Research

Sensitivity, specificity and the meaning of a positive test

Examine test performance and predictive value using a full two-by-two table.

Build the two-by-two table

Compare the test results against a reference classification. Test participants with a positive result who also have the disease/condition being tested for are classified as True Positives. Participants who do not have the disease/condition but receive a positive test result are classified as False Positives. The remaining two classifications include True Negatives (participants without the disease/condition whose test result was Negative) and False Negatives (Participants with the disease/condition who received a Negative test result).

Accuracies reported in studies rely upon the accuracy of the reference methods used in each study and the manner in which participants were selected. Therefore, a reference classification cannot be presumed to be completely accurate.

Use the correct denominator

Sensitivity = TP/(TP + FN): among people with the condition, the proportion testing positive. Specificity = TN/(TN + FP): among people without the condition, the proportion testing negative. Neither gives the chance of the condition among positive results.

Positive predictive value = TP / (TP + FP) Negative predictive value = TN / (TN + FN). Both are based on how you are classified by your test results, not by your status in relation to the condition. The predictive values will be influenced by both the prevalence of the condition as well as the sensitivity and specificity of the test.

Check the setting and uncertainty

Deliberately selecting roughly equal numbers of cases and controls does not directly provide population predictive values. Sensitivity and specificity can also vary with the positivity threshold, patient mix and disease severity.

Reporting Counts and Uncertainty: A small number of subjects who were identified as being affected in the study can make it difficult to determine the sensitivity of the test. There is often a tradeoff when adjusting a cut-off point that changes the degree of sensitivity compared to specificity. Additionally, there is much clinical interpretation involved such as the patient's symptoms, their prior likelihood of having the condition and further testing to confirm if they actually have the condition. These types of calculations help predict results from either hypothetical examples or actual research studies. They cannot be used to individually diagnose someone.

Worked example: 1,000 hypothetical people

To illustrate assume prevalence = 10% , sensitivity = 90 % and specificity = 95 %.

Among 100 affected people: 90 TP and 10 FN. Among 900 unaffected people: 855 TN and 45 FP. PPV = 90/135 = 66.7%; NPV = 855/865 ≈ 98.8%.

In this example, there is roughly a two-thirds chance that someone who tests positive has the condition, even though the test’s sensitivity is 90%.

Sensitivity relates to the test results when compared to people that do have the condition, inclusive of those individuals that have false negative test results. Positive Predictive Value is related to the proportion of people that have the condition in relation to those that tested positive.

Sources for further reading