Average Treatment effect for Restricted Mean for censored competing risks data using IPCW
Source:R/restricted.mean.R
resmeanATE.RdUnder the standard causal assumptions we can estimate the average treatment effect E(Y(1) - Y(0)). We need Consistency, ignorability ( Y(1), Y(0) indep A given X), and positivity.
Usage
resmeanATE(formula, data, model = "exp", outcome = c("rmst", "rmtl"), ...)Details
The first covariate in the specification of the competing risks regression model must be the treatment effect that is a factor. If the factor has more than two levels then it uses the mlogit for propensity score modelling. We consider the outcome mint(T;tau) or I(epsion==cause1)(t- min(T;t)) that gives years lost due to cause "cause" depending on the number of causes. The default model is the exp(X^ beta) and otherwise a linear model is used.
Estimates the ATE using the the standard binary double robust estimating equations that are IPCW censoring adjusted.
References
Scheike, T. and Holst, K. K. Restricted mean time lost for survival and competing risks data using mets in R, WIP
Examples
library(mets); data(bmt); bmt$event <- bmt$cause!=0; dfactor(bmt) <- tcell~tcell
out <- resmeanATE(Event(time,event)~tcell+platelet,data=bmt,time=40,treat.model=tcell~platelet)
summary(out)
#> n events
#> 408 241
#>
#> 408 clusters
#> coeffients:
#> Estimate Std.Err 2.5% 97.5% P-value
#> (Intercept) 2.852563 0.062496 2.730074 2.975052 0.0000
#> tcell1 0.021286 0.122983 -0.219757 0.262329 0.8626
#> platelet 0.303306 0.090772 0.125396 0.481215 0.0008
#>
#> exp(coeffients):
#> Estimate 2.5% 97.5%
#> (Intercept) 17.33214 15.33402 19.591
#> tcell1 1.02151 0.80271 1.300
#> platelet 1.35433 1.13360 1.618
#>
#> Average Treatment effects (G-formula) :
#> Estimate Std.Err 2.5% 97.5% P-value
#> treat0 19.25882 0.95918 17.37887 21.13877 0.0000
#> treat1 19.67316 2.22868 15.30502 24.04129 0.0000
#> treat:1-0 0.41434 2.41151 -4.31213 5.14081 0.8636
#>
#> Average Treatment effects (double robust) :
#> Estimate Std.Err 2.5% 97.5% P-value
#> treat0 19.27793 0.95799 17.40030 21.15556 0.0000
#> treat1 20.34004 2.54146 15.35887 25.32122 0.0000
#> treat:1-0 1.06211 2.71020 -4.24979 6.37402 0.6951
#>
#>
out1 <- resmeanATE(Event(time,cause)~tcell+platelet,data=bmt,cause=1,time=40,
treat.model=tcell~platelet)
summary(out1)
#> n events
#> 408 157
#>
#> 408 clusters
#> coeffients:
#> Estimate Std.Err 2.5% 97.5% P-value
#> (Intercept) 2.80626 0.06962 2.66981 2.94271 0.0000
#> tcell1 -0.37413 0.24769 -0.85960 0.11133 0.1309
#> platelet -0.49164 0.16493 -0.81490 -0.16837 0.0029
#>
#> exp(coeffients):
#> Estimate 2.5% 97.5%
#> (Intercept) 16.54790 14.43717 18.9672
#> tcell1 0.68788 0.42333 1.1178
#> platelet 0.61162 0.44268 0.8450
#>
#> Average Treatment effects (G-formula) :
#> Estimate Std.Err 2.5% 97.5% P-value
#> treat0 14.53165 0.95705 12.65587 16.40742 0.0000
#> treat1 9.99609 2.37815 5.33499 14.65718 0.0000
#> treat:1-0 -4.53556 2.57515 -9.58276 0.51164 0.0782
#>
#> Average Treatment effects (double robust) :
#> Estimate Std.Err 2.5% 97.5% P-value
#> treat0 14.51355 0.95800 12.63590 16.39120 0.0000
#> treat1 9.36465 2.41708 4.62727 14.10203 0.0001
#> treat:1-0 -5.14890 2.59798 -10.24084 -0.05696 0.0475
#>
#>
ratioATE(out,out1,h=function(x) log(x))
#> $ratioG
#> Estimate Std.Err 2.5% 97.5% P-value
#> [treat0] -0.2816 0.1075 -0.4924 -0.07093 9.150e-33
#> [treat1] -0.6771 0.3183 -1.3008 -0.05328 1.369e-07
#> [treat0].1 -0.3954 0.3363 -1.0545 0.26370 3.333e-05
#> [treat1].1 0.3954 0.3363 -0.2637 1.05454 7.221e-02
#>
#> Null Hypothesis:
#> [treat0] = 1
#> [treat1] = 1
#> [treat0] = 1
#> [treat1] = 1
#>
#> chisq = 170.3366, df = 4, p-value < 2.2e-16
#>
#> $ratioDR
#> Estimate Std.Err 2.5% 97.5% P-value
#> [treat0] -0.2839 0.1075 -0.4947 -0.07310 7.454e-33
#> [treat1] -0.7756 0.3460 -1.4538 -0.09753 2.865e-07
#> [treat0].1 -0.4918 0.3619 -1.2011 0.21754 3.756e-05
#> [treat1].1 0.4918 0.3619 -0.2175 1.20109 1.602e-01
#>
#> Null Hypothesis:
#> [treat0] = 1
#> [treat1] = 1
#> [treat0] = 1
#> [treat1] = 1
#>
#> chisq = 168.3765, df = 4, p-value < 2.2e-16
#>