Use of computer adaptive testing in the development of machine learning algorithms.

In this issue, Tighe and colleagues present a novel approach for “classifying large amounts of data into useful information.” Their study demonstrates the use of machine learning (ML) algorithms for clinical predictions and highlights the promise of such advances in mining secondary databases. Recently, a new approach to ML, Computer Adaptive Testing (CAT), has been applied to the collection of patient-reported outcomes (PROs). CAT relies on computer algorithms that select the most informative subset of items for each individual from a larger “bank” of self-report items. The algorithm improves efficiency by administering fewer items and reducing response burden. As the choice of items to be administered is tailored to the individual, measurement precision is not sacrificed. The ML algorithms applied by Tighe and colleagues were used to identify accurate and efficient formulas for predicting subsequent femoral nerve blocks as are the CAT algorithms. In CAT, levels of self-reported health outcomes are estimated (predicted) by administering the most efficient subset of items for any given individual. The methodological advances of ML and CAT have the potential to meaningfully impact both research and clinical practice. This editorial describes ML as it relates to the collection of PROs and explores recent applications of this computer-based measurement strategy. In addition, we discuss how the existence of CAT-assessed PROs might improve the prediction of ML algorithms such as those applied by Tighe and colleagues.