Oct 25

Question 1

What can tests that perform fairly well against the gold standard test be used for?

Answer 1

-not thrown out completely -can be used as a pre diagnostic test (not as good as the diagnostic one but is good enough to pick up cases then use the gold standard to for sure diagnose)

Question 2

What are consequences of a false positive?

Answer 2

-caused by low specificity -emotional: telling someone they are sick when in fact they are healthy -financial: re-testing or needlessly treating

Question 3

What are consequences of a false negative?

Answer 3

-caused by low sensitivity -sick people will go untreated

Question 4

What is the receiver operating characteristic curve?

Answer 4

-useful for comparing screening or diagnostic tests -test closer to the top left corner is the best (high TP and low FP) -anything crossing 0.9 is considered excellent, 0.8 is considered good

Question 5

What is positive predictive value?

Answer 5

• what is the probability you actually have the disease if you test positive (denominator is positive tests)
• TP/(TP+FP)

Question 6

What is negative predictive value?

Answer 6

• probability that the disease is not present when the test is negative (denominator is negative tests)
• TN/(TN+FN)

Question 7

How would you predict a PPV of 20%?

Answer 7

• I just got a positive result back
• What are the chances my patient actually has the disease?

1 in 5

Question 8

How would you predict a 90% NPV value?

Answer 8

-what are the chances my patient doesn’t actually have the disease?

9 in 10

Question 9

What is reliability?

Answer 9

• replicability of the test results over repeated administrations of the test
• reliable test produces the same result over multiple administrations when the patient’s underlying condition remains unchanged throughout the time period of the administrations

Question 10

What are threats to reliability?

Answer 10

• intrasubject: normal biological variability in human characteristics (eg BP)
• intraobserver: variability in two or more readings of the same test by the same examiner

-interobserver: variability in the same test reading by two or more examiners

Question 11

How is reliability measured?

Answer 11

• percent agreement
• Kappa

Question 12

How is percent agreement calculated? What are problems with this?

Answer 12

• take concordant pairs between physician A and B
• add them and divide by total
• can be inflated if physicians mostly agree on a negative result but don’t agree much on a positive result
• only agreed on the negative part not the positive part
• chance agreement: examiners use different criteria to arrive at the same result

Question 13

How is Kappa calculated?

Answer 13

-corrects for chance agreement

• how much of the agreement was caused by chance alone?
• calculate percentage of cases that physician A determined were positive and negative
• take the proportions from the first part and multiply them by how many cases physician B determined to be positive and negative
• take concordant boxes and add them and divide by the total which gives the value of the agreement by chance alone
• then calculate Kappa by taking observed agreement-agreement by chance/100-agreement by chance
• 0.75 is good Kappa
• below 0.4 is not good