Development and validation of an early warning tool for sepsis and decompensation in children during emergency department triage

Abstract

This study was designed to develop and validate an early warning system for sepsis based on a predictive model of critical decompensation. Data from the electronic medical records for 537,837 visits to a pediatric Emergency Department (ED) from March 2013 to December 2019 were collected. A multiclass stochastic gradient boosting model was built to identify early warning signs associated with death, severe sepsis, non-severe sepsis, and bacteremia. Model features included triage vital signs, previous diagnoses, medications, and healthcare utilizations within 6 months of the index ED visit. There were 483 patients who had severe sepsis and/or died, 1102 had non-severe sepsis, 1103 had positive bacteremia tests, and the remaining had none of the events. The most important predictors were age, heart rate, length of stay of previous hospitalizations, temperature, systolic blood pressure, and prior sepsis. The one-versus-all area under the receiver operator characteristic curve (AUROC) were 0.979 (0.967, 0.991), 0.990 (0.985, 0.995), 0.976 (0.972, 0.981), and 0.968 (0.962, 0.974) for death, severe sepsis, non-severe sepsis, and bacteremia without sepsis respectively. The multi-class macro average AUROC and area under the precision recall curve were 0.977 and 0.316 respectively. The study findings were used to develop an automated early warning decision tool for sepsis. Implementation of this model in pediatric EDs will allow sepsis-related critical decompensation to be predicted accurately after a few seconds of triage.

Haemoglobin and C-reactive protein levels are efficient at identifying neoplasms as the most common early mortality of chronic diseases in emergency department patients

“…This can be attributed to the wide range of computational methods available, enabling modern exploration of clinical and epidemiological cause-and-effect relationships. This viewpoint is supported, for example, by the Shapley additive explanation, which models the most important features for sepsis in children during ED triage [ 7 ], as well as by the multi-test evaluation of differences in the National Early Warning Score used in the ED and multi-stage combinations of logistic regression models, linear regression, and survival analysis to predict disease severity and 3-month survival in acutely dyspnoeic patients [ 8 ], or the pooled relative risk estimated by random effects of generalized least square regression models and the dose- response relationship (between CRP and risk of all cause and cause-specific mortality) modelled using restricted cubic splines in the meta-analysis of 22 screened articles (with advanced Begg’s funnel plots) [ 9 ]. Therefore, the data exploration conducted in our study using a package of less conventional statistical methods should not come as a surprise.…”

Validation of the Pediatric Sequential Organ Failure Assessment Score and Evaluation of Third International Consensus Definitions for Sepsis and Septic Shock Definitions in the Pediatric Emergency Department

“…In that vein, work by several groups including our own is ongoing to develop an ED-specific score using machine learning and artificial intelligence methods. [21][22][23] Limitations This study has several limitations. First, we were unable to trend pSOFA scores over time.…”

“…Our study does suggest that a tool derived specifically for ED use that uses variables known in the ED may be necessary. In that vein, work by several groups including our own is ongoing to develop an ED-specific score using machine learning and artificial intelligence methods …”

Identification of sepsis in paediatric emergency departments: A scoping review

“…This When only alerts that fell between 48 hours before and 12 hours after sepsis onset were analyzed, the algorithm demonstrated a sensitivity of 72% (CI, 67-77%) for an episode of severe sepsis; specificity 91.8% (CI, 91.5-92.1%); PPV 8.1% (CI, 7.0-9.2%); negative predictive value (NPV) 99.7% (CI, 99.6-99.8%); likelihood ratio 8.8 (CI, 8.1-9.5); and risk ratio 27 (CI, 21-34). In the more restrictive model examining only alerts that fell between 24 hours before and 2 hours after sepsis onset time, the algorithm had the following test characteristics: sensitivity 67% (CI, 62-72%); specificity 91.8% (CI, 91.5-92.1%); PPV 7.5% (CI, 6.5-8.5%); NPV 99.6% (CI, 99.5-99.7%); likelihood ratio 8.1 (CI, 7.4-8.8); and risk ratio 21 (CI, [16][17][18][19][20][21][22][23][24][25][26]. Also reported the variation in the PPV based on the location in hospital and severity of illness evaluated as reference standard.…”