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.
- 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.
- We use the GitHub issue tracker for all bugs/issues/enhancements.
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.