kevjosey / pm-risk

Risk of Mortality Associated with Fine Particulate Matter Exposure

Mortality Benefits Associated with Decreasing PM_2.5

Assessing the EPA's recent recommendation to reduce PM_2.5 from 12 μg/m³ to 10, 9, and 8 μg/m³.

Contents

Analysis

The scripts described below are presented in the order with which they should be run.

1. Analysis/data_process.R: Creates stratified data sets from National Causal data.
2. Analysis/fit_weights.R: Applies functions from Functions/calibrate.R to fit nuisance inverse-probability/balancing weight models.
3. Analysis/fit_models.R: Applies functions from Functions/gam_models.R to fit nuisance outcome models.
4. Analysis/fit_erf.R: Uses code from Functions/erf_models.R and the output from (2) and (3) to fit the exposure response functions.
5. Analysis/fit_boot.R: Script for generating bootstrap samples used to estimate standard errors of the ERCs from (2)-(4).

Functions

• Functions/calibrate.R: Generic calibration function for fitting covariate balancing weights. Usable for binary, categorical, or continuous exposures. All that is needed is the correct target and the associated constraint. Implemented in Analysis/fit_weights.R.
• Functions/gam_models.R: Wrapper functions for fitting spline outcome models (formerly generalized additive outcome models) assuming a quasi-Poisson likelihood. Functions output components used to construct the doubly-robust pseudo-outcome applied in Functions/erf_models.R. Implemented in Analysis/fit_models.R.
• Functions/erf_models.R: Nonparametric doubly-robust estimator of the exposure response curve using the pseudo-outcomes fit with components found in Functions/calibrate.R and Functions/gam_models.R. Implemented in Analysis/fit_erf.R.

Results

Contains scripts for generating plots and tables from data generated in (1) and the models fit in (4)-(5).

Sensitivity

Contains code for fitting G-Computation and GPS as a regressor sensitivity analyses. Also included is code to visualize covariate balance for various weighting methods.