shrinkDSM (version 1.0.0)

prep_tvinput: Prepare time-varying inputs for estimation of a dynamic survival model

Description

This function pre-processes time-varying inputs in such a way that shrinkDSM can work with time-varying inputs. Its main inputs are two data frames, namely surv_data and covariate_data. surv_data contains meta data about each observation (i.e. survival time and censoring indicator), while covariate_data contains the time-varying covariates (one per observation and time interval) and an index for which time interval each covariate is observed in. The two are merged together via an ID that needs to be unique for each observation and present in both surv_data and covariate_data.

Usage

prep_tvinput(
  surv_data,
  covariate_data,
  id_var,
  surv_var,
  delta_var,
  interval_var,
  covariate_id_var = id_var
)

Value

Returns an object of class data.frame and tvsurv to be used as an input in shrinkDSM.

Arguments

surv_data

data frame containing meta-data for each observation (survival time and censoring indicator) as well as an ID for each observation.

covariate_data

data frame containing the time-varying covariates (one per observation and time interval), an index for which time interval each covariate is observed in as well as an ID for each observation.

id_var

character string specifying the column name of the ID variable. If the name is different in surv_data and covariate_data, id_var will be used in surv_data, whereas covariate_id_var will be used in covariate_data.

surv_var

character string specifying the column name of the survival times in surv_data.

delta_var

character string specifying the column name of the status indicators in surv_data, 0 = censored or 1 = event observed..

interval_var

character string specifying the column name of the time interval ID in covariate_data.

covariate_id_var

character string specifying the column name of the ID variable in covariate_data. Defaults to id_var.

Author

Daniel Winkler

Peter Knaus

Examples

# A toy example with 5 observations and 2 covariates, observed over 3 time periods
set.seed(123)
n_obs <- 5
surv_var <- round(rgamma(n_obs, 1, .1)) + 1
delta_var <- sample(size = n_obs, c(0, 1), prob = c(0.2, 0.8), replace = TRUE)

surv_data <- data.frame(id_var = 1:n_obs, surv_var, delta_var)

# Determine intervals
S <- c(3, 11)

# Create synthetic observations for each individual
covariate_list <- list()

for (i in 1:n_obs) {
  nr_periods_survived <- sum(surv_var[i] > S) + 1
  covariate_list[[i]] <- data.frame(id_var = i,
                                    interval_var = 1:nr_periods_survived,
                                    x1 = rnorm(nr_periods_survived),
                                    x2 = rnorm(nr_periods_survived))
}

# Bind all individual covariate data frames together
# Each observation now has a covariate in each period they
# were observed in.
covariate_data <- do.call(rbind, covariate_list)

# Call prep_tvinput to pre-process for shrinkDSM
merged_data <- prep_tvinput(surv_data,
                            covariate_data,
                            id_var = "id_var",
                            surv_var = "surv_var",
                            delta_var = "delta_var",
                            interval_var = "interval_var")

# Can now be used in shrinkDSM
# Note that delta is now automatically extracted from merged_data,
# providing it will throw a warning
mod <- shrinkDSM(surv_var ~ x1 + x2, merged_data, S = S)