Correcting Measurement Error and Zero Inflation in Functional Covariates for Scalar-on-Function Quantile Regression

Area of Science:

• Biostatistics
• Wearable Technology
• Public Health

Background:

• Wearable devices generate valuable time-varying biobehavioral data for health research.
• Existing statistical methods struggle to simultaneously address measurement error and excess zeros in this data.

Purpose of the Study:

• To develop a novel statistical framework for analyzing zero-inflated and error-prone functional data from wearable devices.
• To accurately estimate latent health behaviors and their impact on outcomes.

Main Methods:

• Introduced a modeling framework with a subject-specific time-varying validity indicator.
• Employed maximum likelihood estimation, basis expansions, and linear mixed models.
• Utilized joint quantile regression to assess covariate effects.

Main Results:

• The proposed approach significantly enhances estimation accuracy compared to methods addressing only measurement error.
• Joint estimation in quantile regression shows substantial improvements over separate analyses.
• The model effectively corrects for zero inflation and measurement error in step count data.

Conclusions:

• The novel framework accurately models complex wearable sensor data, accounting for inherent data issues.
• This approach provides a more reliable method for using physical activity proxies like step counts in health studies.
• Findings support the use of corrected step count data for understanding childhood obesity and physical activity.