DeepMPM: a mortality risk prediction model using longitudinal EHR data

Abstract

Background Accurate precision approaches have far not been developed for modeling mortality risk in intensive care unit (ICU) patients. Conventional mortality risk prediction methods can hardly extract the information in longitudinal electronic medical records (EHRs) effectively, since they simply aggregate the heterogeneous variables in EHRs, ignoring the complex relationship and interactions between variables and the time dependence in longitudinal records. Recently deep learning approaches have been widely used in modeling longitudinal EHR data. However, most existing deep learning-based risk prediction approaches only use the information of a single disease, neglecting the interactions between multiple diseases and different conditions. Results In this paper, we address this unmet need by leveraging disease and treatment information in EHRs to develop a mortality risk prediction model based on deep learning (DeepMPM). DeepMPM utilizes a two-level attention mechanism, i.e. visit-level and variable-level attention, to derive the representation of patient risk status from patient’s multiple longitudinal medical records. Benefiting from using EHR of patients with multiple diseases and different conditions, DeepMPM can achieve state-of-the-art performances in mortality risk prediction. Conclusions Experiment results on MIMIC III database demonstrates that with the disease and treatment information DeepMPM can achieve a good performance in terms of Area Under ROC Curve (0.85). Moreover, DeepMPM can successfully model the complex interactions between diseases to achieve better representation learning of disease and treatment than other deep learning approaches, so as to improve the accuracy of mortality prediction. A case study also shows that DeepMPM offers the potential to provide users with insights into feature correlation in data as well as model behavior for each prediction.

On the evaluation of synthetic longitudinal electronic health records

“…Generating longitudinal instead of row-summarized EHRs provides a rich data representation required in a wide variety of research and real-world applications. Sequential patterns present in EHRs provide additional information for example in early disease detection [19], disease progression modelling [20,21], and mortality prediction [22,23].…”

“…Typical clinical tasks involving patient attributes and diagnoses sequences are, among other things, in-hospital mortality prediction using RNNs [23,51] and next-step diagnoses prediction using RNNs and attention networks [20,21,52]. To assess utility of the generated synthetic EHRs, we can compare performance in these respective tasks with the TSTR approach.…”

“…For showcasing the TSTR approach the comparison in performance is key, the level of performance less soas long as the algorithm has at least some predictive power. When aiming for the best performance possible in these tasks, we recommend to include additional features like socioeconomic status, physiological measurements and medications, which have empirically shown to be important [23,51,[53][54][55] -next to omitting chapter encoding of ICD-9 codes.…”

Enhancing Diagnosis Prediction in Healthcare With Knowledge-Based Recurrent Neural Networks

“…Stacked denoising autoencoders (SDAs) have been employed to identify characteristic physiological patterns in clinical time series data [2]. To capture the temporal dynamics in EHR data, recurrent neural networks (RNNs) have been extensively used for modeling disease progression [8], [27], handling time series healthcare data with missing values [28], [29], and performing diagnosis classification [30] and prediction [10], [13], [16]- [18], [23], [31]- [34].…”

“…This approach ignores the importance of modeling the sequential nature of EHR data. To capture sequential dependencies within healthcare records, state-of-the-art diagnosis prediction methods commonly employ recurrent neural networks (RNNs) [11]- [13], [16]- [18]. For instance, the reverse time attention model (RETAIN) [10] utilizes two time-ordered reverse RNNs with attention mechanisms to further boost the prediction performance.…”