Deep Sequential Models for Suicidal Ideation from Multiple Source Data

TL;DR

This paper introduces a deep sequential modeling approach combining EHR and EMA data with attention mechanisms to improve the prediction of suicidal ideation, achieving higher recall and interpretability.

Contribution

It presents a novel RNN-based method that aligns asynchronous EHR and EMA sequences with attention and pre-training, enhancing prediction accuracy and interpretability.

Findings

EMA data improves recall from 48.13% to 67.78%.

The model offers interpretability via t-SNE visualization.

Most relevant features are identified and medically interpreted.

Abstract

This article presents a novel method for predicting suicidal ideation from Electronic Health Records (EHR) and Ecological Momentary Assessment (EMA) data using deep sequential models. Both EHR longitudinal data and EMA question forms are defined by asynchronous, variable length, randomly-sampled data sequences. In our method, we model each of them with a Recurrent Neural Network (RNN), and both sequences are aligned by concatenating the hidden state of each of them using temporal marks. Furthermore, we incorporate attention schemes to improve performance in long sequences and time-independent pre-trained schemes to cope with very short sequences. Using a database of 1023 patients, our experimental results show that the addition of EMA records boosts the system recall to predict the suicidal ideation diagnosis from 48.13% obtained exclusively from EHR-based state-of-the-art methods to 67.78%. Additionally, our method provides interpretability through the t-SNE representation of the latent space. Further, the most relevant input features are identified and interpreted medically.