This is the third part of a three-segment series on ROC Analysis
Code Development
This particular client had a significant amount of data and was interested in varying the types of analyses that could be extracted from that data set. Â As such, we knew that to complete this project efficiently and effectively, it would be our advantage to develop the analysis not as a spreadsheet (which is typical) but as a Mathematica program, which although is more time-consuming on the front end, pays some serious dividends in time-saving as it is repeatedly used. Â Additionally, especially for this engagement, we knew that creating this program would produce the flexibility needed to quickly change parameters according to the client’s changing specifications. Â Working smart, not hard, not only allows us to fully explore the potentials within a client’s data set, but allows us to provide significant value at reasonable cost.
Mathematica allows good integration between flexible coding and attractive figure preparation, which reduces the amount of time needed to tweak images.
Conclusion
The determination of AUROC allows the client to have a single number in mind when comparing their diagnostic or predictive method with other similar methods in the field. Â The higher the AUROC, the better the diagnostic is overall. Â Additionally, from the ROC curve, we can see the point of interest where the diagnostic must meet the minimum standard (80TP/20 FP), and as long as the ROC curve at least intersects (or preferably rises above that) then the diagnostic will be able to meet that standard. Â Thus armed with both an independent analysis of their diagnostic, and more importantly a rigorous analysis of their data that was focused on an actual need, they were able to walk away from this engagement with something they could actually use.