Predictive Values

We have looked at the definitions of normal, sensitivity and specificity in the context of a test.

We saw the use of sensitivity (SNOUT) and specificity (SPIN) in the context of symptoms/signs or diagnostic tests.

For a clinician, however, sensitivity and specificity, although useful, do not provide all the information they need.

Sensitivity provides information on the ability of the test in persons with the disease…and

Specificity provides information on the ability of the test in persons without disease.

In other words, the sensitivity and specificity help to understand how good the test is at identifying people with the disease and people without the disease.

What the clinician might find useful is, however,

“What proportion of the positive tests predict or correctly identify those with the illness or condition of interest”? What proportion of the negative tests predict or correctly identify those who do not have the disease”?

In other words, the clinician is interested in

“If the test is positive, what is the chance or probability that the person has the disease”?

“If the test is negative, what is the chance or probability that the person does not have the disease”?

These questions can be answered by the positive and negative predictive values

The denominator for the predicitve values is the number of people with a positive or negative test (the denominator for sensitivity and specificity was the number of persons with and without the disease)

The positive predictive value is 215/231=93.07%

The negative predictive value is 114/129=88.37%

Caution

There are two factors that affect the predictive values

1. The prevalence of the disease- As the prevalence of the disease falls, the positive predictive value decreases. Thus, a test with a high positive predictive value will perform better in a high risk population (where the prevalence is higher) than in a low risk or population of apparently normal people. For instance, the predictive value of endoscopy in a surgical gastroenterology clinic will be much higher than when endocscopy is applied in a low risk OP
2. When the prevalence of the disease is low, the specificity of the test influences the predictive value

Let us explore the relationship of prevalence with positive predictive value

Consider the following parameters

1. The population size is 10,000
2. The prevalence of disease X is 1%
3. The sensitivity of the test used is 99%
4. The specificity of the test used is 95%

Let us build the blank 2 X 2 table step by step with the information we have

Step 1- let us put the total population studied

Step 2: Let us estimate the number of diseased persons (1% prevalence)

Step 3: Let us estimate the number of people without the disease

Step 4: Let us estimate the number of true positives (using the sensitivity -99% of 100 persons with the disease)

Step 5: Let us estimate the number of true negatives (using the specificity-95% of 9,900 persons with the disease)

Step 6: Now complete the number of false positives, the number of false negatives, the number of positive tests and the number of negative tests

Step 7: Estimate the positive predictive value (PPV)

PPV= 99/594=17%

Step 8: The negative predictive value is 9405/9406= 99.99%

Now look at what happens when the prevalence increases to 5% keeping the other parameters constant

1. The population size is 10,000
2. The prevalence of disease X is 5%
3. The sensitivity of the test used is 99%
4. The specificity of the test used is 95%

Try creating the table and estimating the predictive values. Work step by step.

You should get a table like this

The PPV is 495/970 =51.03%

Thus, a test is more efficient when it is targeted towards a high risk population

Applications

1. Screening a high risk subgroup is better than screening a low risk subgroup
2. A test that performs very well in a high risk clinic may not perform as efficiently in a low risk clinic (example- fetal surveillance in a high risk pregnancy clinic may pick more cases than fetal surveillance in a routine antenatal low risk OP)

IMPORTANT

The test result has to be interpreted in the context of the population it is applied to.

Relation of specificity with predictive values when the prevalence is low

Consider the following parameters

1. The population size is 10,000
2. The prevalence of disease X is 1%
3. The sensitivity of the test used is 80%
4. The specificity of the test used is 90%

Build the table yourself…(try to practise that)

What is the PPV? 80/1070=7.47%

Let us change just the specificity of the test used

1. The population size is 10,000
2. The prevalence of disease X is 1%
3. The sensitivity of the test used is 80%
4. The specificity of the test used is 95%

The PPV is now 80/575=13.91%

Let us change just the sensitivity of the test used (instead of the specificity)

1. The population size is 10,000
2. The prevalence of disease X is 1%
3. The sensitivity of the test used is 95%
4. The specificity of the test used is 90%

What is the PPV? 95/1085=8.76%

Note:

The Increase in PPV was larger when a test of greater specificity was used than when a test of greater sensitivity was used (applies when the prevalence of the condition of interest is low)

Application

How do we then apply the knowledge of predictive values

1. We can tell the patient- XX% of those with a positive test result may have the target condition of interest
2. We can tell the patient- XX% of those with a negative test result may not have the target condition of interest
3. We can be aware that the predictive ability of the test will decrease if it is applied in a low risk compared to a high risk population
4. We can be aware that we might be better off choosing a more specific than sensitive test when we do a screening program (as population prevalence for conditions that are screened are usually on the lower side).