Wednesday EBDM

Question 0

Probability of disease is close to zero: test or treat?

Answer 0

Neither

Question 1

What are diagnostic tests used for?

Answer 1

• establish a diagnosis in sick patient
• screen for disease in healthy patient
• provide prognosis
• monitor ongoing therapy
• the results are usually dichotomous

Question 2

Probability of disease is moderate: test or treat?

Answer 2

test first

Question 3

Probability of disease is closer to 100%? Test or treat?

Answer 3

treat first

Question 4

What are two basic features of test?

Answer 4

• reliability and precision

- accuracy

Question 5

What is accuracy?

Answer 5

• ability to hit the target

- truth

Question 6

What is precision?

Answer 6

repeatability

Question 7

How will you know if you are accurate/close to truth?

Answer 7

gold standard test

Question 8

A sick person that is correctly diagnosed as sick?

Answer 8

true positive

Question 9

A healthy person wrongly identified as sick?

Answer 9

false positive

Question 10

A healthy person correctly identified as healthy?

Answer 10

true negative

Question 11

A sick person wrongly identified as healthy?

Answer 11

false negative

Question 12

P( A | B )

Answer 12

probability of A given that B is true

Question 13

P ( Julie’s pregnancy test is + | Julie is pregnant )

TP, FP, TN, FN?

Answer 13

True positive

(probability that the test is positive if she really is pregnant

Question 14

P ( Julie’s pregnancy test is - | Julie is pregnant )

TP, FP, TN, FN?

Answer 14

FN

Question 15

P ( Julie’s pregnancy test is + | Julie is not pregnant )

TP, FP, TN, FN?

Answer 15

FP

Question 16

P ( Julie’s pregnancy test is - | Julie is not pregnant )

TP, FP, TN, FN?

Answer 16

TN

Question 18

Why should Diagnostic tests must be compared to Gold Standard tests?

Answer 18

• they define disease

however, they are often time consuming, dangerous, painful, costly

Question 19

A gold standard test should…

Answer 19

be 100% sensitive and 100% specific

Question 19

If a test is very sensitive (99%), what does this mean about how many FN are present?

Answer 19

very few FN
very seNsitive tests have very few false Negatives
N for negative

Question 20

Sensitivity

Answer 20

probability that a test will be positive if a disease is really there
P ( T+ | D+)
sensitivity = number of TPs / (number of TPs + FNs)

Question 22

SnOut

Answer 22

very Sensitive tests are used to rule Out disease
(100% sensitivity means that the test correctly recognizes all sick people, and correctly rules out disease in 100% of people who do not have it, zero FNs)

Question 22

If you have a very specific test, what does this say about FP?

Answer 22

very specific test has very few FP

Question 23

Specificity

Answer 23

• probability that test will be negative when there is no disease
• very good at picking out negatives
  P( T- | D- )
• can only be calculated initially in samples of individuals who do not have the disease
• very sPecific tests have very few false Positives
• highly specific tests to rule in disease SpIn (specificity = ruled in)

Question 24

SpIn

Answer 24

highly specific tests are used to rule in disease

Question 25

ROC curve | Where is the sweet spot? Test performing at ideal?

Answer 25

Inflection point

Question 26

Limitations

Answer 26

- we don't know if our patients actually have the disease - sensitivity and specificity don't fit clinical reality well - all we usually know is what the test result is and our estimate of how likely disease is P( T+ | D+ ) P( T- | D- )

Question 27

But if you want your test to be more sensitive, go further on ROC curve, what does this do to specificity?

Answer 27

increases False Positives, decreases specificity

Question 28

How should the two by two table for true/false positive/negative be organized?

Answer 28

top: patients with disease | patients w/o disease side: test is positive | test is negative (upper left corner is +/+, lower right corner is -/-)

Question 29

Positive Predictive Value

Answer 29

the proportion of people with a positive test who will actually have the disease - depends on the prevalence of the disease in the population PPV = TP/(TP+FP)

Question 30

Negative Predictive Value

Answer 30

the proportion of people with a negative test who will not have the disease - depends on the prevalence of the disease in the population NPV = TN / (TN+FN)

Question 31

Receiver Operating Characteristic (ROC) Curves

Answer 31

plot used to determine the best cut-off points for 'positive' and 'negative' results - Sensitivity vs 1 - Specificity - asterisk is where the curve would be PERFECT (gold standard); this is upper left of graph

Question 32

What are three important things to note about Gold Standard Tests?

Answer 32

- they are the reference or criterion standard for diagnosis (this test defines the presence or absence of disease) - all diagnostic tests should be evaluated in comparison to a Gold Standard reference test - unfortunately, gold standard tests are usually expensive and difficult to obtain, sometimes dangerous or risky for patients