ISOQOL Workshop 2022

Machine-Learning Methods for Differential Item Functioning in Patient-Reported Outcomes

Our goal is to introduce data-driven methods to detect differential item functioning (DIF) in patient-reported outcomes (PROs). The workshop will focus on item-response theory (IRT) models for DIF detection. Extensions of conventional IRT models, which are based on unsupervised machine-learning methods, will be used to explore heterogeneity in PRO responses. The workshop will also briefly explore the use of unsupervised methods to examine response shift in longitudinal data.

Workshop participants will benefit from new insights about the potential role of machine-learning methods as a component of the toolkit for PRO data analyses. Unsupervised machine-learning methods are advantageous for assessing variable importance in PROMs data that contain large numbers of variables (i.e., features) to characterize patients or for assessing the sensitivity of individual items in large items banks to DIF.

Learning Objectives

  1. To examine methods to investigate differential item functioning (DIF) in high-dimensional data (i.e., many PRO scale items or many patient-level variables) and demonstrate the application of machine-learning models for DIF exploration and detection.
  2. To describe the types of data and research problems that will benefit from the application of machine-learning models for DIF detection.
  3. To practice the implementation of machine-learning methods using existing software packages, with a particular emphasis on R software.

Workshop Facilitators

Dr. Lisa Lix – University of Manitoba, Winnipeg, Canada

Dr Tolu Sajobi – University of Calgary, Calgary, Canada

Dr. Yuelin Li – Memorial Sloane Kettering Cancer Center, New York, United States

Workshop Materials

Find all resources (slides and codes) for the ISOQOL conference workshop 2022. Click the links below to download the file. 

Introduction

NB: Please save the Case Example Dataset as a CSV file before importing it into R

 

Machine-Learning Methods for Patient-Reported Outcomes Data

Item Response Theory Models for Detection Differential Item Functioning

Item-Focused Machine-Learning Models for Detection of Differential Item Functioning

Person-Focused Machine-Learning Models for Detection of Differential Item Functioning

Extension of Machine-Learning Methods to Detect Response Shift in Patient-Reported Outcomes Data

Closing