Alicia Alcamo, MD, MPH
Assistant Professor
Children's Hospital of Philadelphia
Philadelphia, Pennsylvania
Disclosure information not submitted.
Gregory Barren, MS
Information Analyst
Children's Hospital of Philadelphia Reseach Institute, United States
Disclosure information not submitted.
Andrew Becker, MD,
Pediatric Critical Care Fellow
Children's Hospital of Philadelphia
Philadelphia, Pennsylvania, United States
Disclosure information not submitted.
Jeffery Pennington, MSCS
Associate Vice President and Chief Research Informatics Officer
Children's Hospital of Philadelphia Research Institute, United States
Disclosure information not submitted.
Martha A.Q. Curley, PhD, RN, FAAN
Ruth M. Colket Endowed Chair in Pediatric Nursing
University of Pennsylvania School of Nursing
Bryn Mawr, Pennsylvania
Disclosure information not submitted.
Robert Tasker, MA, MD, MBBS, FRCP
Senior Associate Staff Physician
Boston Children's Hospital
Milton, Massachusetts
Disclosure information not submitted.
Alexis Topjian, MD, MSCE, FCCM
Associate Professor of Anesthesiology, Critical Care, and Pediatrics
Children's Hospital of Philadelphia, United States
Disclosure information not submitted.
Scott Weiss, MD, FCCM
Associate Professor of Anesthesiology, Critical Care, and Pediatrics
The Children's Hospital of Philadelphia
Glen Mills, Pennsylvania, United States
Disclosure information not submitted.
Title: A Novel Computational Phenotype to Identify Acute Brain Dysfunction in Pediatric Sepsis
Introduction: Acute brain dysfunction (ABD) is associated with increased illness severity and mortality in pediatric sepsis. ABD is defined by Glasgow Coma Scale (GCS), but GCS cannot detect the full spectrum of neurologic changes in children. We hypothesized that a computational phenotype using variables indicative of clinical concern for neurologic change would be a valid measure of ABD in sepsis.
Methods: We performed a retrospective study of 4,288 index sepsis episodes from 2011-2019. A computational phenotype, previously developed in a pediatric transplant cohort, was iterated for optimization in sepsis. Phenotype variables include completion of brain CT or MRI, EEG, or new antipsychotic prescription. We tested criterion validity (sensitivity, specificity, PPV, NPV) of the phenotype against a reference standard established from chart review of 569 random sepsis episodes by 2 raters. Reference standard for ABD was defined as any new neurologic change clinically noted within 30 days of sepsis. Interrater reliability (IRR) of the reference standard was assessed by overall percent agreement and kappa statistic. We tested face and cohort validity of the phenotype in the entire cohort by comparing characteristics and outcomes between those with and without ABD per the phenotype. We compared categorical data with chi-squared and continuous data with Wilcoxon rank sum tests.
Results: IRR of the reference standard was shown by 88% overall agreement (Κ 0.72). Compared to the reference standard, the final ABD phenotype had sensitivity 83% (95%CI 77,88), specificity 94% (95%CI 91,96), PPV 85% (95%CI 79,90), and NPV 93% (95%CI 90,95). In the 4,288 sepsis episodes, 34% (95%CI 33,36) had ABD per the phenotype. The phenotype showed face validity as patients with ABD had higher illness severity indicated by a greater proportion with ICU admission (81% vs 65% no ABD, p< 0.001) and multiple organ dysfunction syndrome (50% vs 25% no ABD, p< 0.001). Cohort validity supported by higher hospital mortality in patients with ABD (18% vs 3% no ABD, p< 0.001).
Conclusions: A computational phenotype based on variables indicative of clinical concern for neurologic change provided a valid measure of ABD in pediatric sepsis and can be used to study risk factors for and outcomes from ABD in large datasets without relying on GCS.