MIICD (version 1.1)

MIICD.crreg: Multiple Imputation for competing risks regression

Description

This function uses the multiple imputation approach of Pan (2000) to handle interval censored data with competing outcomes using the Fine and Gray Propotional Hazards model (or Cox Proportional Hazard Model). It calls iteratively the crr function on imputed datasets and derives multiple estimates from imputed data sets. Finally it combines multiple estimates following multiple imputation rules (Rubin 1987; Schenker and Welsh 1988; Tanner and Wong 1987b) to update parameter estimation. The process stops once the desired number of iteration have been reached. Both the Poor Man's Data Augmentation (PMDA) as well as the Asymptotic Normal Data Augmentation (ANDA) scheme are implemented (Pan 2000; Wei and Tanner 1991; Tanner and Wong 1987a). When ANDA is chosen, the function calls mvrnorm from package MASS.

Usage

MIICD.crreg(formula, data, imax = 25, k = 10, th0 = 1e-3, status, trans,
cens, keep, method = c("PMDA", "ANDA"), model = c("FG", "Cox"))

Arguments

formula

a formula

data

a data frame to look for interval censored data and covariates

imax

maximum number of iteration

k

number of dataset generated at each iteration

th0

note used currently

status

collumn name were the status of observation is indicated

trans

what is the transition of interest in the status comlumn

cens

code used for censored data in the status column

keep

must be the same as the covariate used in formula

method

which imputation scheme shall be used PMDA or ANDA

model

which model shall be used currently FG state for Fine and Gray proportional hazards regression model

Value

• Mean beta
  estimation of coefficients computed using the last iterations
• Mean sigma
  estimation of sigma computed using the last iterations
• betas
  list of coefficient estimates for each iteration
• call
  function call
• df
  results returned in a data frame
• niter
  number of iteration
• conv
  successive mean of coefficient estimates over iterations
• sigma1
  mean of standard errors not augmented for the between imputations composant of the variance

Details

The data frame MUST have one column named 'left' one column 'right'. Interval censored data are observation for which 'left' < 'right'. Inf in the 'right' column stands for right sensored observations. Competing event must have exact failure time, that is 'left' = 'right'.

References

PAN, Wei. A Multiple Imputation Approach to Cox Regression with Interval-Censored Data. Biometrics, 2000, vol. 56, no 1, p. 199-203.

Rubin, D. B. (1987). Multiple imputation for nonresponse in surveys.

Schenker, N. and Welsh, A. (1988). Asymptotic results for multiple imputation. The Annals of Statistics pages 1550-1566.

Tanner, M. A. and Wong, W. H. (1987). An application of imputation to an estimation problem in grouped lifetime analysis. Technometrics 29, 23-32.

Wei, G. C., & Tanner, M. A. (1991). Applications of multiple imputation to the analysis of censored regression data. Biometrics, 47(4), 1297-1309.

See Also

MIICD.coxph crr mvrnorm