Introduction

PLPBenchmarks is an R package for building benchmarks of patient-level prediction models using data in the OMOP Common Data Model format.

The package is build on top of the PatientLevelPrediction, an R package to develop and validate prediction models.

Features

Contains the necessary artifacts to reproduce and replicate the benchmark problems predefined in this package.
Provides problem specifications for all problems, as well as, their model designs used to develop the models.
Takes one or more model designs as generated by PatientLevelPrediction::createModelDesign(). Within a model design, one can specify covariate settings, algorithms, inclusion criteria for the target cohort etc.
Extracts the necessary data from a database in OMOP Common Data Model.
Includes functions to explore model performance.
Includes a shiny app to interactively view and explore results.

A shiny app containing benchmarks developed with LASSO can be found here:

PLPBenchmarks is an R package.

Requires R (version 3.3 or higher). Installation on Windows requires RTools. PatientLevelPrediction use libraries that require Java and Python. See the PatientLevelPrediction package’s installation guide

To install the package call remotes::install_github("mi-erasmusmc/PLPBenchmarks") or fork or clone the package.

Package function reference: Reference

Documentation can be found on the package website.

Read here how you can contribute to this package.

PLPBenchmarks is licensed under Apache License 2.0

PLPBenchmarks is being developed in R Studio.

The package is maintained by Solomon Ioannou, with Ross D. Williams being a contributor. Jenna M. Reps has made substantial suggestions to the structure of the package.
We like to thank Evan Minty, Andreas Weinberger Rosen and Koen Zwart for their major contributions to some of the cohort definitions included in this package.
Of course, extended thanks to the whole OHDSI community for insights into clinically useful prediction problems.