cut_tv: cut_tv - Cuts a survival dataset on a time varying variable

In rEHR: Manipulating and Analysing Electronic Health Record Data

Description

Survival datasets often have time-varying covariates that need to be dealt with. For example a drug exposure may occur after the entry into the cohort and you are interested in how this might affect your outcome.

Usage

cut_tv(.data, entry, exit, cut_var, tv_name, cores = 1, id_var,
  on_existing = c("flip", "increment"))

Arguments

.data	a dataframe
entry	name of the column in .data that defines entry time to cohort. Column must be numeric.
exit	name of the column in .data that defines exit time from cohort. Column must be numeric
cut_var	name of the column in .data that defines the time of the time-varying covariate event. Column must be numeric.
tv_name	name for the constructed time-varying covariate
cores	number of mc.cores to use.
id_var	name of the variable identifying individual cases
on_existing	see details for cutting behaviour

Details

This function cuts up a dataset based on times supplied for the time-varying covariate. If there is already a variable for the time-varying covariate, you can chose to flip the existing values or increment them. This means the function can be called multiple times to, e.g. deal with drugs starting and stopping and also to deal with progression of treatment.

The function is faster than other cutting methods, does not require conversion to Lexis format, and can be parallelised for large datasets and chained with dply workflows. Arguments should not be quoted.

This function can deal with the following scenarios (see examples):

• "Binary chronic covariates"e.g. The time of diagnosis for a chronic (unresolvable) condition. This requires a single column variable of times from entry in the dataset

• "Binary covariates"e.g. times of starting and stopping medication. This requires more than one column variable in the dataset, one for each start or stop event. The state flips with each new change.

• "Incremental time-varying covariates"e.g. different stages of a condition. This requires a single column variable for each incremental stage

• "Any combination of the above"This is achieved by chaining multiple calls together

Examples

# A simple example dataset to be cut
tv_test <- data.frame(id = 1:5, start = rep(0, 5), end = c(1000, 689, 1000, 874, 777), 
                      event = c(0,1,0,1,1), drug_1 = c(NA, NA, NA, 340, 460),
                      drug_2 = c(NA, 234, 554, 123, NA), 
                      drug_3_start = c(110, 110,111, 109, 110),
                      drug_3_stop = c(400, 400, 400, 400, 400),
                      stage_1 = c(300, NA, NA, NA, NA),
                      stage_2 = c(450, NA, NA, NA, NA))

# Binary chronic covariates:
tv_out1 <- cut_tv(tv_test, start, end, drug_1, id_var = id, drug_1_state)
tv_out1 <- cut_tv(tv_out1, start, end, drug_2, id_var = id, drug_2_state)
# Binary covariates:
tv_out3 <- cut_tv(tv_test, start, end, drug_3_start, id_var = id, drug_3_state)
tv_out3 <- cut_tv(tv_out3, start, end, drug_3_stop, id_var = id, drug_3_state)
# incremental covariates:
inc_1 <- cut_tv(tv_test, start, end, stage_1, id_var = id, disease_stage, on_existing = "inc")
inc_1 <- cut_tv(inc_1, start, end, stage_2, id_var = id, disease_stage, on_existing = "inc")
# Chaining combinations of the above 
## Not run: 
library(dplyr)
tv_all <- tv_test %>%
          cut_tv(start, end, drug_1, id_var = id, drug_1_state) %>% 
          cut_tv(start, end, drug_2, id_var = id, drug_2_state) %>%
          cut_tv(start, end, drug_3_start, id_var = id, drug_3_state) %>%
          cut_tv(start, end, drug_3_stop, id_var = id, drug_3_state) %>%
          cut_tv(start, end, stage_1, id_var = id, disease_stage, on_existing = "inc") %>%
          cut_tv(start, end, stage_2, id_var = id, disease_stage, on_existing = "inc")

## End(Not run) 

rEHR documentation built on Jan. 23, 2017, 1 p.m.