The Analyse-it Method evaluation module includes ROC curve analysis, useful in clinical medicine for evaluating the performance of a diagnostic method.

You can plot upto 3 diagnostic tests on the same ROC chart to compare their effectiveness & specificity / sensitivity. To use the ROC curve procedure your dataset must contain two or more variables:

• 1 variable must indicate the true state or classification of the subject - ie.whether the subject is considered normal (no disease) or abnormal (diseased).

• 1-3 variables containing the observations of the diagnostic test(s) you want to evaluate.

Entering your data in Excel

Lets use a simple example dataset to demonstrate how to use the ROC curve procedure. The example dataset we will use is shown alonside.

In this example, Disease is the variable indicating the true state/classification of the subject. You can use whatever names you like for the two categories indicating normal and abnormal. In the example we have used Present & Absent, but you might use diseased & clear, positive & negative, 0 & 1 - or whatever you want.

TestResult is the variable containing observations from the diagnostic test. Your dataset can contain up to 3 variables containing results from diagnostic tests - future versions of Analyse-it will allow more than 3 tests to be plotted & analyzed.

Setting the Dataset & Variable properties

Now, first you must setup the dataset properties (see below). To do so, select a cell in the dataset, then choose Data | Dataset from the menu. For the Layout, select list (where each row is a case, and each column a variable). Choose OK.

Next (and this is the bit that is probably confusing you), you must tell Analyse-it that the variable indicating the true state/classification is a categorical variable. To do so, select a cell in the Disease data column, and choose Data | Variable from the menu (see below). In the Variable dialog box, set the type to Nominal, and choose OK.

Analyzing the data & plotting the ROC curve

Now you are ready to use the ROC curve procedure. Choose Analyse | Diagnostic Testing | ROC curves from the menu bar. The run-test dialog box is shown (below):

First you must first select the diagnostic tests (in the Variables list box) you want to plot on the ROC curve. In this example we have only entered results from 1 diagnostic test (the TestResult variable), but you can plot up to 3 diagnostic tests on the same ROC curve.

Next select the variable containing the true state / classification of the subject. In this example it's the Disease variable. Next indicate which category in the Disease variable indicates an abnormal subject. In our example, the Disease variable contains 2 categories - Present and Absent - and the category Present indicates presence of disease (ie. an abnormal subject). So in the Abnormal indicated by drop-down list, choose Present.

Finally click OK. Analyse-it presents the results of the ROC analysis.