Modeling multivariate clinical event time-series with recurrent temporal mechanisms
作者：
Highlights：
• We propose a novel event time-series model that can predict clinical events.
• Different temporal mechanisms are used to process complex aspects of the time-series.
• Information about distant past is modeled through hidden states from LSTM.
• Information on recently observed events is modeled through discriminative projections.
• Information about periodic events is modeled using an external memory mechanism.
• The memory is based on probability distributions of interevent gaps compiled from past data.
• We evaluate our proposed model on electronic health record (EHRs) data derived from MIMIC-III dataset.
• We show that our approach leads to improved prediction performance compared to multiple baselines.
摘要
•We propose a novel event time-series model that can predict clinical events.•Different temporal mechanisms are used to process complex aspects of the time-series.•Information about distant past is modeled through hidden states from LSTM.•Information on recently observed events is modeled through discriminative projections.•Information about periodic events is modeled using an external memory mechanism.•The memory is based on probability distributions of interevent gaps compiled from past data.•We evaluate our proposed model on electronic health record (EHRs) data derived from MIMIC-III dataset.•We show that our approach leads to improved prediction performance compared to multiple baselines.
论文关键词：Event time series prediction,Recurrent neural network,Sequential models,Clinical time series,Modeling electronic health record data
论文评审过程：Received 22 January 2020, Revised 26 December 2020, Accepted 10 January 2021, Available online 18 January 2021, Version of Record 23 January 2021.
论文官网地址：https://doi.org/10.1016/j.artmed.2021.102021
