Establishing Measurement Invariance Over Time to Enable Data Harmonization: A Case Study in Environmental Health to Harmonize PFAS Burden Scores from 1999 to 2020

Although commonly used in education and psychology, measurement invariance analysis is limited in environmental health research. In this paper, we develop a model for total exposure to per- and polyfluoroalkyl substances (PFAS), also known as forever chemicals, and harmonize this measure from 1999 to 2020. We use data on seven PFAS from ten cycles of the nationally representative U.S. National Health and Nutrition Examination Survey. These data pose several challenges, including severe positive skew, left-censoring due to detection limits, and time-varying multidimensionality. We describe categorizing PFAS measurements and applying multidimensional item response theory to develop a harmonized, partially invariant model. Item threshold parameters over time reflect the phase-out of legacy PFAS and industry introduction of replacement PFAS in response to regulatory pressure and increased consumer awareness. Results are compared with an approach that uses continuous PFAS measurements. We discuss implications for environmental health research and identify directions for software and methodological advancements.