Risk Factors for Deterioration Events in the Pediatric Acute Care Setting.

OBJECTIVES: Rapid response system (RRS) activations resulting in emergency transfers (ETs) and codes outside the ICU are associated with increased mortality and length of stay. We aimed to evaluate the patient and care team characteristics of RRS activations resulting in ETs and codes outside the ICU (together classified as "deterioration events") versus those that did not result in a deterioration event. METHODS: For each RRS activation at our institution from 2019 to 2021, data were gathered on patient demographics and medical diagnoses, care team and treatment factors, and ICU transfer. Descriptive statistics, bivariate analyses, and multivariable logistic regression using a backward elimination model selection method were performed to assess potential risk factors for deterioration events. RESULTS: Over the 3-year period, 1765 RRS activations were identified. Fifty-three (3%) activations were deemed acute care codes, 64 (4%) were noncode ETs, 921 (52%) resulted in nonemergent transfers to an ICU, and 727 (41%) patients remained in an acute care unit. In a multivariable model, any complex chronic condition (adjusted odds ratio, 6.26; 95% confidence interval, 2.83-16.60) and hematology/oncology service (adjusted odds ratio, 2.19; 95% confidence interval, 1.28-3.74) were independent risk factors for a deterioration event. CONCLUSIONS: Patients with medical complexity and patients on the hematology/oncology service had a higher risk of deterioration events than other patients with RRS activations. Further analyzing how our hospital evaluates and treats these specific patient populations is critical as we develop targeted interventions to reduce deterioration events.

Sepsis Prediction in Hospitalized Children: Model Development and Validation.

BACKGROUND AND OBJECTIVES: Early recognition and treatment of pediatric sepsis remain mainstay approaches to improve outcomes. Although most children with sepsis are diagnosed in the emergency department, some are admitted with unrecognized sepsis or develop sepsis while hospitalized. Our objective was to develop and validate a prediction model of pediatric sepsis to improve recognition in the inpatient setting. METHODS: Patients with sepsis were identified using intention-to-treat criteria. Encounters from 2012 to 2018 were used as a derivation to train a prediction model using variables from an existing model. A 2-tier threshold was determined using a precision-recall curve: an "Alert" tier with high positive predictive value to prompt bedside evaluation and an "Aware" tier with high sensitivity to increase situational awareness. The model was prospectively validated in the electronic health record in silent mode during 2019. RESULTS: A total of 55 980 encounters and 793 (1.4%) episodes of sepsis were used for derivation and prospective validation. The final model consisted of 13 variables with an area under the curve of 0.96 (95% confidence interval 0.95-0.97) in the validation set. The Aware tier had 100% sensitivity and the Alert tier had a positive predictive value of 14% (number needed to alert of 7) in the validation set. CONCLUSIONS: We derived and prospectively validated a 2-tiered prediction model of inpatient pediatric sepsis designed to have a high sensitivity Aware threshold to enable situational awareness and a low number needed to Alert threshold to minimize false alerts. Our model was embedded in our electronic health record and implemented as clinical decision support, which is presented in a companion article.

Sepsis Prediction in Hospitalized Children: Clinical Decision Support Design and Deployment.

BACKGROUND: Following development and validation of a sepsis prediction model described in a companion article, we aimed to use quality improvement and safety methodology to guide the design and deployment of clinical decision support (CDS) tools and clinician workflows to improve pediatric sepsis recognition in the inpatient setting. METHODS: CDS tools and sepsis huddle workflows were created to implement an electronic health record-based sepsis prediction model. These were proactively analyzed and refined using simulation and safety science principles before implementation and were introduced across inpatient units during 2020-2021. Huddle compliance, alerts per non-ICU patient days, and days between sepsis-attributable emergent transfers were monitored. Rapid Plan-Do-Study-Act (PDSA) cycles based on user feedback and weekly metric data informed improvement throughout implementation. RESULTS: There were 264 sepsis alerts on 173 patients with an 89% bedside huddle completion rate and 10 alerts per 1000 non-ICU patient days per month. There was no special cause variation in the metric days between sepsis-attributable emergent transfers. CONCLUSIONS: An automated electronic health record-based sepsis prediction model, CDS tools, and sepsis huddle workflows were implemented on inpatient units with a relatively low rate of interruptive alerts and high compliance with bedside huddles. Use of CDS best practices, simulation, safety tools, and quality improvement principles led to high utilization of the sepsis screening process.