MImpute {doMIsaul}

R Documentation

Wrapper functions for multivariate imputation with survival data or left-censored data

Description

Performs imputation of the missing data using MICE and returns a list in the correct format for the unsupMI() and seMIsupcox()functions. MImpute() performs imputation for datasets with missing data only. MImpute_surv() performs imputation for a dataset with survival data. The Nelson Aalen estimator is calculated and used as predictor in the imputation, Time is not used as predictor. MImpute_lcens() performs imputation for a dataset with left-censored data. Note that with MImpute_lcens() pmm imputation is performed for variables not affected by left-censoring.

Usage

MImpute(
  data,
  mi.m,
  method = NULL,
  predMat = NULL,
  maxit = 10,
  return.midsObject = FALSE
)

MImpute_surv(
  data,
  mi.m,
  time.status.names = c("time", "status"),
  return.midsObject = FALSE
)

MImpute_lcens(
  data,
  data.lod,
  standards,
  mi.m,
  mice.log = 10,
  maxit = 10,
  return.midsObject = FALSE
)

MImpute_lcenssurv(
  data,
  mi.m,
  time.status.names = c("time", "status"),
  data.lod,
  standards,
  mice.log = 10,
  var.log = NULL,
  maxit = 10,
  return.midsObject = FALSE
)

Arguments

data	Dataframe with incomplete data. (for MImpute_lcens(), with NA for both missing and left-censored data).
mi.m	Number of imputations to perform.
method	Optional. single string, or a vector of strings specifying the imputation method to be used for each column in data (passed to mice()). If NULL default mice setting are used.
predMat	Optional. supply a predictorMatrix (passed to mice()). If NULL default mice setting are used.
maxit	passed to mice().
return.midsObject	Boolean
time.status.names	Names of the variables for time and status (in that order).
data.lod	Dataframe containing indicators of which observation are left-censored (censoring value for such observations and any other values for not censored observations). The colnames should correspond to variables in data. The variables that are left-censored are thus given in data (with left-censored data as NA) and in data.lod with random values for observed data and the LOD for left-censored data. Note that if the data are to be logged (is.numeric(mice.log)), only the argument data will be logged, therefore, the LOD values given here should be given as log(LOD) with the correct number of decimals: round(log(LOD), mice.log).
standards	Dataframe of 1 row containing the LOD values (not logged, whatever the value for mice.log).
mice.log	set to FALSE if the imputation should be performed on unlogged data. Otherwise, number of decimal to save after taking the log of data (should be 10 unless for specific reasons) ; in that case the data will be unlogged after imputation.
var.log	names of variables to log if mice.log is numeric. If NULL, all variables but those in time.status.names will be logged.

Value

If return.midsObject == FALSE a list of size mi.m, containing the imputed datasets. If return.midsObject == TRUE a list of 2, the first element (imputed.data) being the list of size mi.m as described in the previous sentence, the 2nd element (mids.obj) containing the mids object as returned by mice()

Examples

data(cancer, package = "survival")
cancer.imp <- MImpute(cancer[, -c(1:3)], 2)

## MImpute_surv
data(cancer, package = "survival")
cancer$status <- cancer$status - 1
cancer.imp <- MImpute_surv(cancer, 1)

## MImpute_lcens
toy <- iris[, 1:4]
# censor on variables 3 and 4, with LOD at quantile .1 and .2.
LODs <- toy[1, ]
LODs[1, ] <-c(NA, NA, quantile(toy[,3], .2), quantile(toy[,4], .1))
# Censor indicator
Censored <- data.frame(Petal.Length = runif(150, 50,60),
                       Petal.Width = runif(150, 50,60))
Censored[toy[,3] < LODs[1, 3], 1] <- LODs[1, 3]
Censored[toy[,4] < LODs[1, 4], 2] <- LODs[1, 4]
# NA for censored data
toy[toy[,3] < LODs[1, 3], 3] <- NA
toy[toy[,4] < LODs[1, 4], 4] <- NA
# Additional missing data
toy[sample(1:nrow(toy), 30), 1] <- NA
toy[sample(1:nrow(toy), 30), 3] <- NA
toy[sample(1:nrow(toy), 30), 4] <- NA

toy.imp <- MImpute_lcens(data = toy, data.lod = Censored, standards = LODs,
                         mi.m = 1, mice.log = FALSE)
## MImpute_lcenssurv
data(cancer, package = "survival")
cancer$status <- cancer$status - 1
toy2 <- cancer[, -1]
# censor on variables age and meal.cal, with LOD at quantile .1 and .2.
LODs <- toy2[1, ]
LODs[1, ] <-c(NA, NA, quantile(toy2[, "age"], .2, na.rm = TRUE), NA,  NA,
              NA, NA, quantile(toy2[, "meal.cal"], .1, na.rm = TRUE), NA)
# Censor indicator
Censored <- data.frame(age = runif(nrow(toy2), 300,400),
                       meal.cal = runif(nrow(toy2), 50,60))
Censored[toy2[, "age"] < LODs[1, "age"], "age"] <- LODs[1, "age"]
Censored[!is.na(toy2[, "meal.cal"]) &
           toy2[, "meal.cal"] < LODs[1, "meal.cal"], "meal.cal"] <-
 LODs[1, "meal.cal"]
# NA for censored data
toy2[toy2[, "age"] < LODs[1, "age"], "age"] <- NA
toy2[!is.na(toy2[, "meal.cal"]) &
      toy2[, "meal.cal"] < LODs[1, "meal.cal"],"meal.cal"] <- NA
# Additional missing data
toy2[sample(1:nrow(toy2), 30), 6] <- NA
toy2[sample(1:nrow(toy2), 30), 3] <- NA
toy2[sample(1:nrow(toy2), 30), 4] <- NA
toy2$sex <- factor(toy2$sex)
toy2$ph.ecog <- factor(toy2$ph.ecog)

toy2.imp <- MImpute_lcenssurv(
  data = toy2, mi.m = 1, data.lod = Censored, standards = LODs,
  mice.log = FALSE)

---