Clinicians carry out investigations on patients in order to make a diagnosis. A diagnostic test can range from a physical examination, such as measuring blood pressure, to a blood test, such as haemoglobin, or taking a tissue biopsy. Data obtained from these tests can be either continuous, that is, the result can take any value within certain limits, or categorical, for example, normal or abnormal. Continuous data can be plotted as a frequency distribution (Figure 13a). The ability to label test results as coming from those with and without the condition assumes that there is a ‘gold standard’ test, which can truly identify those with and without the condition being tested for. Comparing the diagnostic test with the gold standard allows the test results to be separated into two distributions: those with and those without the condition of interest. As no diagnostic test can perfectly separate these two groups, the two distributions always overlap and a threshold value has to be set, above which people are classified as probably having the condition being tested for, and below which they are classified as normal. The overlapping distributions of results from the two groups, however, means that there will always be some people with the condition whose test result is below the threshold (false negatives) and some people without the condition whose result is above the threshold (false positives). To complicate matters, the threshold value above which results are considered abnormal may have varied over time or depend on the test method used. For example, the definitions of hypertension (high blood pressure) have varied over time, as our understanding of the disease risk attached to particular values has been refined and more treatment options have become available.