generate_subgroup {SimNPH}

Generate Dataset with different treatment effect in subgroup

Description

Generate Dataset with different treatment effect in subgroup

Create an empty assumtions data.frame for generate_subgroup

Calculate true summary statistics for scenarios with differential treatment effect in subgroup

Calculate hazards in treatment arm in subgroup and compliment

Usage

generate_subgroup(condition, fixed_objects = NULL)

assumptions_subgroup(print = interactive())

true_summary_statistics_subgroup(
  Design,
  cutoff_stats = NULL,
  milestones = NULL,
  fixed_objects = NULL
)

hazard_subgroup_from_PH_effect_size(
  design,
  target_power_ph = NA_real_,
  final_events = NA_real_,
  target_alpha = 0.025
)

cen_rate_from_cen_prop_subgroup(design)

Arguments

condition

condition row of Design dataset

fixed_objects

additional settings, see details

print

print code to generate parameter set?

Design

Design data.frame for subgroup

cutoff_stats

(optionally named) cutoff times, see details

milestones

(optionally named) vector of times at which milestone survival should be calculated

design

design data.frame

target_power_ph

target power under proportional hazards

final_events

target events for inversion of Schönfeld Formula, defaults to condition$final_events

target_alpha

target one-sided alpha level for the power calculation

Details

Condidtion has to contain the following columns:

• n_trt number of paitents in treatment arm

• n_ctrl number of patients in control arm

• hazard_ctrl hazard in the control arm

• hazard_trt hazard in the treatment arm for not cured patients

• hazard_subgroup hazard in the subgroup in the treatment arm

• prevalence proportion of cured patients

assumptions_subgroup generates a default design data.frame for use with generate_subgroup If print is TRUE code to produce the template is also printed for copying, pasting and editing by the user. (This is the default when run in an interactive session.)

cutoff_stats are the times used to calculate the statistics like average hazard ratios and RMST, that are only calculated up to a certain point.

hazard_subgroup_from_PH_effect_size calculates the hazard rate in the subgroup and the compliment of the subgroup in the treatment arm as follows: First, the hazard ratio needed to archive the desired power under proportional hazards is calculated by inverting Schönfeld's sample size formula. Second the median survival times for both arms under this hazard ratio and proportional hazards are calculated. Finally the hazard rate of the treatment arm in the subgroup and its complement are set such that the median survival time is the same as the one calculated under proportional hazards.

This is a heuristic and to some extent arbitrary approach to calculate hazard ratios that correspond to reasonable and realistic scenarios.

cen_rate_from_cen_prop_subgroup takes the proportion of censored patients from the column censoring_prop. This column describes the proportion of patients who are censored randomly before experiencing an event, without regard to administrative censoring.

Value

For generate_subgroup: A dataset with the columns t (time) and trt (1=treatment, 0=control), evt (event, currently TRUE for all observations)

For assumptions_subgroup: a design tibble with default values invisibly

For true_summary_statistics_subgroup: the design data.frame passed as argument with the additional columns

For hazard_subgroup_from_PH_effect_size: the design data.frame passed as argument with the additional columns hazard_trt and hazard_subgroup.

for cen_rate_from_cen_prop_subgroup: design data.frame with the additional column random_withdrawal

Functions

• generate_subgroup(): simulates a dataset with a mixture of cured patients

• assumptions_subgroup(): generate default assumptions data.frame

• true_summary_statistics_subgroup(): calculate true summary statistics for subgroup

• hazard_subgroup_from_PH_effect_size(): Calculate hazards in treatement arm

• cen_rate_from_cen_prop_subgroup(): calculate censoring rate from censoring proportion

Examples

one_simulation <- merge(
    assumptions_subgroup(),
    design_fixed_followup(),
    by=NULL
  ) |>
  head(1) |>
  generate_subgroup()
head(one_simulation)
tail(one_simulation)
Design <- assumptions_subgroup()
Design
my_design <- merge(
    assumptions_subgroup(),
    design_fixed_followup(),
    by=NULL
  )
my_design <- true_summary_statistics_subgroup(my_design)
my_design

my_design <- merge(
  assumptions_subgroup(),
  design_fixed_followup(),
  by=NULL
)

my_design$hazard_trt <- NA
my_design$hazard_subgroup <- NA
my_design$hr_subgroup_relative <- 0.9
my_design$final_events <- ceiling((my_design$n_ctrl + my_design$n_trt) * 0.75)
my_design <- hazard_subgroup_from_PH_effect_size(my_design, target_power_ph=0.9)
my_design
design <- expand.grid(
  hazard_ctrl=0.2,                   # hazard under control and before treatment effect
  hazard_trt=0.02,                   # hazard after onset of treatment effect
  hazard_subgroup=0.01,              # hazard in the subgroup in treatment
  prevalence = c(0.2, 0.5),           # subgroup prevalence
  censoring_prop=c(0.1, 0.25, 0.01), # 10%, 25%, 1% random censoring
  followup=100,                      # followup of 100 days
  n_trt=50,                          # 50 patients treatment
  n_ctrl=50                          # 50 patients control
)
cen_rate_from_cen_prop_subgroup(design)