Modeling Heterogeneity in Multivariate Functional Data with Applications to Ambulatory Blood Pressure Monitoring Restricted; Files & ToC
Parker, Rachel (Summer 2026)
Abstract
Ambulatory blood pressure monitoring (ABPM) provides a detailed characterization of blood pressure (BP) over a 24-hour period, enabling the study of temporal patterns that are not observable through traditional clinic-based measurements. Blood pressure is inherently dynamic, yet most existing analyses rely on static summary measures that obscure clinically important temporal features and variability patterns associated with cardiovascular risk. Motivated by ABPM data from the Mechanisms Underlying Stress and Emotions (MUSE) study in African American women, we develop statistical methodology for modeling and characterizing heterogeneity in multivariate functional data.
In Aim 1, we introduce a flexible semiparametric framework to jointly model the mean and variability of BP trajectories using generalized additive models for location, scale, and shape (GAMLSS). Unlike approaches that treat variability as a secondary summary, this framework models both components (mean and variance) as smooth functions of time and covariates. Results demonstrate that higher obesity status is associated with elevated systolic blood pressure, attenuated nocturnal dipping, and increased variability across the 24-hour period. These patterns are further exacerbated by poor sleep quality, highlighting the joint influence of cardiometabolic and behavioral factors on both the level and stability of blood pressure.
In Aim 2, we develop a multilevel multivariate functional principal component analysis (FPCA) framework to quantify variability in functional data with replicate measures. By jointly modeling systolic and diastolic blood pressure across repeated measurements, the proposed approach decomposes variability into between-subject and within-subject components that evolve over time. This framework enables estimation of time-varying variance components and the functional intraclass correlation coefficient (ICC). Additionally, the approach extends to variance decomposition for composite outcomes, such as mean arterial pressure (MAP). Application to ABPM data reveals distinct variability patterns across BMI groups, with higher obesity categories exhibiting more persistent between-subject variability and reduced nocturnal declines.
In Aim 3, we propose a multilevel multivariate nonparametric functional clustering framework for identifying subgroups based on full blood pressure trajectory patterns. The proposed method simultaneously accounts for functional structure, multivariate dependence, and multilevel variability by leveraging a multivariate FPCA representation. When applied to ABPM data, we identify three clinically meaningful subgroups characterized by differences in both trajectory shape and variability, highlighting heterogeneity not captured by methods based on averaged trajectories.
Collectively, these contributions provide a unified framework for analyzing complex multivariate and multilevel functional data. By leveraging the full 24-hour trajectory, this work advances methodological development and provides new insights into blood pressure dynamics in a high-risk population.
Table of Contents
This table of contents is under embargo until 20 August 2028
About this Dissertation
| School | |
|---|---|
| Department | |
| Degree | |
| Submission | |
| Language |
|
| Research Field | |
| Keyword | |
| Committee Chair / Thesis Advisor | |
| Committee Members |
Primary PDF
| Thumbnail | Title | Date Uploaded | Actions |
|---|---|---|---|
|
|
File download under embargo until 20 August 2028 | 2026-06-30 18:20:45 -0400 | File download under embargo until 20 August 2028 |
Supplemental Files
| Thumbnail | Title | Date Uploaded | Actions |
|---|