parallel-tnd

Replication files for:

Boyer, C., Li, K.Q., Shi, X., & Tchetgen Tchetgen, T. J. (2025). “Identification and estimation of vaccine effectiveness in the test-negative design under equi-confounding”. arXiv. https://doi.org/10.48550/arXiv.2504.20360

Abstract

The test-negative design (TND) is widely used to evaluate vaccine effectiveness in real-world settings. In a TND study, individuals with similar symptoms who seek care are tested, and effectiveness is estimated by comparing vaccination histories of test-positive cases and test-negative controls. The TND is often justified on the grounds that it reduces confounding due to unmeasured health-seeking behavior, although this has not been formally described using potential outcomes. At the same time, concerns persist that conditioning on test receipt can introduce selection bias. We provide a formal justification of the TND under an assumption of odds ratio equi-confounding, where unmeasured confounders affect test-positive and test-negative individuals equivalently on the odds ratio scale. Health-seeking behavior is one plausible example. We also show that these results hold under the outcome-dependent sampling used in TNDs. We discuss the design implications of the equi-confounding assumption and provide alternative estimators for the marginal risk ratio among the vaccinated under equi-confounding, including outcome modeling and inverse probability weighting estimators as well as a semiparametric estimator that is doubly robust. When equi-confounding does not hold, we suggest a straightforward sensitivity analysis that parameterizes the magnitude of the deviation on the odds ratio scale. A simulation study evaluates the empirical performance of our proposed estimators under a wide range of scenarios. Finally, we discuss broader uses of test-negative outcomes to de-bias cohort studies in which testing is triggered by symptoms.

Requirements

This code requires R version 4.0 or higher. The following R packages are required:

Core simulation and analysis:

• data.table - for data manipulation and simulation framework
• progressr - for progress tracking during simulations
• readr - for reading/writing simulation results
• lmtest - for linear model testing (coefci function)
• sandwich - for robust standard errors (vcovHC function)
• survival - for survival analysis functions
• numDeriv - for numerical derivatives in estimating equations

Table and figure generation:

• ggplot2 - for plotting simulation results
• patchwork - for combining plots
• tidyverse - collection of tidy data packages (includes dplyr, tidyr, stringr)
• dplyr - for data manipulation in table generation
• tidyr - for data tidying
• stringr - for string manipulation
• kableExtra - for LaTeX table formatting

Installation

1. Clone this repository:

git clone https://github.com/boyercb/parallel-tnd.git
cd parallel-tnd

2. Install required R packages:

install.packages(c("data.table", "ggplot2", "progressr", "survival", 
                   "lmtest", "sandwich", "readr", "numDeriv",
                   "dplyr", "tidyr", "stringr", "kableExtra", 
                   "patchwork", "tidyverse"))

3. Create the data directory (if it doesn't exist):

dir.create("data", showWarnings = FALSE)

Repository Structure

• code/ - Contains all R scripts for the simulation study
  • run.R - Main script that runs all simulation scenarios
  • datagen.R - Data generation functions
  • estimators.R - Implementation of all estimators
  • sim.R - Simulation wrapper functions
  • table.R - Code to generate results tables
  • plot.R - Code to generate figures
• data/ - Data files (in this project only saved simulation results)
• manuscript/ - LaTeX source files for the manuscript
• results/ - Output files (tables and figures)

How to run

Running the full simulation study

To reproduce all simulation results from the paper, run the following command in R from the project root directory:

source("code/run.R")

Note: The simulation study runs:

• Scenarios 1-7: 1,000 replications each with sample size N=15,000
• Scenario 8: 2,000 replications each with sample size N=15,000 across 4 sub-scenarios

This will take several hours to complete (estimated 6-8 hours depending on your system).

Generating tables and figures

After running the simulations, you can generate the tables and figures using:

# Load simulation results
sims <- readr::read_rds("data/sims.rds")

# Generate tables (creates LaTeX files in results/ directory)
source("code/table.R")

# Generate figures (creates PDF files in results/ directory)
source("code/plot.R")

Individual components

You can also run individual parts of the analysis:

# Load required functions
source("code/datagen.R")
source("code/estimators.R") 
source("code/sim.R")

# Run a single scenario (much faster for testing)
# See run.R for all scenario definitions

Simulation Scenarios

The simulation study evaluates 8 different scenarios:

1. No unmeasured confounding - Baseline scenario where TND assumptions hold
2. Equi-confounding - Main scenario where proposed methods should work
3. Direct effect of vaccination on test-negative infection - Vaccination affects test-negative outcomes (exclusion restriction violated)
4. Equi-confounding violated - Confounding differs between test-positive and test-negative
5. Equi-selection violated - Selection bias differs between test-positive and test-negative
6. Equal effect of vaccination on testing - Equal effects of vaccination on testing behavior
7. Unequal effect of vaccination on testing - Unequal effects of vaccination on testing behavior
8. Effect heterogeneity - Scenarios with treatment effect modification
   • 8a: Both models correctly specified
   • 8b: Propensity score model misspecified
   • 8c: Outcome model misspecified
   • 8d: Both models misspecified

Each scenario compares multiple estimators:

• TND estimators:
  • Logistic regression (logit_reg) - traditional TND approach
  • Risk ratio among vaccinated - outcome modeling (rrv_om)
  • Risk ratio among vaccinated - inverse probability weighting (rrv_ipw)
  • Risk ratio among vaccinated - doubly-robust (rrv_dr)
• Cohort estimators:
  • Cohort with unmeasured confounders (cohort_reg_U) - oracle estimator
  • Cohort without unmeasured confounders (cohort_reg_noU) - naive estimator
  • Difference-in-differences (did_reg) - equivalent to TND under equi-confounding

Output

The simulation generates:

• Performance metrics: Bias, coverage probability, and confidence interval length for each estimator
• LaTeX tables: Saved to results/ directory:
  • sims.tex - Main simulation results
  • sims_dr*.tex - Additional robustness results
• Figures: PDF files showing estimator performance across scenarios (sims1.pdf, sims2.pdf)
• Raw data: Complete simulation results saved as data/sims.rds (R data format)
• Sample sizes: Actual TND sample sizes achieved for each simulation

Citation

If you use this code, please cite:

Boyer, C., Li, K.Q., Shi, X., & Tchetgen Tchetgen, T. J. (2025). "Identification and estimation of vaccine effectiveness in the test-negative design under equi-confounding". arXiv. https://doi.org/10.48550/arXiv.2504.20360