Identification of Latent Patient Phenotype from Electronic Health Records (EHR) Data using Variational Bayes Gaussian Mixture Model for Latent Class Analysis and Variational Bayes regression for Biomarker level shifts, both implemented by Coordinate Ascent Variational Inference algorithms. Variational methods are used to enable Bayesian analysis of very large Electronic Health Records data. For VB GMM details see Bishop (2006,ISBN:9780-387-31073-2). For Logistic VB see Jaakkola and Jordan (2000)
Variational Bayes for Latent Patient Phenotypes in Electronic Health Records (EHR)
VBphenoR is an R package for discovering latent patient phenotypes from realistically large EHR data using Bayesian statistics.
In order to computationally support EHR data, we employ variational Bayes (VB). Currently, it supports latent class discovery
using VB Gaussian Mixture Model implemented with Coordinate-ascent Variational Inference (CAVI) and VB Logistic Regression for
biomarker levels shifted for the latent phenotype. Please note this package is still under development.
Prior to analyzing your EHR data, the R package needs to be installed. The
easiest way to install VBphenoR is through CRAN:
install.packages("VBphenoR")
VBphenoR can also be downloaded and installed via GitHub. This is most useful for downloading a specific version of the package (which
can be found at https://github.com/buckleybrian/VBphenoR/releases):
devtools::install_github("buckleybrian/[email protected]")
# or
devtools::install_version("VBphenoR", version = "x.x.x", repos = "http://cran.us.r-project.org")
The latest developmental version of the package can be downloaded and installed by running:
devtools::install_github("buckleybrian/VBphenoR", build_vignettes = TRUE, build_manual=TRUE)
After successful installation, the package must be loaded into the working space:
library(VBphenoR)
See the vignette for usage instructions and example.
VBphenoR is available under the open source MIT License