L4 Basic statistics for test interpretation

Question 1

Sensitivity and specificity is \_\_\_\_\_ based,
while PPV (positive predicted value) and NPV (negative predicted value) are \_\_\_\_\_ based.

Answer 1

disease;

test

Question 2

Sensitivity is defined as?

It should be high in ______ tests?

Answer 2

TP/(TP+FN)

• given a diseased individual, the chance of a positive test
  Screening tests –> if positive –> confirmatory test

Question 3

Specificity is defined as?

It should be high in ______ tests?

Answer 3

TN/(TN+FP)

- given a healthy individual, the chance of a negative test 
Confirmatory tests (do not want to give the wrong diagnosis)

Question 4

Predictive values are all affected by ?

Briefly explain

Answer 4

Prevalence

- if prevelance is low (rare disease) > True positive is low > PPV (positive predicted value) is very low

Question 5

What is PPV (Positive predicted value)?

Answer 5

TP/TP+FP

- given a positive test, the chance of having disease (true positive)

Question 6

What is NPV (Negative predicted value)?

Answer 6

TN/TN+FN

• given a negative test, the chance of healthy (true negative)
• affected by prevelance

Question 7

2 normal distribution curves intersected > Normal + disease

(i) closer to normal
(ii) intersection point
(iii) close to disease

What does it mean if the cut-off is (i)?

Answer 7

= more sensitive (less false negative - because more normal than diseased) but less specific (more false positive )

Question 8

2 normal distribution curves intersected > Normal + disease

(i) closer to normal
(ii) intersection point
(iii) close to disease

What does it mean if the cut-off is (iii)?

Answer 8

= more specific (less false positive - because mostly real positives) but less sensitive (more false negative)

Question 9

Receiver Operating Characteristic (ROC) curves are curves that we can select the best cutoff point

• X-axis: 1- specificity (false positive rate)
• Y-axis: sensitivity (true positive rate)

We compare AUC and a ______ AUC means a better cutoff.

Answer 9

larger

Question 10

What is pre-test odds?

Answer 10

Prevelance/(1-prevelance)

Question 11

Likelihood ratios (LR) is independent of prevalence.
What is Positive LR?
High LR+ means?

Answer 11

LR+ = sensitivity / (1-specificity)
= true positive rate/ false positive rate

High LR+ = high sensitivity and high specificity

Question 12

Likelihood ratios (LR) is independent of prevalence.
What is Negative LR?

Answer 12

LR- = (1-sensitivity)/specificity

= false negative/true negative

Question 13

Post-test odds =?

Answer 13

Pre-test odds x likelihood ratio (LR+)

Question 14

Example 1
Prevalence = 0.1%
False positive rate = 5%
The chance of a person to have a positive result + actually has the disease =?

(asking about the PPV positive predicted value)

Calculate the sensitivity and specificity as well.

Answer 14

PPV = TP/TP+FP

Let there be 1000 people

• number people who really have the disease = 1
• healthy = 999
• FP = 999 x 5% = 50
• TN = 999-50 = 949

Therefore PPV
= 1/(1+50)
= 2%

sensitivity
 = find disease
= TP/TP+FN
= 1/1+0 
=100%

specificity
= negative in healthy 
= TN/TN+FP
= 949/(949+50) 
=95%

Question 15

Example 2

F/34, 4 months pregnant. HIV +Ve in 1st test, -ve in 2nd test, with the 99% specificity and sensitivity of the test.
The prevalence of HIV is 0.1%.

Find the PPV of 1st and 2nd tests.

Answer 15

PPV = TP/(TP+FP)

Let there be 1000 people

• TP = 1
• normal individual = 999
• FP = 999x(1-99%) = 10

PPV of 1st test = 1/(10+1) = 9%

all +ve individual goes to 2nd test - all +ve = 10+1 =11

PPV of 2nd test = 1+ (0.1+1) = 90.9%