Abstract

Background

Cells of the innate and adaptive immune systems play a critical role in the host response to sepsis. However, whether immune cell abundance of peripheral blood has potential application in sepsis patients’ prediction of 28-day survival is largely unknown. Here we aim to develop a deep learning model to predict 28-day survival in patients with sepsis.

Methods

In this study.a total of 479 sepsis patients were included and patients were randomly divided into training and validation groups in a 9:1 ratio. We built the DeepSurv in TensorFlow, a deep learning survival neural network based model on sepsis patients' data with 28 immune cells. The algorithm was internally validated and the primary end point was 28-day survival.

Results

In the training group, we established a deep learning survival neural network model showed promising results to predict 28-day survival in sepsis patients, patients with low vs high risk score had statistically significantly longer 28-day survival [hazard ratio(HR)=0.022, 95%CI=0.013-0.038, P<0.005]. The immune cell abundance risk score was associated with 28-day survival (AUCs for 7-, 14- and 28-day survival were 0.85, 0.912 an 0.936, respectively).Similarly, in the test group, patients with low vs high risk score had statistically significantly longer 28-day survival[P<0.005] and well AUCs for 14- and 28-day survival.Further, this study identified that model obviously related to immune microenvironment characteristics.

Conclusions

This study developed and validated novel deep learning survival neural network model showed reliable individual 28-day survival information in prognostic evaluation and treatment recommendation in patients with sepsis.