ROC analysis

You might have heard of ROC curves, or AUROC as a useful way of understanding whether tests are any use for disease detection. If you aren't familiar with what these are, do some googling first - there are lots of good overview papers freely available - then come back to this guide on how to make them in R with these dummy data.

Right...

6. Produce ROC curves for tests A,B&C

NB. you can choose to display confidence intervals on the plot - check your target journal's preferences, or other publications you're using as a guide.

7. Produce some summary statistics to help compare the ROC curves numerically

eg. > ROC_output# A tibble: 3 × 9 value optimal_cutpoint sensitivity specificity youden AUC AUC_lower_CI AUC_upper_CI p_value <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>1 testA 7 0.657 0.78 0.437 0.781 0.739 0.835 6.86e-122 testB 10 0.59 0.57 0.16 0.636 0.579 0.713 8.74e- 43 testC 9.5 0.63 0.6 0.23 0.639 0.575 0.701 6.29e- 4

8. How did the computer determine the "optimal cutpoint"?

Consider how this may vary for a clinician considering using these tests for screening, vs. as part of a diagnostic assessment.

Use this R code to help you with tasks 6-8

Page updated

Google Sites

Report abuse