Rerr-package: Excessive Relative Risk (ERR) model fitting and inference

In zhuozhang/Rerr: Excessive Relative Risk (ERR) model fitting and inference

Description

A package that fits and make inference of a ERR model with adjustment for measurement errors described by a Monte Carlo dosimetry system.

Details

Rerr is a package that fits excess relative risk (ERR) models in survival analysis, while allowing for measurement error in exposures, with or without effect modifications. Currently this package only allow one of the exposure variables to be measured with error.

To learn to use this package, check the Rerr Vignette, or see the example provided below.

Author(s)

Zhuo Zhang, Daniel Stram Maintainer: Zhuo Zhang

References

TODO

See Also

simulateERR parseERR fitERR inferERR

Examples

## a simple simulation study

library(Rerr)

n <- 5000                              # size of the cohort
n.year <- 10                           # years of follow-up
n.rep <- 400                           # number of dose realizations to generate
id <- rep(1:n, each = n.year)          # id for py table
start.age <- sample(20:50, n, TRUE)    # age of enrollment of each person
age <- unlist(lapply(start.age, function(x)
    x + 1:n.year - 1))                 # age for py table
sex <- rep(rbinom(n, 1, 0.5),
           each = n.year)              # sex for py table
time <- rep(1:n.year - 1, n)           # time variable for py table
censor.time <- rep(n.year, n * n.year) # censor time for py table
x1 <- rchisq(n * n.year, 1)            # exposure 1 for py table
x2 <- rchisq(n * n.year, 1)            # exposure 2 for py table

sm.var <- 0.2                          # variance of shared multiplicative error
sm.log.se <- log(1 + sm.var)           # standard error on log scale
sm.log.mu <- - 1/2 * sm.var            # mean on log scale
um.var <- 0.5                          # variance of unshared multiplicative error
um.log.se <- log(1 + um.var)           # standard error on log scale
um.log.mu <- - 1/2 * um.var            # mean on log scale
x1.reps <- do.call(cbind, lapply(1:n.rep, function(i) {
    x1 * exp(rnorm(1, sm.log.mu, sm.log.se)) *
        exp(rnorm(n * n.year, um.log.mu, um.log.se))}))

rep.index <- 1:n.rep
rep.index <- rep.index[-1]
x1.mean <- apply(x1.reps[, rep.index], 1, mean)
data <- data.frame(sex = sex,
                   age = age,
                   x1 = x1.reps[, 1],
                   x2 = x2,
                   id = id,
                   time = time,
                   censor.time = censor.time)
data[1:10, ]

model <- case ~ I(log(age/40)^2) + sex +
    ERR(U(x1) | sex) + ERR(x2) +
    offset(py)
theta <- c(-5, log(2), 1, 2, 0.5, 1)

data.new <- simulateERR(model, data, theta,
                        id = id,
                        time = time,
                        censor.time = censor.time)
number.case <- sum(data.new$case)

number.case

data.new[1:10,]

data.new$x1 <- x1.mean[data.new$index]
data.err <- parseERR(model, data.new)
xm <- fitERR(data.err)
xm$conv
xm$theta
theta.CI <- inferERR(data.err, xm$theta,
                     x1.reps[data.new$index, rep.index],
                     naive = TRUE)
theta.CI

zhuozhang/Rerr documentation built on May 4, 2019, 11:22 p.m.