In ziwang89/extendAUG: Package

extendAUG

The goal of extendAUG is to partial replicate and extend the work of 'An augmented estimation procedure for EHR-based association studies accounting for differential misclassification' (2020) "

Installation

You can install the development version of extendAUG from GitHub with:

# install.packages("devtools")
#devtools::install_github("ziwang89/extendAUG")

Example

There are four functions contained in the extendAUG R package, including aug_function, data_generator, one_time_simulation, and homo_AugEst. The main function to replicate and extend the simulation in the augmented estimation paper is homo_AugEst with adding additional covariate; the main function to generate the augmented estimators is aug_function. The data_generator can be used to generate the input data for aug_function in order to compute the augmented estimators, the variance of the estimators, and the confidence interval the estimators for 3-model settings. The usage details are introduced in the following sections. In these functions, if else and tryCatch() are used to insure that the function is able to handle the potential simulation errors in the inputs/arguments. Moreover, the simulation in parallel works for both Mac and Windows system.

Note*: this package passes all checks by R CMD check extendAUG_0.0.0.9000.tar.gz with no errors and no warnings.

Guided by JAMIA paper [@Tong2020], let's replicate Scenario 0 setting: Non-differential Misclassification of the simulation study in the original paper. Here, we focus on model comparison among Model(1), Model(2), and Model(4). Following the original assumptions, the authors set the full data set size is n = 5000, and the investigate validation sets of 4 sizes: m = 100, 200, 400, and 800 (i.e., the corresponding values of $\rho$ are 0.02, 0.04, 0.08, and 0.16). We choose the intercept $\beta_{0}$ to be -1.0, corresponding to a disease prevalence of 36.79%, the association parameters $\beta_{1}$ and $\beta_{2}$ were set to 0.5 and 1, respectively, corresponding to an odds ratio of 1.64 (a moderate effect size) and 2.7 (a relatively large effect size). The whole simulation code is wrapped into the extendAUG R package. The main aug_function function is designed for obtaining the proposed augmented estimator of EHR-based association studies accounting for non-differential misclassification. The main replication and extension of this simulation study can be done by calling function homo_AugEst. Then, we can directly get the simulation results for Scenario 0 setting in the augmented estimation paper.

library(extendAUG)

# Replication of Scenario 0: Not parallelized version
example = homo_AugEst(n = 5000, Nsim = 100, x_beta = 1, parallel_ifrun = FALSE)
# Replication of Scenario 0: parallelized version
example = homo_AugEst(n = 2000, Nsim = 100, x_beta = 1, parallel_ifrun = TRUE)

The output data can be used to create a comparison plot of simulation results of 3 models (1), (2), and (4) as Figure 5 in the JAMIA paper.

# for the investigate validation sets of m = 100
plot200.dat <- data.frame(example$sim_200)

p <- ggplot(plot200.dat, aes(x= ind, y= values, fill= ind)) +
  geom_boxplot(notch=TRUE,fill=c("#FFC0CB","#7F60E4","#5010F7")) +
  labs(title="",x="", y = "Estimated Log Odds Ratio") +
  geom_hline(yintercept = 0.5,linetype="dashed",size = 1.5) +  
  ## yintercept = 0.5 and 1 for beta1 corresponding to b1 and b2 , respectively
  theme_classic(base_size=40) +
  theme(axis.text.x = element_text(angle = 40, hjust = 1)) + 
  ylim(0,1) 

pdf("~/Dropbox/penn_master/BSTA670_ProgComput/FINAL_PROJECT/2000_400beta1", width= 12, height = 9)
p
dev.off()
print(p)

Ref: Tong, J., Huang, J., Chubak, J., Wang, X., Moore, J. H., Hubbard, R. A., & Chen, Y. (2020). An augmented estimation procedure for EHR-based association studies accounting for differential misclassification. Journal of the American Medical Informatics Association, 27(2), 244–253. https://doi.org/10.1093/jamia/ocz180

ziwang89/extendAUG documentation built on Dec. 23, 2021, 10:10 p.m.