An Interpretable Machine Learning Model for Accurate Prediction of Sepsis in the ICU

Nemati, Shamim; Holder, Andre; Razmi, Fereshteh; Stanley, Matthew D.; Clifford, Gari D.; Buchman, Timothy G.

doi:10.1097/ccm.0000000000002936

Cited by 545 publications

(415 citation statements)

References 23 publications

Supporting

Mentioning

406

Contrasting

Unclassified

Order By: Relevance

“…The World Health Organization estimates that more than six million people die of sepsis annually, and many of these deaths are preventable [2]. In the United States, severe sepsis Previous studies have shown that machine learning (ML) models trained from data in individual patient electronic health records (EHR) may be used for the early detection of sepsis [7,8,9,10,11,12]. The ML models for sepsis detection far exceed the predictive ability of existing clinical early warning system scores, such as the National Early Warning Score (NEWS) [7,9,10,12,13,14].…”

Section: Introductionmentioning

confidence: 99%

Early detection of sepsis utilizing deep learning on electronic health record event sequences

Lauritsen

Kalør²,

Kongsgaard³

et al. 2020

Artificial Intelligence in Medicine

108

View full text Add to dashboard Cite

Background: The timeliness of detection of a sepsis incidence in progress is a crucial factor in the outcome for the patient. Machine learning models built from data in electronic health records can be used as an effective tool for improving this timeliness, but so far the potential for clinical implementations has been largely limited to studies in intensive care units. This study will employ a richer data set that will expand the applicability of these models beyond intensive care units. Furthermore, we will circumvent several important limitations that have been found in the literature: 1) Models are evaluated shortly before sepsis onset without considering interventions already initiated. 2) Machine learning models are built on a restricted set of clinical parameters, which are not necessarily measured in all departments. 3) Model performance is limited by current knowledge of sepsis, as feature interactions and time dependencies are hard-coded into the model. Methods: In this study, we present a model to overcome these shortcomings using a deep learning approach on a diverse multicenter data set. We used retrospective data from multiple Danish hospitals over a seven-year period. Our sepsis detection system is constructed as a combination of a convolutional neural network and a long short-term memory network. We suggest a retrospective assessment of interventions by looking at intravenous antibiotics and blood cultures preceding the prediction time. Results: Results show performance ranging from AUROC 0.856 (3 hours before sepsis onset) to AUROC 0.756 (24 hours before sepsis onset). Evaluating the clinical utility of the model, we find that a large proportion of septic patients did not receive antibiotic treatment or blood culture at the time of the sepsis prediction, and the model could therefore facilitate such interventions at an earlier point in time. Conclusion: We present a deep learning system for early detection of sepsis that is able to learn characteristics of the key factors and interactions from the raw event sequence data itself, without relying on a labor-intensive feature extraction work. Our system outperforms baseline models, such as gradient boosting, which rely on specific data elements and therefore suffer from many missing values in our dataset.

show abstract

Section: Introductionmentioning

confidence: 99%

Early detection of sepsis utilizing deep learning on electronic health record event sequences

Lauritsen

Kalør²,

Kongsgaard³

et al. 2020

Artificial Intelligence in Medicine

108

View full text Add to dashboard Cite

show abstract

“…Their model hit its highest prediction performance level when predicting sepsis 4 h prior to its occurrence (AUC 0.85). 15 Machine learning models have also been used to predict common post-surgical complications. 16 The present study achieves moderate sensitivity and specificity (74.19% and 84.61%, respectively) in identifying patients at risk of developing sepsis.…”

Section: Discussionmentioning

confidence: 99%

Artificial intelligence in critical care: prediction of sepsis in patients in intensive care from first initial laboratory parameters

Mitra¹,

Sangwan²

2020

Int J Res Med Sci

View full text Add to dashboard Cite

Background: Sepsis is a leading cause of morbidity and mortality in the critical care setting. The analysis of hemostatic parameters at admission have been proven to be a predictive marker for development of sepsis in the ICU. The present study aims to develop a machine learning model which can predict the development of sepsis after 72 hours of ICU admission, from initial assessment of hemostatic parameters. Methods: A total of 170 ICU admissions over six months (May 2018 -Dec 2018) period were included in the study. Hemostatic parameters including platelet counts, prothrombin time and Sonoclot assay were assayed at time of admission. The patients were followed up for development of sepsis. The data was split in two sets: training (100) and test (70). A machine learning model was developed using the linear discriminant analysis (LDA) model, in the R programming environment. The statistical parameters employed were sensitivity, specificity, positive and negative predictive value. Results: A comparison of incidence of development of clinical sepsis and predicted sepsis by the model showed 74.19% sensitivity and 84.61% specificity over the testing set. 06 false positives and 08 false negative predictions were encountered. Conclusions: The model shows potential to be used as a predictive tool for development of sepsis in the critical care ward. Moderate sensitivity and good specificity were achieved by the model, highlighting the role of hematologic assessment at admission in prediction of development of sepsis. However, further studies with larger datasets are required before implementation in clinical practice.

show abstract

“…Recently many studies have used new definitions of sepsis-3. Nemati [10] demonstrated an interpretable machine learning for predicting sepsis onset 4-12 hour prior to clinical diagnosis. Multiscale blood pressure and heart rate dynamic feature extraction and Elastic Net logistic model were used to predict sepsis 4 hours prior to its onset by Shashikumar [11], Roman Z Wang [12]compared three models (LR/SVM/LMT) by extracting a random time window 48 to 6 hours prior to the onset of sepsis .…”

Section: Introductionmentioning

confidence: 99%

An Ensemble Machine Learning Model for the Early Detection of Sepsis from Clinical Data

Fu¹,

Yuan²,

Lu³

et al. 2019

2019 Computing in Cardiology Conference (CinC)

View full text Add to dashboard Cite

Sepsis is a life-threatening disease with high mortality and expensive cost of treatment. In order to improve the outcomes of patients, it is important to detect atrisk patients with sepsis at an early stage. The Phys-ioNet/Computing in Cardiology Challenge 2019 focused on improving predicting sepsis six hours before the clinical diagnosis by using the latest definition of Sepsis-3. A total of 40,336 ICU patients were provided as public training data, A hidden test dataset was used to evaluate. An ensemble model, which combined boosting and bagging tree models (lightgbm, xgboost and random forest ) were designed to predict sepsis based on the records of the patient's hourly data. We compared the ensemble model and each single model of evaluation metrics results on selected inner test data Offline, the best performance was achieved AUC of 0.792, ACC of 0.727. Finally, the proposed model was evaluated on the full test sets received an official utility score, defined by the organizers, was 0.087, ranked 75/105 (our team name: cinc sepsis pass). While the single model of lightgbm only received a utility score of -0.036. The ensemble model utilized the preprocessing data and achieved better performance than a single tree-based model.

show abstract

An Interpretable Machine Learning Model for Accurate Prediction of Sepsis in the ICU

Cited by 545 publications

References 23 publications

Early detection of sepsis utilizing deep learning on electronic health record event sequences

Early detection of sepsis utilizing deep learning on electronic health record event sequences

Artificial intelligence in critical care: prediction of sepsis in patients in intensive care from first initial laboratory parameters

An Ensemble Machine Learning Model for the Early Detection of Sepsis from Clinical Data

Contact Info

Product

Resources

About