Longitudinal Data Analysis (LDA) 2025

This repository contains coursework, homework assignments, and analytic projects for the Longitudinal Data Analysis course at UHasselt (2025). The project focuses on applying advanced statistical methods to longitudinal clinical data, with a primary case study on Alzheimer’s disease progression.

Project Overview

The main objective is to model the evolution of psychiatric symptoms (BPRS) and cognitive-functional status (CDR-SB) over time, accounting for within-subject correlation and missing data due to dropout.

Key Methodologies

• Exploratory Data Analysis: Visualization of individual trajectories (spaghetti plots), mean structures, and variance functions (variograms).
• Linear Mixed-Effects Models (LMM): Analyzing continuous longitudinal outcomes (BPRS) with random intercepts and slopes.
• Generalized Linear Mixed Models (GLMM): Modeling binary outcomes (severe cognitive impairment, CDR-SB > 10) to estimate subject-specific risks.
• Generalized Estimating Equations (GEE): Estimating population-averaged effects, including Weighted GEE (WGEE) to correct for bias under Missing At Random (MAR) assumptions.
• Missing Data Analysis:
  • Investigation of missingness patterns (Monotone vs. Intermittent).
  • Sensitivity analysis using Pattern-Mixture Models (PMM) and $\delta$-adjustment to assess robustness against Missing Not At Random (MNAR) deviations.

Key Project Deliverables

• lda_hm3_final_2025.pdf - Final Report: Longitudinal Analysis of Alzheimer’s Disease Progression. This comprehensive report details the statistical analysis of psychiatric symptoms and cognitive decline, including methodology, results, and discussion.
  • Key Analysis: Includes Linear Mixed Models for BPRS, Logistic GLMM for severe impairment, and Weighted GEE for population-averaged effects, with sensitivity analysis for non-random dropout.
• lda_hm3_final_2025.qmd - Quarto Source Code for Analysis. The complete source file containing all R code for data cleaning, visualization, model fitting, and report generation.
  • Relevance: Useful for reproducing the analysis, inspecting model specifications (lme4, geepack), and understanding the implementation of pattern-mixture models for sensitivity analysis.

Repository Structure

• data/: Datasets used for analysis (e.g., alzheimer25.sas7bdat).
• hm3/: Final assignment resources, including the main Quarto report (lda_hm3_final_2025.qmd), bibliography, and generated PDFs.
• R/: R scripts and RMarkdown/Quarto source files for Homework 1, 2, and other exploratory scripts (critic_hm2.rmd, diff_gee_random.rmd).
• doc/: Course documentation and supplementary materials.
• presentation_mm/: Presentation slides and source files.
• results/: Output files, such as CSVs of variance components or model summaries.

Getting Started

Prerequisites

• R
• RStudio
• Quarto (for rendering .qmd files)

Key Packages

The analysis relies on the following R packages: * lme4 (Mixed models) * geepack (GEE models) * nlme (Linear mixed models) * dplyr, data.table (Data manipulation) * ggplot2 (Visualization)

Rendering Reports

To generate the final report for Homework 3: 1. Open hm3/lda_hm3_final_2025.qmd. 2. Render using Quarto to produce the PDF or HTML output.