Inverse Probability Weighted (IPW) Estimation of Average Treatment Effect (ATE) with Misclassified Binary Outcome

Description

This package is an implementation of the correction methods proposed by Shu and Yi (2017) for the inverse probability weighted (IPW) estimation of average treatment effect (ATE) with misclassified binary outcomes. Logistic regression model is assumed for treatment model for all implemented correction methods, and is assumed for the outcome model for the implemented doubly robust correction method. Misclassification probability given a true value of the outcome is assumed to be the same for all individuals.

Details

The ipwErrorY package implements correction methods developed by Shu and Yi (2017) to adjust for misclassification in binary outcomes in the inverse probability weighted estimation of average treatment effect. The function KnownError implements the correction method with known outcome misclassification probabilities. The function EstValidation implements the optimal linear combination correction method when validation data are available. The function Est2Replicates implements the correction method when two independent replicates of the outcome are available. The function KnownErrorDR implements the doubly robust correction method with known outcome misclassification probabilities.

Author(s)

Di Shu and Grace Y. Yi

Maintainer: Di Shu <shudi1991@gmail.com>

References

Shu D, Yi GY. (2017). Causal inference with measurement error in outcomes: bias analysis and estimation methods. Statistical Methods in Medical Research <doi:10.1177/0962280217743777>

Estimation of ATE with Two Replicates

Description

Estimation of average treatment effect when misclassification probabilities are unknown but two independent replicates of the outcome are available

Usage

Est2Replicates(data, indA, indYerror, indX,
  constraint = c("sensitivity equals specificity", "known sensitivity",
  "known specificity", "known prevalence"), sensitivity = NULL,
  specificity = NULL, prevalence = NULL, confidence = 0.95)

Arguments

Value

A list of the estimate of average treatment effect, sandwich-variance-based standard error, confidence interval, imposed constraint, and the information on sensitivity and specificity

Examples

#create a dataset with sensitivity=0.95 and specificity=0.85
set.seed(100)
X1=rnorm(2000)   
A=rbinom(2000,1,1/(1+exp(-0.2-X1)))
Y=rbinom(2000,1,1/(1+exp(-0.2-A-X1)))
y1=which(Y==1)
y0=which(Y==0) 
Yast1=Y
Yast1[y1]=rbinom(length(y1),1,0.95)
Yast1[y0]=rbinom(length(y0),1,0.15)
Yast2=Y
Yast2[y1]=rbinom(length(y1),1,0.95)  
Yast2[y0]=rbinom(length(y0),1,0.15)
da=data.frame(A=A,X1=X1,Yast1=Yast1,Yast2=Yast2)
head(da)
#apply the correction method assuming specificity=0.85
Est2Replicates(da,"A",c("Yast1","Yast2"),"X1","known specificity",NULL,0.85,NULL,0.95)

Estimation of ATE with Validation Data

Description

Estimation of average treatment effect using the optimal linear combination method when misclassification probabilities are unknown but validation data are available

Usage

EstValidation(maindata, validationdata, indA, indYerror, indX, indY,
  confidence = 0.95)

Arguments

Value

A list of the estimate of average treatment effect, sandwich-variance-based standard error, confidence interval, and the estimated sensitivity and specificity

Examples

#create main data and validation data with sensitivity=0.95 and specificity=0.85
set.seed(100)
X1=rnorm(1200)   
A=rbinom(1200,1,1/(1+exp(-0.2-X1)))
Y=rbinom(1200,1,1/(1+exp(-0.2-A-X1)))
y1=which(Y==1)
y0=which(Y==0) 
Yast=Y
Yast[y1]=rbinom(length(y1),1,0.95)
Yast[y0]=rbinom(length(y0),1,0.15)
mainda=data.frame(A=A,X1=X1,Yast=Yast)
X1=rnorm(800)   
A=rbinom(800,1,1/(1+exp(-0.2-X1)))
Y=rbinom(800,1,1/(1+exp(-0.2-A-X1)))
y1=which(Y==1)
y0=which(Y==0) 
Yast=Y
Yast[y1]=rbinom(length(y1),1,0.95)
Yast[y0]=rbinom(length(y0),1,0.15)
validationda=data.frame(A=A,X1=X1,Y=Y,Yast=Yast)
head(mainda)
head(validationda)
#apply the optimal linear combination correction method
EstValidation(mainda,validationda,"A","Yast","X1","Y",0.95)

Estimation of ATE with Known Error

Description

Estimation of average treatment effect with known outcome misclassification probabilities, i.e., known sensitivity and specificity

Usage

KnownError(data, indA, indYerror, indX, sensitivity, specificity,
  confidence = 0.95)

Arguments

Value

A list of the estimate of average treatment effect, sandwich-variance-based standard error and confidence interval

Examples

#create a dataset with sensitivity=0.95 and specificity=0.85
set.seed(100)
X1=rnorm(2000) 
A=rbinom(2000,1,1/(1+exp(-0.2-X1)))
Y=rbinom(2000,1,1/(1+exp(-0.2-A-X1)))
y1=which(Y==1)
y0=which(Y==0) 
Yast=Y
Yast[y1]=rbinom(length(y1),1,0.95)
Yast[y0]=rbinom(length(y0),1,0.15)
da=data.frame(X1=X1,A=A,Yast=Yast)
head(da)
#apply the correction method with sensitivity=0.95 and specificity=0.85
KnownError(da,"A","Yast","X1",0.95,0.85,0.95)

Doubly Robust Estimation of ATE with Known Error

Description

Doubly robust estimation of average treatment effect with known outcome misclassification probabilities, i.e., known sensitivity and specificity

Usage

KnownErrorDR(data, indA, indYerror, indXtrt, indXout, sensitivity, specificity,
  sharePara = FALSE, confidence = 0.95)

Arguments

Value

A list of the estimate of average treatment effect, sandwich-variance-based standard error and confidence interval

Examples

#create a dataset with sensitivity=0.95 and specificity=0.85
set.seed(100)
X=rnorm(2000)   
xx=X^2
A=rbinom(2000,1,1/(1+exp(-0.1-X-0.2*xx)))
Y=rbinom(2000,1,1/(1+exp(1-A-0.5*X-xx)))
y1=which(Y==1)
y0=which(Y==0) 
Y[y1]=rbinom(length(y1),1,0.95)
Y[y0]=rbinom(length(y0),1,0.15)
Yast=Y
da=data.frame(A=A,X=X,xx=xx,Yast=Yast)
head(da)
#apply the doubly robust correction method with sensitivity=0.95 and specificity=0.85
KnownErrorDR(da,"A","Yast",c("X","xx"),c("X","xx"),0.95,0.85,FALSE,0.95)