Why we cannot calculate an ROC curve in cost sensitive learning?

Question

In the Applied Predictive Modeling book, cost sensitivity learning approach, the author(s) write:

One consequence of this approach is that class probabilities cannot be generated for the model, at least in the available implementation. Therefore we cannot calculate an ROC curve and must use a different performance metric. Instead we will now use the Kappa statistic, sensitivity, and specificity to evaluate the impact of weighted classes.

Can you explain to me why not ROC/AUC but Kappa Statistic, sensitivity and specificity instead? I think sensitivity or specificity is also ROC or AUC?

Link for the book: https://cloudflare-ipfs.com/ipfs/bafykbzacedepga3g6t7b6rq6irwhy5gzpc47bamquhygup4eqggidvkjcztqs?filename=Max%20Kuhn%2C%20Kjell%20Johnson%20-%20Applied%20Predictive%20Modeling-Springer%20%282013%29.pdf