refBasedCts: Reference based imputation of repeated measures continuous...

In jwb133/mlmi: Maximum Likelihood Multiple Imputation

Reference based imputation of repeated measures continuous data

Description

Performs multiple imputation of a repeatedly measured continuous endpoint in a randomised clinical trial using reference based imputation as proposed by \Sexpr[results=rd]{tools:::Rd_expr_doi("10.1080/10543406.2013.834911")}Carpenter et al (2013). This approach can be used for imputation of missing data in randomised clinical trials.

Usage

refBasedCts(
  obsData,
  outcomeVarStem,
  nVisits,
  trtVar,
  baselineVars = NULL,
  baselineVisitInt = TRUE,
  type = "MAR",
  M = 5
)

Arguments

obsData	The data frame to be imputed.
outcomeVarStem	String for stem of outcome variable name, e.g. y if y1, y2, y3 are the outcome columns
nVisits	The integer number of visits (not including baseline)
trtVar	The string variable name of the randomised treatment group variable. The reference arm is assumed to correspond to trtVar==0.
baselineVars	A string or vector of strings specfying the baseline variables. Often this will include the baseline measurement of the outcome
baselineVisitInt	TRUE/FALSE indicating whether to allow for interactions between each baseline variable and visit. Default is TRUE.
type	A string specifying imputation type to use. Valid options are "MAR", "J2R"
M	Number of imputations to generate.

Details

Unlike most implementations of reference based imputation, this implementation imputes conditional on the maximum likelihood estimates of the model parameters, rather than a posterior draw. If one is interested in frequentist valid inferences, this is ok provided the bootstrapping used, for example with using the bootImpute package.

Intermediate missing values are imputed assuming MAR, based on the mixed model fit to that patient's treatment arm. Monotone missing values are imputed using the specified imputation type.

Baseline covariates must be numeric variables. If you have factor variables you must code these into suitable dummy indicators and pass these to the function.

Value

A list of imputed datasets, or if M=1, just the imputed data frame.

References

Carpenter JR, Roger JH, Kenward MG. Analysis of Longitudinal Trials with Protocol Deviation: A Framework for Relevant, Accessible Assumptions, and Inference via Multiple Imputation. (2013) 23(6) 1352-1371

von Hippel PT & Bartlett JW (2019) Maximum likelihood multiple imputation: Faster imputations and consistent standard errors without posterior draws arXiv:1210.0870v10.

Examples

#take a look at ctsTrialWide data
head(ctsTrialWide)

#impute the missing outcome values twice assuming MAR
imps <- refBasedCts(ctsTrialWide, outcomeVarStem="y", nVisits=3, trtVar="trt",
                    baselineVars=c("v", "y0"), type="MAR", M=2)

#now impute using jump to reference method
imps <- refBasedCts(ctsTrialWide, outcomeVarStem="y", nVisits=3, trtVar="trt",
                    baselineVars=c("v", "y0"), type="J2R", M=2)

#for frequentist valid inferences we use bootstrapping from the bootImpute package
## Not run: 
  #bootstrap 10 times using 2 imputations per bootstrap. Note that to do this
  #we specify nImp=2 to bootImpute by M=1 to the refBasedCts function.
  #Also, 10 bootstraps is far too small to get reliable inferences. To do this
  #for real you would want to use a lot more (e.g. at least nBoot=1000).
  library(bootImpute)
  bootImps <- bootImpute(ctsTrialWide, refBasedCts, nBoot=10, nImp=2,
                         outcomeVarStem="y", nVisits=3, trtVar="trt",
                         baselineVars=c("v", "y0"), type="J2R", M=1)

  #write a small wrapper function to perform an ANCOVA at the final time point
  ancova <- function(inputData) {
    coef(lm(y3~v+y0+trt, data=inputData))
  }
  ests <- bootImputeAnalyse(bootImps, ancova)
  ests

## End(Not run)

jwb133/mlmi documentation built on June 4, 2023, 9:39 a.m.