Deep Learning from Heterogeneous Sequences of Sparse Medical Data for Early Prediction of Sepsis

Abstract

Sepsis is a life-threatening complication to infections, and early treatment is key for survival. Symptoms of sepsis are difficult to recognize, but prediction models using data from electronic health records (EHRs) can facilitate early detection and intervention. Recently, deep learning architectures have been proposed for the early prediction of sepsis. However, most efforts rely on high-resolution data from intensive care units (ICUs). Prediction of sepsis in the non-ICU setting, where hospitalization periods vary greatly in length and data is more sparse, is not as well studied. It is also not clear how to learn effectively from longitudinal EHR data, which can be represented as a sequence of time windows. In this article, we evaluate the use of an LSTM network for early prediction of sepsis according to Sepsis-3 criteria in a general hospital population. An empirical investigation using six different time window sizes is conducted. The best model uses a two-hour window and assumes data is missing not at random, clearly outperforming scoring systems commonly used in healthcare today. It is concluded that the size of the time window has a considerable impact on predictive performance when learning from heterogeneous sequences of sparse medical data for early prediction of sepsis.

Federated Semi-Supervised Multi-Task Learning to Detect COVID-19 and Lungs Segmentation Marking Using Chest Radiography Images and Raspberry Pi Devices: An Internet of Medical Things Application

“…Moreover, it is very crucial to consider the amount of time spent in some cases as well. For instance, sepsis detection [ 59 , 60 ] could be one such issue. It has also been argued that analyzing only one part of the whole data gathered for a diagnosis is not convincing enough to predict a particular disease, therefore using only chest radiography images to detect COVID-19 has also been criticized as it may not be quite convincing from a medical explanation perspective [ 61 ].…”

Predicting sepsis onset using a machine learned causal probabilistic network algorithm based on electronic health records data

“…There have been prior works of sepsis prediction models for the non-ICU setting, but many of these studies focus on the technical rather than clinical aspects, use outdated sepsis definitions not accounting for chronic organ dysfunction, or limit their evaluation to specific patient populations 19,[40][41][42][43][44][45] . In this study, we put emphasis on simulating the performance as it would be if it was implemented in a real-world setting.…”

“…As with all similar machine learning scores, the model performance is dependent on correct and accessible input data and we cannot rule out that missing variables, or differences in documentation of clinical data within the hospital, affected our results. Then again, missing data in EHR systems is generally not missing at random, but reflective of clinical decisions, and studies indicate that methods to reduce missing data in sepsis machine learning prediction models does not improve performance 45 .…”