Advances in Modern Longitudinal Data Analysis and Applications

Nikki Freeman Chair
Duke University
 
Thursday, Aug 8: 10:30 AM - 12:20 PM
5189 
Contributed Papers 
Oregon Convention Center 
Room: CC-E148 

Main Sponsor

Biometrics Section

Presentations

A Joint Normal-Ordinal(Probit) Model for Ordinal and Continuous Longitudinal Data

In biomedical studies, continuous and ordinal longitudinal variables are frequently encountered. In many of these studies it is of interest to estimate the effect of one of these longitudinal variables on the other. Time dependent covariates have however several limitations; they can for example not be applied when the data is not collected at fixed intervals. The issues can be circumvented by implementing joint models. In a joint model, both variables are modeled with a random-effects model, and the random effects are allowed to correlate. We propose a normal-ordinal(probit) joint model. First, we derive closed-form formulas to estimate the manifest correlations between the responses as observed. In addition, we derive the marginal model, where the interpretation is no longer conditional on the random effects. As a consequence, we can make predictions for a subvector of one response conditional on the other response and potentially a subvector of the history of the response . Next, we extend the approach to a high-dimensional case with more than two ordinal and/or continuous longitudinal variables. The methodology will be presented by means of a case study. 

Keywords

Longitudinal data analysis

Joint model

Random effects model

Time dependent effects

Probit link 

View Abstract 1948

Co-Author(s)

Geert Molenberghs, Universiteit Hasselt & Katholieke Universiteit Leuven
Steffen Fieuws, Biostatistical Centre
Geert Verbeke, I-Biostat

First Author

Margaux Delporte, KU Leuven

Presenting Author

Margaux Delporte, KU Leuven

Associating Longitudinal Trajectory as a Covariate to Longitudinal Outcomes via a Joint Model

Research on progressive disorders typically involves understanding if the change in biomarker levels can monitor the worsening in medical conditions over time. Their relationship can be quantified by fitting a mixed model for longitudinal medical outcomes with the subject-specific rate of change in biomarker as a fixed effect. As the true long-term change in the biomarker levels is unobservable, biased estimates and invalid inferences may arise if it is replaced by the estimated random slope from a mixed effects model for biomarker measurements. We thus propose a joint modeling method where both longitudinal biomarker measurements and medical outcomes are modeled by linear mixed effects models. We show that the resulting estimators are consistent and asymptotically normal with a sandwich variance estimator. The method is evaluated via simulations and applied to analyze the association between the rate of change in the cerebrospinal fluid biomarker measurements and cognitive decline for participants with mild cognitive impairment in the Alzheimer's Disease Neuroimaging Initiative. 

Keywords

Longitudinal data

Measurement Error

Biomarker 

View Abstract 3245

Co-Author

Sharon Xie, University of Pennsylvania, Perelman School of Medicine

First Author

Yuen Tsz Abby Lau

Presenting Author

Yuen Tsz Abby Lau

Dynamic Latent Factor Models To Infer Dietary Patterns From Longitudinal Nutrition Survey Data

A growing body of research has shown that poor diet is a leading risk factor for death, especially in connection with chronic diseases such as cardiovascular disease. However, these studies are limited because they use simplistic measures of diet measured at a single timepoint. To address this issue, we develop a Bayesian dynamic latent factor model that measures how multivariate dietary patterns change over time. Our approach flexibly incorporates multivariate, longitudinal nutrition survey data such as food frequency questionaires with multiple outcome types (e.g. ordinal, continuous, etc.). A multiplicative gamma process prior is placed on the factor loadings to adaptively estimate low-dimensional dietary patterns. Importantly, our model also incorporates covariates such as demographics to assess how dietary patterns differ across subpopulations. We evaluate the Frequentist operating characteristics of the method in a simulation study. Our motivating application is the Black Women's Health Study, where we construct dynamic measures of diet that will be used in downstream analyses to better understand cardiovascular disease risk among black women in the United States. 

Keywords

nutrition

Bayesian inference

latent factor model

health disparities

survey data 

View Abstract 3844

Co-Author(s)

Briana Stephenson, Harvard T.H. Chan School of Public Health
Xihong Lin, Harvard T.H. Chan School of Public Health

First Author

Daniel Schwartz, Harvard T.H. Chan School of Public Health

Presenting Author

Daniel Schwartz, Harvard T.H. Chan School of Public Health

Extensions of PROLONG: Penalized Regression On Longitudinal Multi-Omics Data with Network and Group

There is a growing interest in longitudinal omics data, but there are gaps in existing high-dimensional methodology. In particular, we are focused on modeling general continuous longitudinal outcomes with continuous longitudinal multi-omics predictors. Simple univariate longitudinal models do not leverage the correlation across predictors, thus losing power. Our method, PROLONG, leverages the first differences of the data to address the piecewise linear structure and the observed time dependence and applies penalties that induce sparsity while incorporating the dependence structure of the data. This presentation will review PROLONG and discuss recent extensions to multiple treatment arms, mixed effects, and general multi-omic data. 

Keywords

Omics

Longitudinal

High-dimensional

Biomarkers

TB

Metabolomics 

Abstracts


Co-Author(s)

Sumanta Basu, Cornell University
Myung Hee Lee, Weill Cornell Medicine
Martin Wells, Cornell University

First Author

Steven Broll, Cornell University

Presenting Author

Steven Broll, Cornell University

Imputing Censored Covariates in Longitudinal Models of Huntington Disease Progression

Huntington disease is a rare neurodegenerative disorder for which no effective therapies have yet been discovered. To help clinicians search for effective therapies that halt or slow the progression of the disease, analysts seek to model the progression of symptoms before and after diagnosis. However, many studies that track symptom progression end before all participants have been diagnosed. This presents a challenge: How do we model symptom progression given time of diagnosis, a right-censored covariate? Analysts frequently meet this challenge by imputing the time of diagnosis for undiagnosed participants. However, if the model used to impute the time to diagnosis is misspecified, it can introduce bias into the symptom progression model. This bias arises because the misspecified imputation model generates values that are prone to significant errors. To mitigate this estimation bias, we adopt a semiparametric technique to correct for imputation errors, enabling us to reliably estimate linear longitudinal models even in the presence of covariate censoring. Our novel approach is presented, and we assess its performance when applied to an observational study of Huntington disease. 

Keywords

censored covariate

imputation

longitudinal data

measurement error

semiparametric theory 

View Abstract 3179

Co-Author(s)

Tanya Garcia
Sarah Lotspeich, Wake Forest University

First Author

Kyle Grosser, University of North Carolina

Presenting Author

Kyle Grosser, University of North Carolina

Supervised Fusion Learning of Physical Activity Features: Longitudinal Functional Accelerometer Data

Accelerometry data collected by high-capacity sensors present a primary data type in smart mobile health. I holistically summarize an individual subject's activity profile using Occupation Time curves (OTCs). Being a functional predictor, OTCs describe the percentage of time spent at or above a continuum of activity count levels. The resulting functional curve is informative to capture time-course individual variability of physical activities both on the underlying functional variables of interest, as well as the specific health outcomes. I leverage the OTC curves to develop a longitudinal functional framework with repeated wearable data to understand the influence of serially measured functional accelerometer data on longitudinal health outcomes. I develop a new one-step method that can simultaneously conduct fusion via change-point detection and parameter estimation through a new L0 constraint formulation, invoking Quadratic Inference Functions (QIF), with an aim to detect physical activity intensity windows and assess their population-average effects on children health outcomes. 

Keywords

L0 regularization

changepoint detection

accelerometer

functional data analysis 

View Abstract 3554

Co-Author

Peter Song, University of Michigan

First Author

Margaret Banker, University of Michigan

Presenting Author

Margaret Banker, University of Michigan