Early Prediction of Sepsis Considering Early Warning Scoring Systems

TY - GEN

T1 - Early Prediction of Sepsis Considering Early Warning Scoring Systems

AU - Biglarbeigi, P.

AU - Kennedy, A.

AU - McLaughlin, J.

AU - McLaughlin, D.

AU - Rjoob, K.

AU - Abdullah, Abdullah

AU - McCallan, N.

AU - Jasinska-Piadlo, A.

AU - Bond, R.

AU - Finlay, D.

AU - Ng, K. Y.

N1 - Funding Information: This project is supported by the European Union’s INTER-REG VA Programme, managed by the Special EU Programmes Body (SEUPB).

PY - 2020/2/24

Y1 - 2020/2/24

N2 - Sepsis is a noted cause of mortality in hospitalised patients, particularly patients in the ICU. Early prediction of sepsis facilitates a better targeted therapy which in turn reduces patient mortality rates. This study developed a methodology to allow automatic prediction of sepsis 6 hours prior to its clinical presentation. For this purpose, four vital signs comprising of HR, SBP, Temperature and respiratory rate, along with laboratory results for Platelets, WBC, Glucose and Creatinine are scored using Prehospital Early Sepsis Detection (PRESEP) and Sequential Organ Failure Assessment (SOFA) Early Warning Scoring (EWS) systems or screening tools and Systemic Inflammatory Response Syndrome (SIRS) criteria to allow under-sampling. The weighted scores obtained from the screening tools are also used to categorise patients into 4 groups with different probabilities of facing sepsis in ICU. The hourly data of each group is then trained through a KNN classifier to detect sepsis hours. The ensemble of classifiers are used to predict sepsis in all available dataset. The proposed model developed by UlsterTeam is trained on training setA and evaluated on training setB. The evaluation of the model on the training setB of the publically available dataset shows the Utility Score, accuracy, AUROC and AUPRC of the model are 0.27, 0.97, 0.71 and 0. 07 respectively.

AB - Sepsis is a noted cause of mortality in hospitalised patients, particularly patients in the ICU. Early prediction of sepsis facilitates a better targeted therapy which in turn reduces patient mortality rates. This study developed a methodology to allow automatic prediction of sepsis 6 hours prior to its clinical presentation. For this purpose, four vital signs comprising of HR, SBP, Temperature and respiratory rate, along with laboratory results for Platelets, WBC, Glucose and Creatinine are scored using Prehospital Early Sepsis Detection (PRESEP) and Sequential Organ Failure Assessment (SOFA) Early Warning Scoring (EWS) systems or screening tools and Systemic Inflammatory Response Syndrome (SIRS) criteria to allow under-sampling. The weighted scores obtained from the screening tools are also used to categorise patients into 4 groups with different probabilities of facing sepsis in ICU. The hourly data of each group is then trained through a KNN classifier to detect sepsis hours. The ensemble of classifiers are used to predict sepsis in all available dataset. The proposed model developed by UlsterTeam is trained on training setA and evaluated on training setB. The evaluation of the model on the training setB of the publically available dataset shows the Utility Score, accuracy, AUROC and AUPRC of the model are 0.27, 0.97, 0.71 and 0. 07 respectively.

KW - Tools

KW - Training

KW - Sugar

KW - Blood platelets

KW - Input variables

KW - Cardiology

UR - https://www.scopus.com/pages/publications/85081116187

U2 - 10.22489/CinC.2019.051

DO - 10.22489/CinC.2019.051

M3 - Conference contribution

SN - 9781728159423

VL - 46

BT - 2019 Computing in Cardiology (CinC)

PB - IEEE

T2 - CinC 2019

Y2 - 8 September 2019 through 11 September 2019

ER -