The purpose of the simulation was to compare the bias in indirect or mixed direct/indirect treatment effect estimation made by an aggregated-data (AD) based approach vs an Individual Patients Data (IPD)-based approach in a simple 3 nodes network.
The simulation parameters were as follows:
Two configurations :
- Network without a closed loop
- Network with a closed loop
Two treatment effects (log HR):
- -0.2
- -0.5
Two random effects were simulated: one for the trial specific baseline risk and one for the treatement effect. Both with variance :
- 0.1
- 0.01
Three scenarios :
- None: same age distribution and no interaction
- Interaction: interaction in one comparison not the other but same age distribution
- Both: interaction in one comparison not the other and age distribution differences between comparison
There is a lot of simulation so the simulation that will usually excess the RAM, so the data are splits in chunk.
The code is parallelized by furrr
sim_parallel_3a : generate the simulation data (allowing to divide the data in chunks before processing) sim_parallel_3b : perform the analysis (chunk by chunk) sim_parrallel_3c : encapsulate all the analysis into a single function so that it can be efficiently run in parallel fonctions : the fonctions used.