Machine-learning to predict in-hospital sepsis mortality

Time is of the essence in the treatment of sepsis; early and aggressive treatment is important to reduce mortality as indicated in recent studies. In 30-50% of patients, sepsis treatment is initiated in the emergency department (ED). Sepsis is responsible for 2% of hospitalizations and 17% of in-hospital deaths. Worldwide around 1400 patients die as a result of sepsis every day and the incidence of sepsis increases annually. Despite treatment, over 20% of the patients with sepsis deteriorate during their first 48 hours in the hospital. Deterioration can result in mortality and/or morbidity, including shock or (multiple) organ failure, when not promptly responded to. Some small studies, mainly in the ICU and general ward population, have shown that patient deterioration was preceded by changes in vital signs in over 80% of the cases, often hours before the deterioration was actually clinically noticed. However, this has not yet been investigated in the ED population of patients presenting with sepsis.

Automated evaluation of heart rate variability as predictor for deterioration and response to treatment in sepsis


The host response to sepsis can be viewed as a complex non-linear system. Such a system will naturally settle into a stable state configuration, while different variables and their interactions are seemingly chaotic and constantly changing. When physiologic parameters are measured precisely and continuously, high degrees of variability mark health, while disease is associated with reduced variability. These variability patterns are hidden within the information generally used for monitoring absolute values of vital signs. Recognizing variability patterns, might provide valuable information for the early detection and prediction of patient deterioration. Early warning and prediction of patient deterioration can help to identify patients at risk. Identifying them may provide an opportunity to anticipate on or even prevent deterioration, thereby potentially reducing mortality, morbidity and decreased quality of life. In the SepsiVit study, we aim to gain insight into the factors involved in deterioration of patients with sepsis and to create a model for early detection and prediction of patient deterioration. To achieve this aim patients will be continuously monitored during the first 48-hours in-hospital using a bedside patient monitor to record their vital signs. Using this recordings, heart rate variability (HRV) will be calculated and we will investigate the association of HRV with patient deterioration and response to treatment.


Financial support: Junior Scientific Masterclass (UMCG)