Diagnostic and screening tests (e.g. HIV blood test, colonoscopy) respectively confirm a condition or detect high risk individuals. Measures such as sensitivity and predictive values are used describe these tests. They can help calibrate tests (e.g. set criteria for positive diagnosis) or indicate the suitability of tests to different applications (i.e. a less accurate but cheaper test may be more suited to screening than diagnosis). These measures are summarised below:

	With condition	Without condition	Total
Tested positive	a (True positive)	c (False positive)	a + c	Positive predictive value = a / (a+c)
Tested negative	b (False negative)	d (True negative)	b + d	Negative predictive value = b / (b+d)
Total	a + b	c + d	a+ b + c + d
	Specificity = a / (a + b)	Sensitivity = d / (c + d)

Sensitivity & specificity

Sensitivity is the proportion of individuals who have the condition (e.g. diseased) correctly identified by a test.

Sensitivity
= a / (a + b)
= true positives / with condition
= 1 – false negative rate

Specificity is the proportion of individuals without the condition (e.g. non-diseased) correctly identified by a test.

Specificity
= d / (c + d)
= true negatives / without condition
= 1 – false positive rate

There is an inverse relationship between specificity and sensitivity – tightening or relaxing criteria (e.g. cut-off concentration for positive diagnosis in a blood test) to improve one will have the effect of decreasing the other.

Predictive value

Positive predictive value (PPV) or yield, is the proportion of true positive cases out of all test-identified positives.

 PPV = a / (a + c)

Negative predictive value (NPV) is the proportion of true negatives cases out of all test-identified negatives.

NPV = d / (b + d)

While sensitivity and specificity are independent of prevalence of the tested condition, positive predictive value are dependent and decreases as prevalence decreases.

Accuracy

The accuracy of a diagnostic test is the proportion of all cases correctly identified (true positives and negatives).

Accuracy = (a + d) / (a + b + c + d)