Skip to contents

Graphical representation of potential safety signal detection for a given ongoing trial sample size.

Source: R/approx_bssd_graph.R
approx_bssd_graph.Rd

This function creates a graph that represents all possible occurrences of observed events for an ongoing trial given a fixed sample size. The x-axis represents the number of participants (also equivalent to time since time can be expressed in a completed-trial-participant unit). The y-axis represents the number of participants with one AE. Risk levels, i.e., Pr(active AE rate > placebo AE rate), are expressed via color coding: green, yellow, and red representing risk levels that would warrant no action, a warning, and an alert, respectively. These three levels are delimited by two thresholds that are set to 80% and 90% by default. This function assumes that all the enrolled patients are completers (no dropout). The threshold levels can be changed.

Usage

approx_bssd_graph(
  time,
  prob_arm1,
  pi1_alpha,
  pi1_beta,
  pi2_alpha = 0.3,
  pi2_beta = 0.3,
  num_samples = 50000
)

Arguments

time

numeric, total time at risk (units = study duration). It can also be construed as the actual number of participants in the ongoing trial, and not the effective number of patients. This parameter expects a positive whole number.

prob_arm1

numeric, probability of being randomized to placebo (arm1).

pi1_alpha

numeric, first shape param of beta prior distribution on unit-time event probability in placebo (arm1).

pi1_beta

numeric, second shape param of beta prior distribution on unit-time event probability in placebo (arm1).

pi2_alpha

numeric, first shape param of beta prior distribution on unit-time event probability in active (arm2).

pi2_beta

numeric, second shape param of beta prior distribution on unit-time event probability in active (arm2).

num_samples

integer, number of replication of the simulated samples. More is better, up to a point. The default is 50000 and gives robust estimations of Pr(active AE rate > placebo AE rate). This parameter can be changed to a smaller number to fasten the computation time, but the estimation will be less accurate.

Value

list, ggplot2 graph.

Examples

# Note: Set seed for reproducible graph.
set.seed(2021)
approx_bssd_graph(
  time = 80, prob_arm1 = 1 / 3,
  pi1_alpha = 151.3025005, pi1_beta = 1784.3675521, pi2_alpha = 0.3,
  pi2_beta = 0.3, num_samples = 1000
)