2020/06/29 by Emma Rocheteau, Píetro Lió, Rocheteau, Emma +3
Computer Science · Medicine · #Machine Learning in Healthcare #Intensive Care Unit Cognitive Disorders #Sepsis Diagnosis and Treatment
paper · pdf · doi:10.48550/arxiv.2006.16109
The pressure of ever-increasing patient demand and budget restrictions make\nhospital bed management a daily challenge for clinical staff. Most critical is\nthe efficient allocation of resource-heavy Intensive Care Unit (ICU) beds to\nthe patients who need life support. Central to solving this problem is knowing\nfor how long the current set of ICU patients are likely to stay in the unit. In\nthis work, we propose a new deep learning model based on the combination of\ntemporal convolution and pointwise (1x1) convolution, to solve the length of\nstay prediction task on the eICU critical care dataset. The model - which we\nrefer to as Temporal Pointwise Convolution (TPC) - is specifically designed to\nmitigate for common challenges with Electronic Health Records, such as\nskewness, irregular sampling and missing data. In doing so, we have achieved\nsignificant performance benefits of 18-51% (metric dependent) over the commonly\nused Long-Short Term Memory (LSTM) network, and the multi-head self-attention\nnetwork known as the Transformer.\n