run_simulation: Run Simulation

In jiyingz/clinicalMPI: The novel modified probability interval (mPI) phase I clinical trial design

Run Simulation

Description

Run simulations to get performance characteristics with given parameters. Performance outcomes are averaged over 1000 simulation trials by default.

Usage

library(stats)
true_probs = matrix(data = c(0.95, 0.0, 0.05,
                             0.5, 0.4, 0.1,
                             0.2, 0.65, 0.15,
                             0.1, 0.4, 0.5,
                             0.05, 0.0, 0.95), nrow = 5, ncol = 3, byrow = TRUE)

dose_levels = c(60,80,100,120,140)
starting_dose = 100
PF = 0.2
PF_tolerance = 0.05
eta = 0.8
PT = 0.2
PT_tolerance = 0.05
zeta = 0.8
#cohort size is 3, max sample size is 24, using defaults

results = run_simulation(true_probs, dose_levels, starting_dose, PF, PF_tolerance, eta, PT, PT_tolerance, zeta)

Arguments

true_probs	A Dx3 matrix containing the true chances of treatment futility, efficacy, and toxicity, where D = # of dose levels
dose_levels	A length D vector containing dose levels to be included in the trial. Can be in index or true value form (e.g. could be either c(1,2,3,4,5) or c(40,60,80,100,120)).
starting_dose	Which dose to start at, must be a value within dose_levels
PF	Target tolerable futility probability
PF_tolerance	Half-length of futility equivalence interval (epsilon_F)
eta	Upper futility threshold beyond which invokes the futility rule
PT	Target tolerable toxicity probability
PT_tolerance	Half-length of toxicity equivalence interval (epsilon_T)
zeta	Upper toxicity threshold beyond which invokes the safety rule
cohort_size	How many new patients to test in each round of the trial
max_samp_size	The maximum number of subjects for the entire trial (default = 24)
num_sims	How many trials to perform in simulation (default = 1000)
filter	Posterior sum threshold to determine dose availability for final selection (pf+pt <= filter) (default = 0.5)
seed	A random seed (default = 111)

Details

In order to run simulations, decision tables must be pre-specified (e.g. by using the make_decision_table() function) and loaded into the working environment. All decision tables must be named in the format "decision_table#", where # = dose sample size.

For simulations the cohort size must be a fixed value, although this may not necessarily be true in real-life trials.

Simulations are meant to give an idea of how variable trial results may be given the chosen parameters, and is not to be taken as a statistical guarantee of any sort. Results depend on the assumptions that

1. the "true" probabilities of futility and toxicity are valid, and

2. no unexpected cohort additions or dropouts occur throughout the trial.

These assumptions are difficult to adhere to in real-life scenarios, and so the actual performance characteristics may differ from those outputted by the simulation. In other words, take the simulation performance results with a grain of salt. However, the closer the real-life trial adheres to the ideal simulation conditions, and the more sure we are about the "true" chances of toxicity and futility, the closer the actual outcome will adhere to what is expected from simulation results! For example, if simulation results show that the correct dose is chosen in 73% of trials, then in one real-life trial there is a 73% chance that the trial outcomes will point to a truly acceptable dose.

Value

A list containing:

• prop_time_selected – Percent of trials where each dose was selected as the final dose recommendation

• times_early_stopping – How many trials stopped before reaching maximum sample size

• mean_patients_perdose – Mean number of patients allocated to each dose level

• min_patients_perdose – Minimum number of patients allocated to each dose level

• max_patients_perdose – Maximum number of patients allocated to each dose level

jiyingz/clinicalMPI documentation built on Jan. 19, 2024, 5:58 a.m.