Excel example 1 - extrapolation calculations

Introduction

This vignette describes how to work with the included example excel templates that are compatible to the survival models estimated with flexsurvPlus. These examples are deliberately simple and are intended to illustrate calculations in excel rather than as a basis for a real economic model. In this example the basic calculations needed to extrapolate survival are illustrated.

Set up packages and data

Install packages

The following packages are required to run this example:

rm(list = ls())
# Libraries
library(flexsurvPlus)
library(tibble)
library(dplyr)
library(boot)
library(ggplot2)

Generate the data

To perform survival analyses, patient level data is required for the survival endpoints. In this example, we analyze progression-free survival (PFS). For more details on these steps please refer to the other vignettes.

# make reproducible
set.seed(1234)

# used later
(simulation_seed <- floor(runif(1, min = 1, max = 10^8)))
#> [1] 11370342
(bootstrap_seed <- floor(runif(1, min = 1, max = 10^8)))
#> [1] 62229940

# low number for speed of execution given illustrating concept
n_bootstrap <- 10

adtte <- sim_adtte(seed = simulation_seed)
head(adtte)
#>   USUBJID ARMCD             ARM PARAMCD                     PARAM AVAL AVALU
#> 1       1     A Reference Arm A     PFS Progression Free Survival  108  DAYS
#> 2       2     A Reference Arm A     PFS Progression Free Survival  150  DAYS
#> 3       3     A Reference Arm A     PFS Progression Free Survival  372  DAYS
#> 4       4     A Reference Arm A     PFS Progression Free Survival   73  DAYS
#> 5       5     A Reference Arm A     PFS Progression Free Survival  137  DAYS
#> 6       6     A Reference Arm A     PFS Progression Free Survival  103  DAYS
#>   CNSR
#> 1    0
#> 2    0
#> 3    0
#> 4    0
#> 5    0
#> 6    0

# subset PFS data and rename
PFS_data <- adtte %>%
  filter(PARAMCD == "PFS") %>%
  transmute(USUBJID,
    ARMCD,
    PFS_days = AVAL,
    PFS_event = 1 - CNSR
  )

Fitting the models

More information about each function can be used by running the code ?runPSM or viewing the other vignettes.

psm_PFS_all <- runPSM(
  data = PFS_data,
  time_var = "PFS_days",
  event_var = "PFS_event",
  model.type = c(
    "Common shape",
    "Independent shape",
    "Separate"
  ),
  distr = c(
    "exp",
    "weibull",
    "gompertz",
    "lnorm",
    "llogis",
    "gengamma",
    "gamma",
    "genf"
  ),
  strata_var = "ARMCD",
  int_name = "B",
  ref_name = "A"
)

Bootstrap the estimated parameters

As described in other vignettes we can use boot to explore uncertainty.

# fix seed for reproducible samples
set.seed(bootstrap_seed)

boot_psm_PFS_all <- do.call(boot, args = c(psm_PFS_all$config, statistic = bootPSM, R = n_bootstrap))

Exporting to Excel

Once the values are calculated we can export to Excel. The following code prepares two tibbles that can be exported. One containing the main estimates. A second containing the bootstrap samples.

main_estimates <- psm_PFS_all$parameters_vector %>%
  t() %>%
  as.data.frame()


boot_estimates <- boot_psm_PFS_all$t %>%
  as.data.frame()

colnames(main_estimates) <- colnames(boot_estimates) <- names(psm_PFS_all$parameters_vector)

# can preview these tables

main_estimates[, 1:5] %>%
  pander::pandoc.table()
#> 
#> ----------------------------------------------------------------
#>  comshp.exp.rate.int   comshp.exp.rate.ref   comshp.exp.rate.TE 
#> --------------------- --------------------- --------------------
#>       0.002711              0.005584              -0.7226       
#> ----------------------------------------------------------------
#> 
#> Table: Table continues below
#> 
#>  
#> -----------------------------------------------------
#>  comshp.weibull.scale.int   comshp.weibull.scale.ref 
#> -------------------------- --------------------------
#>           370.9                      193.4           
#> -----------------------------------------------------

boot_estimates[1:3, 1:5] %>%
  pander::pandoc.table()
#> 
#> ----------------------------------------------------------------
#>  comshp.exp.rate.int   comshp.exp.rate.ref   comshp.exp.rate.TE 
#> --------------------- --------------------- --------------------
#>       0.002409              0.006054              -0.9214       
#> 
#>       0.002592              0.005716              -0.7907       
#> 
#>       0.002603              0.005959              -0.8283       
#> ----------------------------------------------------------------
#> 
#> Table: Table continues below
#> 
#>  
#> -----------------------------------------------------
#>  comshp.weibull.scale.int   comshp.weibull.scale.ref 
#> -------------------------- --------------------------
#>           410.9                       177            
#> 
#>           383.7                      188.9           
#> 
#>           384.4                      183.9           
#> -----------------------------------------------------

# the following code is not run in the vignette but will export this file

# require(openxlsx)
# wb <- openxlsx::createWorkbook()
# openxlsx::addWorksheet(wb, sheetName = "Exported data")
# openxlsx::writeDataTable(wb, sheet = "Exported data", main_estimates, startRow = 2, startCol = 2)
# openxlsx::writeDataTable(wb, sheet = "Exported data", boot_estimates, startRow = 5, startCol = 2)
# openxlsx::createNamedRegion(wb, sheet = "Exported data",
#                             cols = 2:(2+length(main_estimates)), rows = 3, name = "Estimates")
# openxlsx::createNamedRegion(wb, sheet = "Exported data",
#                             cols = 2:(2+length(main_estimates)), rows = 6:(6-1+nrow(boot_estimates)), name = "Samples")
# openxlsx::saveWorkbook(wb, file = "export_data.xlsx", overwrite = TRUE)

The Excel model

Included with the package is an example Excel file called ex1_calculation.xlsx. This can be extracted using the below code (not run). It can also be found in the github repository at https://github.com/Roche/flexsurvPlus/tree/main/inst/extdata

installed_file <- system.file("extdata/ex1_calculation.xlsx", package = "flexsurvPlus")
installed_file
#> [1] "/tmp/RtmpPdCdmH/Rinst10126e32932f/flexsurvPlus/extdata/ex1_calculation.xlsx"

# not run but will give you a local copy of the file
# file.copy(from = installed_file, to ="copy_of_ex1_calculation.xlsx")

This illustrates how all the included survival models can be extrapolated in Excel.

Exported data tab

This contains a copy of the data exported in the last step.

Exported data tab
Exported data tab

Extrapolations tab

This contains example calculations to extrapolate survival.

Extrapolations tab
Extrapolations tab

We can compare the approximate estimates of mean survival with those calculated in R. As the excel model only goes until time t=2000 we can more directly compare to the estimates of restricted mean survival time (rmst) until this time.

means_est <- psm_PFS_all %>%
  summaryPSM(type = c("mean", "rmst"), t = 2000)

# match to selected model in screenshot
means_est %>%
  dplyr::filter(Model == "Common shape", Strata == "Intervention") %>%
  tidyr::pivot_wider(
    id_cols = c("Dist"),
    names_from = c("type"),
    values_from = "value"
  ) %>%
  pander::pandoc.table()
#> 
#> -----------------------------------
#>        Dist          mean    rmst  
#> ------------------- ------- -------
#>     Exponential      368.9   367.3 
#> 
#>       Weibull         339     339  
#> 
#>      Gompertz        326.3   326.3 
#> 
#>     Log Normal       414.2   395.6 
#> 
#>    Log Logistic      459.4   405.6 
#> 
#>  Generalized Gamma   338.8   338.8 
#> 
#>        Gamma         346.8   346.8 
#> 
#>    Generalized F     342.6   342.5 
#> -----------------------------------