Recent Temporal Pattern Mining for Septic Shock Early Prediction

Abstract

Sepsis is a leading cause of in-hospital death over the world and septic shock, the most severe complication of sepsis, reaches a mortality rate as high as 50%. Early diagnosis and treatment can prevent most morbidity and mortality. In this work, Recent Temporal Patterns (RTPs) are used in conjunction with SVM classifier to build a robust yet interpretable model for early diagnosis of septic shock. This model is applied to two different prediction tasks: visit-level early diagnosis and event-level early prediction. For each setting, this model is compared against several strong baselines including atemporal method called Last-Value, six classic machine learning algorithms, and lastly, a state-of-the-art deep learning model: Long Short-Term Memory (LSTM). Our results suggest that RTP-based model can outperform all aforementioned baseline models for both diagnosis tasks. More importantly, the extracted interpretative RTPs can shed lights for the clinicians to discover progression behavior and latent patterns among septic shock patients.