We use stratified K-fold cross-validation a lot nowadays, but wonder where it started to got prominent (as an idea)?

This paper, A Study of CrossValidation and Bootstrap for Accuracy Estimation and Model Selection, by Rohavi 1995 explains how when in the older past, researchers tend to use Leave-one-out cross validation (wiki please), and how K-fold (k=10/20, though I use 5 generally) actually yield better results.