A latent topic model for mining heterogenous non-randomly missing electronic health records data

TL;DR

This paper introduces mixEHR, a novel unsupervised model that effectively analyzes heterogeneous and biased electronic health records to uncover disease patterns, impute missing data, and predict patient outcomes.

Contribution

The paper presents mixEHR, a new generative model combining collaborative filtering and latent topic modeling for EHR data analysis, addressing heterogeneity and bias.

Findings

mixEHR outperforms previous methods in simulations and real data

It reveals meaningful multi-disease insights from EHR data

The model improves data imputation and mortality prediction

Abstract

Electronic health records (EHR) are rich heterogeneous collection of patient health information, whose broad adoption provides great opportunities for systematic health data mining. However, heterogeneous EHR data types and biased ascertainment impose computational challenges. Here, we present mixEHR, an unsupervised generative model integrating collaborative filtering and latent topic models, which jointly models the discrete distributions of data observation bias and actual data using latent disease-topic distributions. We apply mixEHR on 12.8 million phenotypic observations from the MIMIC dataset, and use it to reveal latent disease topics, interpret EHR results, impute missing data, and predict mortality in intensive care units. Using both simulation and real data, we show that mixEHR outperforms previous methods and reveals meaningful multi-disease insights.