Diagnosis (part 2)

Question 1

When calculating CIs for sensitivity/specificity, what happens with increased sample size?

Answer 1

CIs get smaller

Question 2

What are people who don’t have disease but are positive on the test?

Answer 2

False alarms (false positives)

Question 3

How is the probability of a false alarm calculated?

Answer 3

1 - specificity

Question 4

What is this?

Answer 4

A receiver-operator curve (ROC)

Question 5

What is the point of an ROC?

Answer 5

• Allows visual assessment of the usefulness of a diagnostic test
• A useless test would just be a straight line from bottom left to top right

Question 6

Where is the appropriate cut-point?

Answer 6

Question 7

What is the ROC AUC?

Answer 7

ROC area under the curve.

• Global assessment of the performance of a diagnostic test
• Probability that a random person with the disease has a higher value of the measurement than someone without the disease

Question 8

What is the ROC used for?

Answer 8

• Comparing the results of 2+ tests
• If the curve of one test lies wholly above another, it is better.

Question 9

If CIs have cross-over, which test should you choose?

Answer 9

• Generally the higher average
  
  • But if a really low CI, consider the highest other with the lowest CI

Question 10

What do -ve likelihood ratios mean and how are they calculated?

Answer 10

How much more likely a negative test finding is in people who have the condition than in those who don’t.

-LR = (1-sensitivity)/specificity

Question 11

How is pretest probability estimated?

Answer 11

Usually best estimate is the prevalence of the condition in the population of interest.

Can use 2x2 table (e.g. Present on MRI/total participants)

Question 12

How do we calculate post-test probability?

Answer 12

Use prevalence (pre-test probability) and LR to plot on a Fagan’s Nomogram.

OR

[Pre-test probability (p)]

Pretest odds = p / (1-p)

Post-test odds (o) = pretest odds x likelihood ratio

Post-test probability = o / (1+o)