R: Restructuring Data for Survival Analysis - r

Question

I am working with the R programming language.
I have the following data on medical patients:
my_data = data.frame(id = c(1,2,3), status_2017 = c("alive", "alive", "alive"), status_2018 = c("alive", "dead", "alive"), status_2019 = c("alive", "dead", "dead"), height_2017 = rnorm(3,3,3), height_2018 = rnorm(3,3,3),
height_2019 = rnorm(3,3,3) , weight_2017 = rnorm(3,3,3), weight_2018 = rnorm(3,3,3), weight_2019 = rnorm(3,3,3))
cols <- colnames(my_data)
ix <- my_data[, startsWith(cols, "status")] == "dead"
my_data[, startsWith(cols, "height")][ ix ] <- NA
my_data[, startsWith(cols, "weight")][ ix ] <- NA
This looks something like this:
id status_2017 status_2018 status_2019 height_2017 height_2018 height_2019 weight_2017 weight_2018 weight_2019
1 1 alive alive alive 3.7276706 4.524869 -1.648458 -1.702781 7.755581 3.369895
2 2 alive dead dead 0.7539518 NA NA 1.060408 NA NA
3 3 alive alive dead 6.6213771 2.122374 NA 5.114120 1.851467 NA
My Question: I want to restructure this data such that:
Each patient has its own row for each year
There is a "year" column
Status_2017, Status_2018, Status_2019 are all combined into a single column (i.e. "status")
Height_2017, Height_2018, Height_2019 are all combined into a single column (i.e. "height")
Weight_2017, Weight_2018, Weight_2019 are all combined into a single column (i.e. "weight")
A new variable ("new_var") is created such that if a patient id has a row with 2019 then new_var is always 0 - for all other patient id's, new_var is 0 until the max year (and new_var then is 1)
I tried to do this with the following code:
library(dplyr)
library(tidyr)
my_data_long <- na.omit(my_data %>%
pivot_longer(cols = -c(id, status_2017),
names_to = c(".value", "year"),
names_pattern = "(height|weight)_(\\d{4})") %>%
arrange(id, year))
final = my_data_long %>%
group_by(id) %>%
mutate(
new_var = ifelse(any(year == "2019"), 0, 1),
max_year = max(year)
) %>%
ungroup() %>%
mutate(
new_var = ifelse(year == max_year & new_var == 1, 1, 0),
max_year = NULL
)
The final result looks something like this:
> final
# A tibble: 6 x 6
id status_2017 year height weight new_var
<dbl> <chr> <chr> <dbl> <dbl> <dbl>
1 1 alive 2017 2.39 2.27 0
2 1 alive 2018 -0.541 1.63 0
3 1 alive 2019 -1.93 10.1 0
4 2 alive 2017 4.18 -3.35 1
5 3 alive 2017 -1.35 7.12 0
6 3 alive 2018 1.42 1.70 1
My end goal is to restructure this dataset such that I can fit a "time varying survival analysis model" (e.g. cox-ph) to this data (e.g. https://atm.amegroups.com/article/view/18820/html, https://cran.r-project.org/web/pacacages/survival/vignettes/timedep.pdf)
Can someone please tell me if I have done this correctly?
Thanks!
Note: I tried to add time differences for each ID
This looks something like this:
library(stringr)
final %>%
group_by(id) %>%
mutate(start = 0:(n() - 1),
end = 1:n()) %>%
ungroup()
# A tibble: 6 x 8
id status_2017 year height weight new_var start end
<dbl> <chr> <chr> <dbl> <dbl> <dbl> <int> <int>
1 1 alive 2017 2.39 2.27 0 0 1
2 1 alive 2018 -0.541 1.63 0 1 2
3 1 alive 2019 -1.93 10.1 0 2 3
4 2 alive 2017 4.18 -3.35 1 0 1
5 3 alive 2017 -1.35 7.12 0 0 1
6 3 alive 2018 1.42 1.70 1 1 2

Answer 1

if we need the status column, then we have to include those also in the pivoting to long i.e. cols = -c(id, status_2017) removes the 'status_2017' from reshaping. In addition, the names_pattern needs to include the status as well in addition to height and weight
library(dplyr) # version >= 1.1.0
library(tidyr)
my_data %>%
pivot_longer(cols = -id, names_to = c(".value", "year"),
names_pattern = "(height|weight|status)_(\\d{4})") %>%
drop_na() %>%
mutate(new_var = +(2019 %in% year), max_year = max(year), .by = "id") %>%
mutate(new_var = +(year == max_year & new_var), max_year = NULL)
-output
# A tibble: 6 × 6
id year status height weight new_var
<dbl> <chr> <chr> <dbl> <dbl> <int>
1 1 2017 alive 9.54 7.47 0
2 1 2018 alive 6.49 5.23 0
3 1 2019 alive 3.75 1.93 1
4 2 2017 alive 4.21 0.619 0
5 3 2017 alive 1.97 5.32 0
6 3 2018 alive -0.406 8.00 0