Loading…
Interpretation of diagnostic data: 4. How to do it with a more complex table
A more complex table is especially useful when a diagnostic test produces a wide range of results and your patient's levels are near one of the extremes. The following guidelines will be useful: Identify the several cut-off points that could be used. Fill in a complex table along the lines of T...
Saved in:
Published in: | Canadian Medical Association journal 1983-10, Vol.129 (8), p.832-835 |
---|---|
Format: | Article |
Language: | English |
Subjects: | |
Online Access: | Get full text |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | A more complex table is especially useful when a diagnostic test produces a wide range of results and your patient's levels are near one of the extremes. The following guidelines will be useful: Identify the several cut-off points that could be used. Fill in a complex table along the lines of Table I, showing the numbers of patients at each level who have and do not have the target disorder. Generate a simple table for each cut-off point, as in Table II, and determine the sensitivity (TP rate) and specificity (TN rate) at each of them. Select the cut-off point that makes the most sense for your patient's test result and proceed as in parts 2 and 3 of our series. Alternatively, construct an ROC curve by plotting the TP and FP rates that attend each cut-off point. If you keep your tables and ROC curves close at hand, you will gradually accumulate a set of very useful guides. However, if you looked very hard at what was happening, you will probably have noticed that they are not very useful for patients whose test results fall in the middle zones, or for those with just one positive result of two tests; the post-test likelihood of disease in these patients lurches back and forth past 50%, depending on where the cut-off point is. We will show you how to tackle this problem in part 5 of our series. It involves some maths, but you will find that its very powerful clinical application can be achieved with a simple nomogram or with some simple calculations. |
---|---|
ISSN: | 0008-4409 |