The Use of Autoencoders for Discovering Patient Phenotypes

TL;DR

This paper explores the application of autoencoders to identify patient phenotypes by creating low-dimensional representations, comparing different autoencoder architectures on a large clinical dataset.

Contribution

It introduces a comparison between fixed-input and sequence-to-sequence autoencoders for patient phenotype discovery using clinical time series data.

Findings

Autoencoders effectively capture meaningful patient phenotypes.

Sequence-to-sequence autoencoders outperform fixed-input models.

The approach scales to large clinical datasets like MIMIC III.

Abstract

We use autoencoders to create low-dimensional embeddings of underlying patient phenotypes that we hypothesize are a governing factor in determining how different patients will react to different interventions. We compare the performance of autoencoders that take fixed length sequences of concatenated timesteps as input with a recurrent sequence-to-sequence autoencoder. We evaluate our methods on around 35,500 patients from the latest MIMIC III dataset from Beth Israel Deaconess Hospital.