Simulation study for volume-outcome analyses

This R code provides tools for volume-outcome analyses as they are proposed in the paper Modelling volume-outcome relationships in health care and to ascertain the reproducibility of the simulation study conducted in that paper.

Files

The simulation is contained in four R scripts. Successively executing simulation_generate.R, simulation_analyze.R and simulation_plot.R produces the numbers and figures as included in the paper.

1. simulation_generate.R: This file contains code to simulate data to be used in the analysis. One may freely adjust all parameters introduced in the paper.

2. simulation_analyze.R: A volume-outcome-analysis is conducted on the data generated through simulation_generate.R. This includes coefficient estimates, p-values for the volume effect and odds ratios with respect to the volume effect.

3. simulation_plot.R: This file contains code for constructing and exporting a number of plots based on the results obtained in simulation_analyze.R. It covers the estimated (versus underlying) volume effects and corresponding p-values, random effects standard deviations and odds ratios with respect to the volume effects.

The full simulation study with all combinations of parameter values as in the paper takes around two to three days on a server with 16 cores at 2.9 GHz. For a single combination of the parameter values, generating and analyzing the data can be done within a few minutes on a normal desktop computer.

System requirements

The code assumes a recent version of R and makes use of the libraries mgcv, dplyr, tidyr, ggplot2, scales, extrafont and sessioninfo. The simulation results in the paper were obtained with R-4.0.2 on a Windows server (x64).

License

MIT