Contextual Embeddings from Clinical Notes Improves Prediction of Sepsis

Abstract

Sepsis, a life-threatening organ dysfunction, is a clinical syndrome triggered by acute infection and affects over 1 million Americans every year. Untreated sepsis can progress to septic shock and organ failure, making sepsis one of the leading causes of morbidity and mortality in hospitals. Early detection of sepsis and timely antibiotics administration is known to save lives. In this work, we design a sepsis prediction algorithm based on data from electronic health records (EHR) using a deep learning approach. While most existing EHR-based sepsis prediction models utilize structured data including vitals, labs, and clinical information, we show that incorporation of features based on clinical texts, using a pre-trained neural language representation model, allows for incorporation of unstructured data without an explicit need for ontology-based named-entity recognition and classification. The proposed model is trained on a large critical care database of over 40,000 patients, including 2805 septic patients, and is compared against competing baseline models. In comparison to a baseline model based on structured data alone, incorporation of clinical texts improved AUC from 0.81 to 0.84. Our findings indicate that incorporation of clinical text features via a pre-trained language representation model can improve early prediction of sepsis and reduce false alarms.

Sepsis prediction, early detection, and identification using clinical text for machine learning: a systematic review

“…Of the 9 identified articles, 2 studies aimed at identifying infection, 47 , 48 6 studies focused on early detection of sepsis, 51 , 53 , 55 severe sepsis, 49 or septic shock, 50 , 54 and 1 study considered both identification and early detection for a combination of sepsis, severe sepsis, and septic shock. 52 Most studies focused on intensive care unit (ICU) 48 , 50 , 52–55 or emergency department (ED) 47 , 51 data; only 1 used inpatient care data.…”

“…49 Four studies utilized data from hospitals, 47 , 49 , 51 , 52 1 utilized MIMIC-II 54 and 4 utilized MIMIC-III. 48 , 50 , 53 , 55 MIMIC-II and MIMIC-III are publicly available ICU datasets created from Boston’s Beth Israel Deaconess Medical Center; MIMIC-II contains data from 2001–2007 76 and MIMIC-III contains data from 2001–2012. 77 Eight studies used data from the United States 47–51 , 53–55 and 1 study used data from Singapore.…”

“… 48 , 50 , 53 , 55 MIMIC-II and MIMIC-III are publicly available ICU datasets created from Boston’s Beth Israel Deaconess Medical Center; MIMIC-II contains data from 2001–2007 76 and MIMIC-III contains data from 2001–2012. 77 Eight studies used data from the United States 47–51 , 53–55 and 1 study used data from Singapore. 52 Sample sizes varied greatly in terms of the number of patients or notes used.…”

“… 52 Sample sizes varied greatly in terms of the number of patients or notes used. To select patient cohorts or notes associated with sepsis, 3 studies used International Statistical Classification of Diseases and Related Health Problems (ICD) codes, 47 , 49 , 52 5 applied sepsis definition criteria, 49–51 , 53 , 55 1 utilized descriptions of antibiotics usage, 48 and another 54 applied criteria from Henry et al 78 that include ICD codes, sepsis criteria, and notes mentioning sepsis or septic shock. Table 1 summarizes the study characteristics and additional details are in Supplementary Table S2 (for Culliton et al, 49 the 8 structured variables for the Modified Baystate clinical definition of severe sepsis and 29 structured variables used in models were provided through personal communications with the corresponding author of Culliton et al, 49 Steve Gallant, on June 4, 2021).…”

“…Hence, a search on May 15, 2021, including those terms, found 2 additional articles. 53 , 54 To ensure inclusion of other relevant articles, a broader search was conducted on September 3, 2021 to include the following terms: “unstructured data,” “unstructured text,” “clinical note,” “clinical notes,” “clinical text,” “free-text,” “free text,” “record text,” “narrative,” or “narratives.” This resulted in 1 additional article. 55…”