Multiple imputation in functional regression with applications to EEG data in a depression study

TL;DR

This paper develops a novel multiple imputation method for handling missing scalar and functional data in EEG-based depression research, enabling more accurate analysis of brain activity's link to depression.

Contribution

It introduces an extension to multiple imputation by chained equations and Rubin's Rules for functional and scalar data, improving analysis of incomplete EEG datasets.

Findings

CSD power asymmetry is associated with depression, modulated by age and sex.

Proposed methods perform well in simulation and real data applications.

Handling missing data with the new approach yields more reliable inference.

Abstract

Current source density (CSD) power asymmetry, a measure derived from electroencephalography (EEG), is a potential biomarker for major depressive disorder (MDD). Though this measure is functional in nature (defined on the frequency domain), it is typically reduced to a scalar value prior to analysis, possibly obscuring the relationship between brain function and MDD. To overcome this issue, we sought to fit a functional regression model to estimate the association between CSD power asymmetry and MDD diagnostic status, adjusting for age, sex, cognitive ability, and handedness using data from a large clinical study. Unfortunately, nearly 40\% of the observations were missing either their functional EEG data, their cognitive ability score, or both. In order to take advantage of all of the available data, we propose an extension to multiple imputation by chained equations that handles both scalar and functional data. We also propose an extension to Rubin's Rules for pooling estimates from the multiply imputed data sets in order to conduct valid inference. We investigate the performance of the proposed extensions in a simulation study and apply them to our clinical study data. Our analysis reveals that the association between CSD power asymmetry and diagnostic status depends on both age and sex.