Code of study: "The impact of differential exposure measurement error on the PM2.5 -- preterm birth association: A simulation study"

Code by: Vijay Kumar Reviewer: Alan Domínguez Date updated: March 17, 2026

================================================================================ REQUIREMENTS

R version: 4.2.2 (or later)

Required packages: haven, dplyr, readr, MASS, lubridate, lme4, ggplot2, cowplot, xtable

Install all at once: install.packages(c("haven","dplyr","readr","MASS","lubridate","lme4", "ggplot2","cowplot","xtable"))

================================================================================ FOLDER STRUCTURE

The code is structured as a main folder (project folder) with a subfolder exp, which contains three sub-subfolders: exp_001, exp_002, and exp_003.

Each exp_00X folder contains:

• Sims_all.R runs all 10 scenarios in one session
• sims1.R - sims10.R individual scenario scripts (for parallel execution)
• Results/ simulation outputs and visualization scripts saved here

Data files required in the project root:

• zipall_1719.sas7bdat (preterm birth data, provided)
• pm25_js_zip_2000_2016_ny_daily.csv (daily PM2.5 data, provided separately)

Generated by Step 1 (do not use pre-existing copies -- always regenerate):

• truedta_null_Final.csv
• truedta_ass_Final.csv

================================================================================ STEP 1 -- DATA CREATION

Navigate to the project root folder, then run:

cd ~/path/to/project folder/ Rscript CreateDataFinal.R

This generates: truedta_null_Final.csv (null association data, true RR = 1.00) truedta_ass_Final.csv (harmful association data, true RR = 1.115)

Data are created using ZIP code-level PTB counts (2017-2019) and daily 1 km PM2.5 predictions for NYC (2006-2016).

================================================================================ STEP 2 -- EXPERIMENTS

Three experiments are provided, each repeating the same 10 simulation scenarios with a different error mean structure:

exp_001 (E1): Random error only -- v ~ N(0, sigma^2) exp_002 (E2): Random + positive systematic -- v ~ N(+0.25, sigma^2) exp_003 (E3): Random + negative systematic -- v ~ N(-0.25, sigma^2)

Each exp_00X folder contains 10 scenario scripts (sims1.R - sims10.R) and and run_all_sims.sh:

Scenario 1: Non-differential error, Null association Scenario 2: Non-differential error, Harmful association Scenario 3: Differential -- less error in at-risk population, Null Scenario 4: Differential -- more error in at-risk population, Null Scenario 5: Differential -- less error in at-risk + 50% non-risk, Null Scenario 6: Differential -- more error in at-risk + 50% non-risk, Null Scenario 7: Differential -- less error in at-risk population, Harmful Scenario 8: Differential -- more error in at-risk population, Harmful Scenario 9: Differential -- less error in at-risk + 50% non-risk, Harmful Scenario 10: Differential -- more error in at-risk + 50% non-risk, Harmful run_all_sims.sh: opens all screens and runs each script in one line of code "./run_all_sims.sh"

================================================================================ STEP 3 -- RUNNING SIMULATIONS

NOTE: Scripts must be run from inside the exp_00X folder (not the project root).

---------- OPTION A: Run all scenarios together (one screen session) ----------

Estimated runtime: 10-12 hours on Mac M2 with 16 GB RAM. Data are loaded once and shared across all 10 scenarios.

screen -S sims_all cd ~/path/to/project folder/exp/exp_001 Rscript Sims_all.R

Detach screen (script keeps running in background): Press Ctrl+A, then D

Reattach later: screen -r sims_all

List all active screen sessions: screen -ls

Repeat for exp_002 and exp_003 (in separate screen sessions if desired).

---------- OPTION B: Run scenarios individually in parallel (~2.5 hrs total) --

Each sims#.R script is self-contained and can be run in its own screen session. Run all 10 in parallel for fastest completion.

Example for sims1.R: screen -S sims1 cd ~/path/to/project folder/exp/exp_001 Rscript sims1.R

---------- OPTION C (more direct): Run scenarios individually in parallel using .sh file (~2.5 hrs total) --

cd ~/path/to/project folder/ chmod +x run_all_sims.sh # only first time ./run_all_sims.sh # creates 10 screens named simsX and executes each simsX.R in that screen screen -ls # to check if all screens created and are running

Repeat for sims2.R through sims10.R using screen names sims2, sims3, etc.

---------- RECOMMENDED EXECUTION METHODS (in order of preference) ------------

1. Screen sessions through RStudio terminal (recommended)
2. HPC cluster (optional, for large-scale runs)
3. Direct R console (slowest -- not recommended for full runs)

================================================================================ STEP 4 -- VISUALIZING RESULTS (Figures and Tables)

Two scripts produce all figures and tables. Both must be run after Step 3 completes for a given experiment.

---------- Script 1: VisualizeResults.R (Figures 2, 3 and Tables A3-A5) ------

Navigate to the Results folder of each experiment and run:

cd ~/path/to/project folder/exp/exp_001/Results Rscript VisualizeResults.R

Produces:

• null_results.pdf Figure 2 (null panel)
• harm_results.pdf Figure 2 (harmful panel)
• null_bias_perc_heatmap.pdf Figure 3 (null bias heatmap)
• ass_bias_perc_heatmap.pdf Figure 3 (association bias heatmap)
• results.csv Pooled RR estimates (Tables A3-A5)
• bias_results.csv Bias and coverage by scenario
• coverage_results.csv CI coverage summary

Repeat for exp_002/Results and exp_003/Results.

---------- Script 2: CI_check.Rmd (Tables A1 and A6) -------------------------

This notebook must be run from the project root (not a Results subfolder). It processes all three experiments in one knit and saves one output CSV per experiment.

cd ~/path/to/project folder/ Rscript -e "rmarkdown::render('CI_check.Rmd')"

Or open CI_check.Rmd in RStudio and click "Knit".

Produces (one per experiment):

• exp/exp_001/Results/Var_Corr.csv Table A1 (E1 variance and correlation)
• exp/exp_002/Results/Var_Corr.csv Table A2 (E2 variance and correlation)
• exp/exp_003/Results/Var_Corr.csv Table A3 (E3 variance and correlation)
• exp/exp_001/Results/bias_results.csv Table A4 (E1 Bias 95% intervals)
• exp/exp_002/Results/bias_results.csv Table A5 (E2 Bias 95% intervals)
• exp/exp_003/Results/bias_results.csv Table A6 (E3 Bias 95% intervals)
• CI_check.html Rendered notebook with all three tables

NOTE: CI_check.Rmd must be placed in the project root folder alongside CreateDataFinal.R, not inside any exp_00X subfolder.

---------- Overall structure ----------

Note: Overall structure:

project folder/ ├── CreateDataFinal.R ← create null and association data ├── CI_check.Rmd ← run at the end ├── code_review.R ← filled by code reviewer ├── zipall_1719.sas7bdat ← PTB data (provided) ├── pm25_js_zip_2000_2016_ny_daily.csv ← PM2.5 data (provided separately, 117 MB) ├── truedta_null_Final.csv ← generated by CreateDataFinal.R ├── truedta_ass_Final.csv ← generated by CreateDataFinal.R ├── exp/ │ ├── meta │ ├── exp_001/ │ │ ├── readme │ │ ├── Sims_all.R │ │ ├── sims1.R – sims10.R │ │ └── Results/ │ │ ├── VisualizeResults.R │ │ └── (CSV outputs saved here) │ ├── exp_002/ │ │ └── (same structure) │ └── exp_003/ │ └── (same structure)