Example 2: Two arm | Interim | Continuous

library(rxsim)

This example extends Example 1 by introducing two named analyses — interim and final — that fire at different enrollment milestones. Understanding how analysis names propagate through collect_results() is the foundation for multi-analysis designs shown in later examples.

Scenario

Capture scenario parameters. We will assume piece-wise linear enrollment.

sample_size <- 100
arms        <- c("pbo", "trt")
allocation  <- c(1, 1)
delta       <- 0.2  # true treatment - placebo mean difference
enrollment_fn <- function(n) rexp(n, rate = 1)
dropout_fn    <- function(n) rexp(n, rate = 0.01)
scenario <- tidyr::expand_grid(
  sample_size = sample_size,
  allocation  = list(allocation),
  delta       = delta
)

allocation = c(1, 1) specifies balanced randomisation. tidyr::expand_grid() embeds design parameters directly into each result row for traceability across parameter sweeps.

Populations

Define population generators.

population_generators <- list(
  pbo = function(n) data.frame(
    id = 1:n,
    value = rnorm(n),
    readout_time = 1
  ),
  trt = function(n) data.frame(
    id = 1:n,
    value = rnorm(n, mean = delta),
    readout_time = 1
  )
)

Each generator is a function of n returning a data.frame with one row per subject. readout_time = 1 means the endpoint is observed 1 time unit after enrollment. The treatment arm has a mean shift of δ = 0.2 SD units over placebo.

Triggers & Analysis

Final analysis at full enrollment.

analysis_generators <- list(
  interim = list(
    trigger = rlang::exprs(
      sum(!is.na(enroll_time)) >= (!!sample_size) / 2
    ),
    analysis = function(df, timer){
      df_enrolled <- df |> subset(!is.na(enroll_time))
      tt <- t.test(value ~ arm, data = df_enrolled)
      data.frame(
        scenario,
        n_total = nrow(df_enrolled),
        mean_pbo = mean(df_enrolled$value[df_enrolled$arm == "pbo"]),
        mean_trt = mean(df_enrolled$value[df_enrolled$arm == "trt"]),
        p_value = unname(tt$p.value),
        stringsAsFactors = FALSE
      )
    }
  ),
  
  final = list(
    trigger = rlang::exprs(
      sum(!is.na(enroll_time)) >= !!sample_size
    ),
    analysis = function(df, timer){
      df_enrolled <- df |> subset(!is.na(enroll_time))
      tt <- t.test(value ~ arm, data = df_enrolled)
      data.frame(
        scenario,
        n_total = nrow(df_enrolled),
        mean_pbo = mean(df_enrolled$value[df_enrolled$arm == "pbo"]),
        mean_trt = mean(df_enrolled$value[df_enrolled$arm == "trt"]),
        p_value = unname(tt$p.value),
        stringsAsFactors = FALSE
      )
    }
  )
)

Two analyses are defined. The interim analysis fires at half enrollment and the final analysis fires when all subjects are enrolled. These names appear in the analysis column of collect_results() output, making it straightforward to distinguish interim from final results when stacking rows across timepoints.

Trial

Make multiple trial replicates.

trials <- replicate_trial(
  trial_name = "test_trial",
  sample_size = sample_size,
  arms = arms,
  allocation = allocation,
  enrollment = enrollment_fn,
  dropout = dropout_fn,
  analysis_generators = analysis_generators,
  population_generators = population_generators,
  n = 3
)

Simulate

To simulate all replicates:

run_trials(trials)
#> [[1]]
#> <Trial>
#>   Public:
#>     clone: function (deep = FALSE) 
#>     conditions: list
#>     initialize: function (name, seed = NULL, timer = NULL, population = list(), 
#>     locked_data: list
#>     name: test_trial_1
#>     population: list
#>     results: list
#>     run: function () 
#>     seed: NULL
#>     timer: Timer, R6
#> 
#> [[2]]
#> <Trial>
#>   Public:
#>     clone: function (deep = FALSE) 
#>     conditions: list
#>     initialize: function (name, seed = NULL, timer = NULL, population = list(), 
#>     locked_data: list
#>     name: test_trial_2
#>     population: list
#>     results: list
#>     run: function () 
#>     seed: NULL
#>     timer: Timer, R6
#> 
#> [[3]]
#> <Trial>
#>   Public:
#>     clone: function (deep = FALSE) 
#>     conditions: list
#>     initialize: function (name, seed = NULL, timer = NULL, population = list(), 
#>     locked_data: list
#>     name: test_trial_3
#>     population: list
#>     results: list
#>     run: function () 
#>     seed: NULL
#>     timer: Timer, R6

Bind one row per replicate into one data frame.

replicate_results <- collect_results(trials)
replicate_results
#>   replicate timepoint analysis sample_size allocation delta n_total    mean_pbo
#> 1         1  48.42750  interim         100       1, 1   0.2      50 -0.15458732
#> 2         1  87.36847    final         100       1, 1   0.2     100 -0.02475451
#> 3         2  40.26049  interim         100       1, 1   0.2      50 -0.04934773
#> 4         2  83.92580    final         100       1, 1   0.2     100 -0.05831826
#> 5         3  43.51109  interim         100       1, 1   0.2      50 -0.14213989
#> 6         3  96.44952    final         100       1, 1   0.2     100 -0.25739017
#>      mean_trt    p_value
#> 1  0.42624025 0.01626231
#> 2  0.32833062 0.05877739
#> 3  0.02698932 0.80070480
#> 4  0.13052493 0.36397787
#> 5 -0.13708169 0.98578393
#> 6  0.08945477 0.09253380

collect_results() prepends replicate, timepoint, and analysis columns to your result data. Each row shows either "interim" or "final" in the analysis column, reflecting which named analysis produced it. This naming convention becomes essential in later examples where multiple named analyses fire per replicate.

Example 2: Two arm | Interim | Continuous | t-test

Scenario

Populations

Triggers & Analysis

Trial

Simulate