Longitudinal Latent Class Modeling for Measurement Error Correction

• Activity Number: 501 - Innovative Methods for Measurement Error Correction
• Type: Topic Contributed
• Date/Time: Wednesday, July 31, 2019 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Caroline P Groth* and David Aaby and Linda Van Horn and Michael Daniels and Juned Siddique
• Companies: Feinberg School of Medicine, Northwestern University and Northwestern University Feinberg School of Medicine and Northwestern University Feinberg School of Medicine and University of Florida and Feinberg School of Medicine, Northwestern University
• Keywords: measurement error correction; epidemiology; latent class regression; self-report; dietary intake

Abstract

Measurement error correction in self-reported diet is typically assessed by considering a single overarching relationship between 24-hour urinary biomarkers and the self-reported dietary intake within a validation study. The results of this relationship are then applied to a new study that lacks biomarker data. Since these relationships may change over time and differ based on participant characteristics, relying upon a single cross-sectional relationship may undermine accurate correction of nutrient intake. To better account for different measurement error relationships, we developed a longitudinal latent class regression model that allows us to form latent classes that represent different measurement error relationship patterns over time between urinary biomarkers and self-reported diet. We allow class membership to be determined based on known characteristics that influence measurement error. Our model can then be applied to studies without biomarkers to correct for measurement error, and assess the true relationship between intake and epidemiological outcomes. We illustrate our method using data on sodium intake from a longitudinal lifestyle intervention trial.