Biomarker analysis of functional reserve in patients with heart failure with reduced ejection fraction (HFrEF), using data from the HF-ACTION randomized clinical trial.
This repository accompanies a completed master's project in the Department of Biostatistics & Bioinformatics at Duke University.
This project evaluates whether a panel of 32 baseline protein biomarkers is associated with functional reserve baseline cardiopulmonary exercise capacity (peak VO₂) in HFrEF, using a prespecified four-model robustness framework.
Scope: This repository covers the completed analysis of functional reserve. Two extensions were scoped but are outside the delivered work see Planned extensions.
- Clinical Trial Data Analysis
- Biomarker Discovery
- Multiple Linear Regression
- False Discovery Rate (FDR) Correction
- Reproducible Research in R
- Statistical Analysis Plan (SAP) Development
- Clinical Data Cleaning and Validation
- Publication-Style Statistical Reporting
flowchart TD
A[HF-ACTION Trial Data]
B[Data Cleaning & Validation]
C[Biomarker Screening]
D[Four-Model Regression Framework]
E[FDR Correction]
F[Robust Biomarker Identification]
G[Clinical Interpretation]
A --> B
B --> C
C --> D
D --> E
E --> F
F --> G
Sample. HFrEF participants from the HF-ACTION biomarker substudy. The analytic sample for regression models is n = 263 (from n = 300 with biomarker data; exclusions driven primarily by missing covariate data). The correlation structure is described on the full biomarker-complete cohort (n = 300).
Aim 1 : functional reserve. Each z-scored biomarker is related to peak VO₂ under a prespecified four-model framework crossing adjustment (unadjusted vs. adjusted for sex, BMI, BUN, and KCCQ symptom burden) with outcome scale (original vs. log-transformed). Multiplicity is controlled with the Benjamini–Hochberg procedure (FDR q < 0.10). A biomarker is considered robust when it shows (1) direction consistency across all four model specifications and (2) q < 0.10 in both adjusted models.
Headline result. 13 of 32 biomarkers met both robustness criteria; associations with peak VO₂ were predominantly inverse.
- 32 baseline biomarkers evaluated
- 263 participants included in adjusted analyses
- 13 biomarkers met predefined robustness criteria
- Associations with functional reserve were predominantly inverse
- Findings were consistent across all four model specifications
The functional reserve analysis was designed to support two further directions, which are not part of this repository's completed work:
- Functional recovery Predicting change in peak VO₂ at 3 months from baseline biomarkers using LASSO penalized regression.
- Functional resilience A residual-based analysis testing whether biomarkers add predictive value beyond clinical measures.
These are noted to document the project's intended scope, not as delivered results.
.
├── R/ # Analysis scripts and R Markdown notebooks (Aim 1)
├── data/ # Data access instructions only (no data committed)
├── output/
│ ├── figures/ # Generated figures
│ └── tables/ # Generated publication tables
└── docs/ # Statistical analysis plan, report
The HF-ACTION trial data are not redistributed in this repository. They are available to qualified investigators through the NHLBI Biologic Specimen and Data Repository Information Coordinating Center (BioLINCC) under a Data Use Agreement.
- Request access: https://biolincc.nhlbi.nih.gov/studies/hf_action/
To reproduce the analysis, obtain the data through BioLINCC and place the files in a
local data/ directory matching the paths referenced in R/01_data_prep.R. The
.gitignore is configured to prevent data files from being committed.
Analyses were conducted in R using R Markdown.
Core packages:
| Package | Use |
|---|---|
gtsummary, gt |
Descriptive and publication tables |
readxl |
Reading source spreadsheets |
tidyverse |
Data manipulation and plotting |
# Install dependencies
install.packages(c("tidyverse", "gtsummary", "gt", "readxl"))Run the scripts in numerical order. Set the working directory to the repository root before sourcing.
If you reference this work, please cite it using the metadata in
CITATION.cff.
Patrick Bautista Master's candidate, Biostatistics & Bioinformatics, Duke University
Advisor: Marissa Ashner, PhD Committee: Sarah Peskoe, PhD · Leanna Ross, PhD
Code in this repository is released under the MIT License. The license applies to the analysis code only and does not extend to the HF-ACTION data, which remain governed by the NHLBI BioLINCC Data Use Agreement.
This research uses data from the HF-ACTION trial, supported by the National Heart, Lung, and Blood Institute (NHLBI) and obtained through NHLBI BioLINCC. The views expressed here are the author's own and do not reflect those of the NHLBI.