PharmacoVigilance Comparison Study

This repository contains all the R code for a comparison study of 27 statistical measures in their ability to detect associations between drugs and adverse events in spontaneous reporting data. The code heavily relies on the following two R packages:

SRSim, see www.github.com/bips-hb/srsim, and
PVM, see www.github.com/bips-hb/pvm.

The code is published under the GPL-3 license.

Shiny

All the results can be explored interactively at https://srs.bips.eu/

Acknowledgements

We gratefully acknowledge the financial support from the innovation fund (“Innovationsfonds”) of the Federal Joint Committee in Germany (grant number: 01VSF16020).

Installation

To install, simply type in R

devtools::install_github("bips-hb/pvmcomparison")

Usage

In order to generate all the results as presented in the paper and at www.srs.bips.eu, install the package and run the run.R script:

source("run.R")

Example Simulated Data Set

The file example-SRS.rds contains an example of a simulated spontaneous reporting data set. The number of reports is 50,000. There are no innocent bystanders and the average odds ratio between drugs and adverse drug reactions (ADRs) are drawn for a truncated normal distribution with mean 1.5. The odds ratios can not be smaller than 1. The object stored in example-SRS.rds has six entries:

$sr - a binary matrix with 50,000 rows and 1,000 columns. Each row represents a report. The first 500 columns represent the drugs. The last 500 columns represent athe adverse drug events (ADEs) and the ADRs. In case that the entry is 1, the report contains that particular drug or that particular ADE/ADR
$prob_drugs - a vector with 500 entries. Each entry represents the baseline probability of that drug being prescribed when no other drugs are described
$prob_events - a vector with 500 entries. Each entry represents the baseline probability of that ADE/ADR occuring when no other drugs are prescribed

$nodes - a table with a 1,000 rows and 7 columns. Each row represents either a drug or an ADE/ADE. The columns represent:

label - the label, e.g., drug7 or event44
in_degree - the number of nodes in the directed acyclic graph (DAG) pointing to that variable (in this setting, either 0 or 1).
id - an integer ID (1 to 1,000)
parent_id - the ID of the node pointing to this node (1 to 1,000). If no node points to this variable, the value is set to -1
margprob - probability of the drug being prescribed when no other drugs are prescribed, or the probability of that ADE/ADR occurs when no drugs are prescribed
beta0 - the intercept of the logistic regression function
beta1 - the odds ratio with the parent node

$tables is a data frame with 250,000 rows and contains all 2 x 2 tables for each drug and ADE/ADR pair. It has the following columns:

drug_id - the ID of the drug
event_id - the ID of the event
prob_drug - the baseline probability of the drug being prescribed when no other drugs are prescribed
prob_event - the probability of the ADE/ADR occuring when no drugs are prescribed
or - the true odds ratio
a, b, c and d: the entries of the 2 x 2 table

References

Please cite

Adverse Drug Reaction or Innocent Bystander? A Systematic Comparison of Statistical Discovery Methods for Spontaneous Reporting Systems
L.J. Dijkstra, M. Garling, R. Foraita & I. Pigeot
Pharmacoepidemiology and Drug Safety (2020)
DOI:10.1002/PDS.4970

Acknowledgments

Funded by the Innovation Fund of the German Joint Federal Committee (G-BA, 01VSF16020)

Contact

Louis Dijkstra
Leibniz Institute for Prevention Research & Epidemiology
E-mail: dijkstra (at) leibniz-bips.de

tanboyu/pvmcomparison