sureLDA: A Novel Multi-Disease Automated Phenotyping Method for the EHR

Overview

Surrogate-guided ensemble Latent Dirichlet Allocation (sureLDA) is a label-free multidimensional phenotyping method. It first uses the PheNorm or MAP algorithm to initialize probabilities based on two surrogate features for each target disease, and then leverages these probabilities to guide the LDA topic model to generate phenotype-specific topics. Finally, it combines phenotype-feature counts with surrogates via clustering ensemble to yield final phenotype probabilities.

Installation

Install stable version from CRAN:

install.packages("sureLDA")

Install development version from GitHub:

# install.packages("remotes")
devtools::install_github("remotes/sureLDA")

Citation

Ahuja Y, Zhou D, He Z, Sun J, Castro VM, Gainer V, Murphy SN, Hong C, Cai T. sureLDA: A multidisease automated phenotyping method for the electronic health record. J Am Med Inform Assoc. 2020 Aug 1;27(8):1235-1243. doi: 10.1093/jamia/ocaa079. PMID: 32548637; PMCID: PMC7481024.