sensitivity + specificity - 1

Front 10

PSI

Back 10

PPV + NPV - 1

Predictive Summary Index

Front 11

Likelihood ratio for a positive test (result)

Back 11

= sensitivity / (1-specificity)

Front 12

Likelihood ratio for a negative test (result)

Back 12

= (1-sensitivity) / specificity

Front 13

Kappa statistic

Back 13

= (O-C) / (1-C)

where O =(a+d) / (a+b+c+d)

and C = [(a+c)(a+b) + (b+d)(c+d)] / [(a+b+c+d)^2]

------------------------

Use an estimate table with the products of the margins to calculate C the same way as O for an estimate table. May be easier.

Front 14

Post-test Odds

Back 14

= pre-test odds * LR

Front 15

Pre-test Odds

Back 15

If you have no other information about a subject healthstatus before administering a test, prevalence servedas the subject’s pre-test probability of disease

Front 16

Interpret Kappa

Back 16

• Numerator: Observed agreement beyond chance
  
• Denominator: Maximum agreement possiblebeyond chance
  
• Kappastatistic quantifies how much beler their levelof agreement is than that which results fromchance alone as a proportion of the maximalagreement beyond chance.
  
• Scale of -1 to 1, with 1 being the best and -1 being the worst generally
• Good Kappa is > 0.7

Front 17

Interpret Youden's J

Back 17

• This is a summary measure of a test’s validity – howclose the test results (T+/T-) correspond to the truth(D+/D-).
  
• It’s theoretical range is -1 to 1, though a practicalrange for a well functioning test is 0 to 1.• Youden’s J is negative when a test’s results are misleading,when they are negatively associated with the truth
  
• Assigns equal weight to sensitivity and specificity• Thus, in settings where you care more about one than theother, you should use the weighted version of Youden’s J.
  

Front 18

Youden's J: 1/J

Back 18

= NND = Number of diseased patientsneeded to examine in order to correctlydiagnose one person with the disease.

Front 19

Interpret PPV

Back 19

If the test result is positive in thepatient, it is the probability the patient actually has thedisease.

Front 20

Interpret NPV

Back 20

If the test result is negative in thepatient, it is the probability the patient actually doesnot have the disease

Front 21

Interpret PSI

Back 21

• Summary measure of the predictability of diseasestatus for a diagnostic test.
  
• Specifically, PSI is the net gain in certainty of a test’spredictability of disease status • The pre-test (prior) probability of disease is the prevalence• A gain in certainty that disease is present occurs when theposterior probability of disease presence (=PPV by Bayes’sRule) exceeds the prior probability of disease• A gain in certainty that disease is absent occurs when theposterior probability of no disease (=NPV by Bayes’s Rule)exceeds the prior probability of no disease (1-prevalence)
  

Front 22

1/PSI

Back 22

= NNP = Number of test-positivepatients needed to examine in order tocorrectly predict the diagnosis of one personwith a positive test result

Front 23

Interpreting Likelihood Ratios

Back 23

The LR reflects the magnitude of evidence that aparticular test result (t) provides in favor of thedisease being present relative to the disease beingabsent

• LR = 1: The test result is equally likely in patients withand without the disease
  
• LR > 1: The test result is more likely among patientswith the disease than without
  
• LR < 1: The test result is more likely among patientswithout the disease than with
  

Front 24

If you want to avoid false positives and invasive/costly follow-up...

Back 24

maximize specificity

Front 25

If you want to avoid false negatives and hence catch a disease early...

Back 25

maximize sensitivity

Front 26

Are sensitivity and specificity inherent to a given test

Back 26

No, since they depend on the degree of diseaseseverity in a population

Sensitivity and specificity condition on the totalsof true disease status (a+c for sensitivity and b+dfor specificity), so they are independent of thedisease prevalence if the gold standard is prefect

Front 27

Are PPV and NPV are dependent on theprevalence on the disease

Back 27

Yes, as they do notcondition on the totals of true disease status

PPV and NPV must be interpreted within theprevalence context of the popula9on of interest

Front 28

Positivity criterion

Back 28

cutoff for a positive test

Front 29

Relation between sensitivity and specificity

Back 29

• Increasing sensitivity decreases specificity
• Increasing specificity decreases sensitivity
  

Front 30

Receiver operator characteristic (ROC) curves

Back 30

The ROC curve is a plot of the true positive rate (i.e.sensitivity) versus the false positive rate (1-specificity)

Front 31

Interpreting ROC Curves

Back 31

• The closer an ROC curve moves to the upper leftcorner, the more accurate a test it represents– i.e. the upper left hand corner represents a test with 100%true positivity and 0% false positivity
  
• As the (cut-point) criterion for a test becomes morestringent, the point on the curve corresponding tosensitivity and specificity moves down and to the left(lower sensitivity, higher specificity)
  
• As the criterion for a test becomes less stringent, thepoint on the curve corresponding to sensitivity andspecificity moves up and to the right (highersensitivity and lower specificity)
  

Front 32

measure of intra-raterreliability is needed

Back 32

When one person measures the same itemtwice (or more)

Front 33

measure of interraterreliability is needed

Back 33

When two (or more) people measure the sameitem once (or more) each

Front 34

Coefficient of variation

Back 34

= (standard deviation / mean) x (100%)

Front 35

Interpret Coefficient of variation

Back 35

• generally used to report thereliability of laboratory or assay measurements ofcontinuous variables
  
• most useful in comparing the variability of severaldifferent samples, each with different arithmetic means
  
• This is true because higher variability usually is to beexpected when the arithmetic mean increases--the C.V.accounts for this variability
  

Front 36

Utility of Bayes’ theorem

Back 36

• Bayes’ theorem demonstrates, in probabilisticterms, that the posterior probability of an eventcan be determined from knowledge of sensitivity,specificity, and prevalence.
  
• It also makes more explicitly obvious than theformulae presented earlier for PVP the role ofprevalence in the determination of posteriorprobabilities.