Reference: Anatomic and Clinical Pathology Board Review. Atif A. Ahmed & Ronald M. Pryzgodzki
How do I choose the "best test"?
Of course, to answer such a question, you have to keep in mind the specific end-goal that is desired as a result of the test. Let's play a 'game' to illustrate this.
Would you rather have a test that...
OPTION 1: ... will pick up every single person with the disease (a highly sensitive test), but may result in a lot of false positives?
or
OPTION 2: ... is positive only in the presence of disease (highly specific), but some patients with the disease may not have this particular finding so the test may be negative?
In the real world, we try to find a balance between these two extremes. Ideally, the "best test" is one that is both highly sensitive and specific for the disease.
In screening programs, sensitivity is the most important characteristic of test performance.
Confirmatory tests require high specificity
How do I interpret a ROC Curve?
ROC curves compare the sensitivity and specificity of various tests. They are prepared by calculating the sensitivity and specificity of a test at a number of different cutoff points, then plotting these on a graph.
y-axis = TRUE-positive rate or sensitivity
x-axis = FALSE-positive rate (100 – specificity)
(NOTE: In some versions of the plot, specificity is plotted on the x axis, but in that case the numbers on the x axis go from 100 on the left to 0 on the right.)
Using this method, if the outcome of a test was due to chance alone (i.e. you could flip a coin and it either be pos or neg), it would be depicted by a straight line running from the lower left-hand corner to the upper right-hand corner of the graph.
In evaluating an ROC curve comparing more than one test, several features of the graph can be used to compare test performance against expectations. Inspection of the graphs is the best way to identify tests that may perform better as screening tests (high sensitivity) or confirmatory tests (high specificity).
1. Area under the curve
(AUC; i.e., the proportion of the square below and to the right of the ROC curve)
AUC is a common way to evaluate overall test performance
AUC by itself does NOT determine which test is better for screening or for confirmation of disease. This is because two tests can have the same AUC but may have different shapes to their ROC curves indicating which test might serve better as a screening test or a confirmatory test. Since a straight line from the lower left to upper right corners (performance of chance alone) will include an AUC of 0.50, an AUC of 0.6 is only slightly better than chance alone.
The HIGHER the AUC (with 1.0 being a perfect test), the BETTER the test performance
2. Test Efficiency
Efficiency refers to the proportion of all persons who are correctly classified by a test (true positive + true negative/total tested).
Efficiency will be HIGHEST at that point closest to the UPPER LEFT-HAND CORNER of the graph.
(Examples/pics will be posted soon)
コメント