ElasticIntegrative

Overview

The goal of ElasticIntegrative is to implement integrative analyses for the heterogenous treatment effect combining a randomized trial and a real-world evidence study from Yang et al., (2022).

Two datasets

• The randomized trial contains observations on (A,X,Y), where the treatment assignment A is randomized.

• The real-world evidence study contains observations on (A,X,Y), where the treatment assignment A may be confounded.

Installation with devtools:

# it takes around 1 minute for the downloads, including the vignettes
devtools::install_github("Gaochenyin/ElasticIntegrative", 
                         build_vignettes = TRUE) 
library(ElasticIntegrative)

Usage

This is an example for illustration when $Y$ is a continuous outcome.

Value

Argument	Description
est	the HTE estimators including
i)	covj.t the covariate adjustment estimator;
ii)	ee.rt(ml) the RCT estimating equation estimator with sieve models;
iii)	opt.ee(ml) the RT&RW estimating equation estimator with sieve models;
iv)	elasthe elastic integrative estimator.
ve	the bootstrap variance estimates for est.
CI	95% percentile bootstrap confidence intervals. Note covj.t, ee.rt(ml), and opt.ee(ml) use the Wald CI, and elas uses the adaptive confidence with threshold sqrt(log(2000)).

#> est
#>     covj.t.1     covj.t.2     covj.t.3  ee.rt(ml).1  ee.rt(ml).2  ee.rt(ml).3 
#>   0.02851533   1.00451768   0.94310969   0.05562283   1.00080762   0.97255215 
#> opt.ee(ml).1 opt.ee(ml).2 opt.ee(ml).3       elas.1       elas.2       elas.3 
#>   0.07641376   0.94788773   1.05146075   0.07641376   0.94788773   1.05146075
#> ve
#>     covj.t.1     covj.t.2     covj.t.3  ee.rt(ml).1  ee.rt(ml).2  ee.rt(ml).3 
#>  0.030100673  0.017169858  0.020132153  0.030455816  0.017169809  0.020936258 
#> opt.ee(ml).1 opt.ee(ml).2 opt.ee(ml).3       elas.1       elas.2       elas.3 
#>  0.006598755  0.004010907  0.003080497  0.005862381  0.004043671  0.003188111
#> CI
#>                    lower     upper
#> covj.t.1     -0.31152951 0.3685602
#> covj.t.2      0.74769614 1.2613392
#> covj.t.3      0.66501468 1.2212047
#> ee.rt(ml).1  -0.28642214 0.3976678
#> ee.rt(ml).2   0.74398645 1.2576288
#> ee.rt(ml).3   0.68895777 1.2561465
#> opt.ee(ml).1 -0.08279945 0.2356270
#> opt.ee(ml).2  0.82375983 1.0720156
#> opt.ee(ml).3  0.94267838 1.1602431
#> elas.1       -0.13145638 0.2803858
#> elas.2        0.86455094 1.1823684
#> elas.3        0.84800403 1.1233989

Extension

Additional functionality is incorporated to handle binary outcomes by specifying family.Y = binomial(). Note that $E(Y(1)-Y(0))=(\exp(Z^{\intercal}\psi_0)-1)/(\exp(Z^{\intercal}\psi_0)+1)$ for evaluating treatment effect in terms of binary outcomes may not be optimal.

result.elastic <- elasticHTE(mainName = c('X1', 'X2'),
                             contName = c('X1', 'X2'),
                             propenName = c('X1', 'X2'),
                             dat.t = dat.t,
                             dat.os = dat.os,
                             family.Y = binomial(),
                             thres.psi = sqrt(log(m)), # threshold for ACI psi
                             fixed = FALSE,
                             cvControl = list(V = 1L),
                             nboot = 50)

More examples

browseVignettes("ElasticIntegrative")

• Simulation: adaptive selection
• Simulation: comparing AIPW and SES
• Simulation: fixed threshold