The Pediatric Data Science and Analytics Subgroup of the Pediatric Acute Lung Injury and Sepsis Investigators Network: Use of Supervised Machine Learning Applications in Pediatric Critical Care Medicine Research.

OBJECTIVE: Perform a scoping review of supervised machine learning in pediatric critical care to identify published applications, methodologies, and implementation frequency to inform best practices for the development, validation, and reporting of predictive models in pediatric critical care. DESIGN: Scoping review and expert opinion. SETTING: We queried CINAHL Plus with Full Text (EBSCO), Cochrane Library (Wiley), Embase (Elsevier), Ovid Medline, and PubMed for articles published between 2000 and 2022 related to machine learning concepts and pediatric critical illness. Articles were excluded if the majority of patients were adults or neonates, if unsupervised machine learning was the primary methodology, or if information related to the development, validation, and/or implementation of the model was not reported. Article selection and data extraction were performed using dual review in the Covidence tool, with discrepancies resolved by consensus. SUBJECTS: Articles reporting on the development, validation, or implementation of supervised machine learning models in the field of pediatric critical care medicine. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: Of 5075 identified studies, 141 articles were included. Studies were primarily (57%) performed at a single site. The majority took place in the United States (70%). Most were retrospective observational cohort studies. More than three-quarters of the articles were published between 2018 and 2022. The most common algorithms included logistic regression and random forest. Predicted events were most commonly death, transfer to ICU, and sepsis. Only 14% of articles reported external validation, and only a single model was implemented at publication. Reporting of validation methods, performance assessments, and implementation varied widely. Follow-up with authors suggests that implementation remains uncommon after model publication. CONCLUSIONS: Publication of supervised machine learning models to address clinical challenges in pediatric critical care medicine has increased dramatically in the last 5 years. While these approaches have the potential to benefit children with critical illness, the literature demonstrates incomplete reporting, absence of external validation, and infrequent clinical implementation.

Transcriptomic profiles of multiple organ dysfunction syndrome phenotypes in pediatric critical influenza.

Background: Influenza virus is responsible for a large global burden of disease, especially in children. Multiple Organ Dysfunction Syndrome (MODS) is a life-threatening and fatal complication of severe influenza infection. Methods: We measured RNA expression of 469 biologically plausible candidate genes in children admitted to North American pediatric intensive care units with severe influenza virus infection with and without MODS. Whole blood samples from 191 influenza-infected children (median age 6.4 years, IQR: 2.2, 11) were collected a median of 27 hours following admission; for 45 children a second blood sample was collected approximately seven days later. Extracted RNA was hybridized to NanoString mRNA probes, counts normalized, and analyzed using linear models controlling for age and bacterial co-infections (FDR q<0.05). Results: Comparing pediatric samples collected near admission, children with Prolonged MODS for ≥7 days (n=38; 9 deaths) had significant upregulation of nine mRNA transcripts associated with neutrophil degranulation (RETN, TCN1, OLFM4, MMP8, LCN2, BPI, LTF, S100A12, GUSB) compared to those who recovered more rapidly from MODS (n=27). These neutrophil transcripts present in early samples predicted Prolonged MODS or death when compared to patients who recovered, however in paired longitudinal samples, they were not differentially expressed over time. Instead, five genes involved in protein metabolism and/or adaptive immunity signaling pathways (RPL3, MRPL3, HLA-DMB, EEF1G, CD8A) were associated with MODS recovery within a week. Conclusion: Thus, early increased expression of neutrophil degranulation genes indicated worse clinical outcomes in children with influenza infection, consistent with reports in adult cohorts with influenza, sepsis, and acute respiratory distress syndrome.

DDX58 Is Associated With Susceptibility to Severe Influenza Virus Infection in Children and Adolescents.

BACKGROUND: Seasonal influenza virus infection causes a range of disease severity, including lower respiratory tract infection with respiratory failure. We evaluated the association of common variants in interferon (IFN) regulatory genes with susceptibility to critical influenza infection in children. METHODS: We performed targeted sequencing of 69 influenza-associated candidate genes in 348 children from 24 US centers admitted to the intensive care unit with influenza infection and lacking risk factors for severe influenza infection (PICFlu cohort, 59.4% male). As controls, whole genome sequencing from 675 children with asthma (CAMP cohort, 62.5% male) was compared. We assessed functional relevance using PICFlu whole blood gene expression levels for the gene and calculated IFN gene signature score. RESULTS: Common variants in DDX58, encoding the retinoic acid-inducible gene I (RIG-I) receptor, demonstrated association above or around the Bonferroni-corrected threshold (synonymous variant rs3205166; intronic variant rs4487862). The intronic single-nucleotide polymorphism rs4487862 minor allele was associated with decreased DDX58 expression and IFN signature (P < .05 and P = .0009, respectively) which provided evidence supporting the genetic variants' impact on RIG-I and IFN immunity. CONCLUSIONS: We provide evidence associating common gene variants in DDX58 with susceptibility to severe influenza infection in children. RIG-I may be essential for preventing life-threatening influenza-associated disease.

Host Respiratory Transcriptome Signature Associated with Poor Outcome in Children with Influenza-Staphylococcus aureus Pneumonia.

Respiratory coinfection of influenza with Staphylococcus aureus often causes severe disease; methicillin-resistant S. aureus (MRSA) coinfection is frequently fatal. Understanding disease pathogenesis may inform therapies. We aimed to identify host and pathogen transcriptomic (messenger RNA) signatures from the respiratory compartment of pediatric patients critically ill with influenza-S. aureus coinfection (ISAC), signatures that predict worse outcomes. Messenger RNA extracted from endotracheal aspirate samples was evaluated for S. aureus and host transcriptomic biosignatures. Influenza-MRSA outcomes were worse, but of 190 S. aureus virulence-associated genes, 6 were differentially expressed between MRSA-coinfected versus methicillin-susceptible S. aureus-coinfected patients, and none discriminated outcome. Host gene expression in patients with ISAC was compared with that in patients with influenza infection alone. Patients with poor clinical outcomes (death or prolonged multiorgan dysfunction) had relatively reduced expression of interferons and down-regulation of interferon γ-induced immune cell chemoattractants CXCL10 and CXCL11. In ISAC, airway host but not pathogen gene expression profiles predicted worse clinical outcomes.

Change in Frequency of Invasive and Noninvasive Respiratory Support in Critically Ill Pediatric Subjects.

BACKGROUND: Noninvasive respiratory support has become more popular in the pediatric population and may prevent or replace invasive procedures, such as endotracheal intubation, in certain circumstances. The objective was to examine the frequency of invasive and noninvasive respiratory support from 2009 to 2017 in critically ill pediatric patients and to determine patient-related factors associated with invasive support using the Virtual Pediatric Systems, LLC database. METHODS: This was an analysis of prospectively collected data on admissions with respiratory support from 17 pediatric ICUs from 2009 to 2017 reported within the Virtual Pediatric Systems database. We determined the frequency of invasive and noninvasive respiratory support over the study period by measuring the number of admissions with either invasive or noninvasive support within a given year divided by the total number of pediatric ICU admissions with respiratory support during the same year. Factors associated with invasive support were examined in univariate and multivariate regressions. RESULTS: A total of 69,262 cases of respiratory support were included. There was a decrease in the rate of invasive support over the study period from 66.9% to 48.5% (P value for test of trend < .001) and an increase in the rate of noninvasive support from 28.7% to 57.7% (P value for test of trend < .001). Trauma cases and subjects < 1 month old were more likely to receive invasive support. Cases occurring in later years and subjects with Black or Hispanic race were less likely to receive invasive support. CONCLUSIONS: From 2009 to 2017, the frequency of admissions with invasive respiratory support decreased, and those with noninvasive respiratory support increased. By 2017, the frequency of noninvasive respiratory support was greater than that of invasive respiratory support.

Pediatric Cardiac Intensive Care Distribution, Service Delivery, and Staffing in the United States in 2018.

OBJECTIVES: To assess the distribution, service delivery, and staffing of pediatric cardiac intensive care in the United States. DESIGN: Based on a 2016 national PICU survey, and verified through online searching and clinician networking, medical centers were identified with a separate cardiac ICU or mixed ICU. These centers were sent a structured web-based survey up to four times, with follow-up by mail and phone for nonresponders. SETTING: Cardiac ICUs were defined as specialized units, specifically for the treatment of children with life-threatening primary cardiac conditions. Mixed ICUs were defined as separate units, specifically for the treatment of children with life-threatening conditions, including primary cardiac disease. PARTICIPANTS: Cardiac ICU or mixed ICU physician medical directors or designees. MEASUREMENTS AND MAIN RESULTS: One-hundred twenty ICUs were identified: 61 (51%) were mixed ICUs and 59 (49%) were cardiac ICUs. Seventy five percent of institutions at least sometimes used a neonatal ICU prior to surgery. The most common temporary cardiac support beyond extracorporeal membrane oxygenation was a centrifugal pump such as Centrimag. Durable cardiac support devices were far more common in separate cardiac ICUs (84% vs 20%; p < 0.0001). Significantly less availability of electrophysiology, heart failure, and cardiac anesthesia consultation was available in mixed ICUs (p = 0.0003, p < 0.0001, p = 0.042 respectively). ICU attending physicians were in-house day and night 98% of the time in mixed ICUs and 87% of the time in cardiac ICUs. Nurse practitioners were consistent front-line providers in the ICUs caring for children with primary cardiac disease staffing 88% of cardiac ICUs and 56% of mixed ICUs. Mixed ICUs were more commonly staffed with pediatric residents, and critical care fellows were found in more cardiac ICUs (83% vs 77%; p < 0.0001). CONCLUSIONS: Mixed ICUs and cardiac ICUs have statistically different staffing models and available services. More evaluation is needed to understand how this may impact patient outcomes and training programs of physicians and nurses.

RIG-I and TLR4 responses and adverse outcomes in pediatric influenza-related critical illness.

BACKGROUND: Decreased TNF-α production in whole blood after ex vivo LPS stimulation indicates suppression of the Toll-like receptor (TLR)4 pathway. This is associated with increased mortality in pediatric influenza critical illness. Whether antiviral immune signaling pathways are also suppressed in these patients is unclear. OBJECTIVES: We sought to evaluate suppression of the TLR4 and the antiviral retinoic acid-inducible gene-I (RIG-I) pathways with clinical outcomes in children with severe influenza infection. METHODS: In this 24-center, prospective, observational cohort study of children with confirmed influenza infection, blood was collected within 72 hours of intensive care unit admission. Ex vivo whole blood stimulations were performed with matched controls using the viral ligand polyinosinic-polycytidylic acid-low-molecular-weight/LyoVec and LPS to evaluate IFN-α and TNF-α production capacities (RIG-I and TLR4 pathways, respectively). RESULTS: Suppression of either IFN-α or TNF-α production capacity was associated with longer duration of mechanical ventilation and hospitalization, and increased organ dysfunction. Children with suppression of both RIG-I and TLR4 pathways (n = 33 of 103 [32%]) were more likely to have prolonged (≥7 days) multiple-organ dysfunction syndrome (30.3% vs 8.6%; P = .004) or prolonged hypoxemic respiratory failure (39.4% vs 11.4%; P = .001) compared with those with single- or no pathway suppression. CONCLUSIONS: Suppression of both RIG-I and TLR4 signaling pathways, essential for respective antiviral and antibacterial responses, is common in previously immunocompetent children with influenza-related critical illness and is associated with bacterial coinfection and adverse outcomes. Prospective testing of both pathways may aid in risk-stratification and in immune monitoring.

Risk Factors for Mortality and Pre-ICU Fluid Balance Among Critically Ill Hematopoietic Stem Cell Transplant Patients.

OBJECTIVE: To examine if fluid balance surrounding pediatric intensive care unit (PICU) admission in hematopoietic stem cell transplant (HSCT) patients was associated with mortality, ventilator-free days, and intensive care unit (ICU)-free days. To explore other population-specific factors associated with poor outcome. MATERIALS AND METHODS: Retrospective review of HSCT patients admitted to 2 quaternary PICUs, Children's Hospital Los Angeles and University of California San Francisco Benioff Children's Hospital from January 2009 to December 2014. RESULTS: Of 144 patients, 92 were identified with complete fluid balance data available. No difference in fluid balance between survivors and nonsurvivors in the 24 hours preceding PICU admission (P = .81) or when the first 24 hours of PICU stay were taken into account (P = .48) was identified. There was no difference in ventilator-free or ICU-free days. Comparing Pediatric Index of Mortality (PIM)-2, Pediatric Risk of Mortality (PRISM)-3, and a multivariable model using independent risk factors identified through multivariable analysis, the receiver operating characteristic plot for the multivariable model (area under the curve = 0.844 [95% confidence interval: 0.77-0.92]) was superior to both PIM-2 and PRISM-3 in discriminating mortality. CONCLUSIONS: Fluid balance immediately preceding and early in the course of admission was not associated with mortality in PICU HSCT patients. A subset of variables was identified which better discriminated mortality in this cohort than accepted PICU severity of illness scores.

Safety, Efficacy, and Timeliness of Intravenous Potassium Chloride Replacement Protocols in a Pediatric Cardiothoracic Intensive Care Unit.

OBJECTIVE: Hypokalemia in children following cardiac surgery occurs frequently, placing them at risk of life-threatening arrhythmias. However, renal insufficiency after cardiopulmonary bypass warrants careful administration of potassium (K+). Two different nurse-driven protocols (high dose and tiered dosing) were implemented to identify an optimal K+ replacement regimen, compared to an historical low-dose protocol. Our objective was to evaluate the safety, efficacy, and timeliness of these protocols. DESIGN: A retrospective cohort review of pediatric patients placed on intravenous K+ replacement protocols over 1 year was used to determine efficacy and safety of the protocols. A prospective single-blinded review of K+ repletion was used to determine timeliness. PATIENTS: Pediatric patients with congenital or acquired cardiac disease. SETTING: Twenty-four-bed cardiothoracic intensive care unit in a tertiary children's hospital. INTERVENTIONS: Efficacy was defined as fewer supplemental potassium chloride (KCl) doses, as well as a higher protocol to total doses ratio per patient. Safety was defined as a lower percentage of serum K+ levels ≥4.8 mEq/L after a dose of KCl. Between-group differences were assessed by nonparametric univariate analysis. RESULTS: There were 138 patients with a median age of 3.0 (interquartile range: 0.23-10.0) months. The incidence of K+ levels ≥4.8 mEq/L after a protocol dose was higher in the high-dose protocol versus the tiered-dosing protocol but not different between the low-dose and tiered-dosing protocols (high dose = 2.2% vs tiered dosing = 0.5%, P = .05). The ratio of protocol doses to total doses per patient was lower in the low-dose protocol compared to the tiered-dosing protocol (P < .05). Protocol doses were administered 45 minutes faster (P < .001). CONCLUSION: The tiered-dosed, nurse-driven K+ replacement protocol was associated with decreased supplemental K+ doses without increased risk of hyperkalemia, administering doses faster than individually ordered doses; the protocol was effective, safe, and timely in the treatment of hypokalemia in pediatric patients after cardiac surgery.

Growth and Changing Characteristics of Pediatric Intensive Care 2001-2016.

OBJECTIVES: We assessed the growth, distribution, and characteristics of pediatric intensive care in 2016. DESIGN: Hospitals with PICUs were identified from prior surveys, databases, online searching, and clinician networking. A structured web-based survey was distributed in 2016 and compared with responses in a 2001 survey. SETTING: PICUs were defined as a separate unit, specifically for the treatment of children with life-threatening conditions. PICU hospitals contained greater than or equal to 1 PICU. SUBJECTS: Physician medical directors and nurse managers. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: PICU beds per pediatric population (< 18 yr), PICU bed distribution by state and region, and PICU characteristics and their relationship with PICU beds were measured. Between 2001 and 2016, the U.S. pediatric population grew 1.9% to greater than 73.6 million children, and PICU hospitals decreased 0.9% from 347 to 344 (58 closed, 55 opened). In contrast, PICU bed numbers increased 43% (4,135 to 5,908 beds); the median PICU beds per PICU hospital rose from 9 to 12 (interquartile range 8, 20 beds). PICU hospitals with greater than or equal to 15 beds in 2001 had significant bed growth by 2016, whereas PICU hospitals with less than 15 beds experienced little average growth. In 2016, there were eight PICU beds per 100,000 U.S. children (5.7 in 2001), with U.S. census region differences in bed availability (6.8 to 8.8 beds/100,000 children). Sixty-three PICU hospitals (18%) accounted for 47% of PICU beds. Specialized PICUs were available in 59 hospitals (17.2%), 48 were cardiac (129% growth). Academic affiliation, extracorporeal membrane oxygenation availability, and 24-hour in-hospital intensivist staffing increased with PICU beds per hospital. CONCLUSIONS: U.S. PICU bed growth exceeded pediatric population growth over 15 years with a relatively small percentage of PICU hospitals containing almost half of all PICU beds. PICU bed availability is variable across U.S. states and regions, potentially influencing access to care and emergency preparedness.

The Top Ten Websites in Critical Care Medicine Education Today.

INTRODUCTION:: The number of websites for the critical care provider is rapidly growing, including websites that are part of the Free Open Access Med(ical ed)ucation (FOAM) movement. With this rapidly expanding number of websites, critical appraisal is needed to identify quality websites. The last major review of critical care websites was published in 2011, and thus a new review of the websites relevant to the critical care clinician is necessary. METHODS:: A new assessment tool for evaluating critical care medicine education websites, the Critical Care Medical Education Website Quality Evaluation Tool (CCMEWQET), was modified from existing tools. A PubMed and Startpage search from 2007 to 2017 was conducted to identify websites relevant to critical care medicine education. These websites were scored based on the CCMEWQET. RESULTS:: Ninety-seven websites relevant for critical care medicine education were identified and scored, and the top ten websites were described in detail. Common types of resources available on these websites included blog posts, podcasts, videos, online journal clubs, and interactive components such as quizzes. Almost one quarter of websites (n = 22) classified themselves as FOAM websites. The top ten websites most often included an editorial process, high-quality and appropriately attributed graphics and multimedia, scored much higher for comprehensiveness and ease of access, and included opportunities for interactive learning. CONCLUSION:: Many excellent online resources for critical care medicine education currently exist, and the number is likely to continue to increase. Opportunities for improvement in many websites include more active engagement of learners, upgrading navigation abilities, incorporating an editorial process, and providing appropriate attribution for graphics and media.

Vancomycin Monotherapy May Be Insufficient to Treat Methicillin-resistant Staphylococcus aureus Coinfection in Children With Influenza-related Critical Illness.

Background: Coinfection with influenza virus and methicillin-resistant Staphylococcus aureus (MRSA) causes life-threatening necrotizing pneumonia in children. Sporadic incidence precludes evaluation of antimicrobial efficacy. We assessed the clinical characteristics and outcomes of critically ill children with influenza-MRSA pneumonia and evaluated antibiotic use. Methods: We enrolled children (<18 years) with influenza infection and respiratory failure across 34 pediatric intensive care units 11/2008-5/2016. We compared baseline characteristics, clinical courses, and therapies in children with MRSA coinfection, non-MRSA bacterial coinfection, and no bacterial coinfection. Results: We enrolled 170 children (127 influenza A, 43 influenza B). Children with influenza-MRSA pneumonia (N = 30, 87% previously healthy) were older than those with non-MRSA (N = 61) or no (N = 79) bacterial coinfections. Influenza-MRSA was associated with increased leukopenia, acute lung injury, vasopressor use, extracorporeal life support, and mortality than either group (P ≤ .0001). Influenza-related mortality was 40% with MRSA compared to 4.3% without (relative risk [RR], 9.3; 95% confidence interval [CI], 3.8-22.9). Of 29/30 children with MRSA who received vancomycin within the first 24 hours of hospitalization, mortality was 12.5% (N = 2/16) if treatment also included a second anti-MRSA antibiotic compared to 69.2% (N = 9/13) with vancomycin monotherapy (RR, 5.5; 95% CI, 1.4, 21.3; P = .003). Vancomycin dosing did not influence initial trough levels; 78% were <10 µg/mL. Conclusions: Influenza-MRSA coinfection is associated with high fatality in critically ill children. These data support early addition of a second anti-MRSA antibiotic to vancomycin in suspected severe cases.

CALIPSO: A Randomized Controlled Trial of Calfactant for Acute Lung Injury in Pediatric Stem Cell and Oncology Patients.

To assess if calfactant reduces mortality among children with leukemia/lymphoma or after hematopoietic cell transplantation (HCT) with pediatric acute respiratory distress syndrome (PARDS), we conducted a multicenter, randomized, placebo-controlled, double-blinded trial in 17 pediatric intensive care units (PICUs) of tertiary care children's hospitals. Patients ages 18 months to 25 years with leukemia/lymphoma or having undergone HCT who required invasive mechanical ventilation for bilateral lung disease with an oxygenation index (OI) > 10 and <37 were studied. Interventions used were intratracheal instillation of either calfactant or air placebo (1 or 2 doses). Forty-three subjects were enrolled between November 2010 and June 2015: 26 assigned to calfactant and 17 to placebo. There were no significant differences in the primary outcome, which was survival to PICU discharge (adjusted hazard ratio of mortality for calfactant versus placebo, 1.78; 95% confidence interval, .53 to 6.05; P = .35), OI, functional outcomes, or ventilator-free days, adjusting for risk strata and Pediatric Risk of Mortality (PRISM) score. Despite the risk-stratified randomization, more allogeneic HCT patients received calfactant (76% and 39%, respectively) due to low recruitment at various sites. This imbalance is important because independent of treatment arm and while adjusting for PRISM score, those with allogeneic HCT had a nonsignificant higher likelihood of death at PICU discharge (adjusted odds ratio, 3.02; 95% confidence interval, .76 to 12.06; P = .12). Overall, 86% of the patients who survived to PICU discharge also were successfully discharged from the hospital. These data do not support the use of calfactant among this high mortality group of pediatric leukemia/lymphoma and/or HCT patients with PARDS to increase survival. In spite of poor enrollment, allogeneic HCT patients with PARDS appeared to be characterized by higher mortality than even other high-risk immunosuppressed groups. Conducting research among these children is challenging but necessary, because survival to PICU discharge usually results in successful discharge to home.

Association of 24/7 In-House Intensive Care Unit Attending Physician Coverage With Outcomes in Children Undergoing Heart Operations.

BACKGROUND: Multicenter data regarding the around-the-clock (24/7) presence of an in-house critical care attending physician with outcomes in children undergoing cardiac operations are limited. METHODS: Patients younger than 18 years of age who underwent operations (with or without cardiopulmonary bypass [CPB]) for congenital heart disease at 1 of the participating intensive care units (ICUs) in the Virtual PICU Systems (VPS, LLC) database were included (2009-2014). The study population was divided into 2 groups: the 24/7 group (14,737 patients; 32 hospitals), and the No 24/7 group (10,422 patients; 22 hospitals). Propensity-score matching was performed to match patients 1:1 in the 24/7 group and in the No 24/7 group. RESULTS: Overall, 25,159 patients from 54 hospitals qualified for inclusion. By propensity matching, 9,072 patients (4,536 patient pairs) from 51 hospitals were matched 1:1 in the 2 groups. After matching, mortality at ICU discharge was lower among the patients treated in hospitals with 24/7 coverage (24/7 versus No 24/7, 2.8% versus 4.0%; p = 0.002). The use of extracorporeal membrane oxygenation (ECMO), the incidence of cardiac arrest, extubation within 48 hours after operation, the rate of reintubation, and the duration of arterial line and central venous line use after operation were significantly improved in the 24/7 group. When stratified by surgical complexity, survival benefits of 24/7 coverage persisted among patients undergoing both high-complexity and low-complexity operations. CONCLUSIONS: The presence of 24-hour in-ICU attending physician coverage in children undergoing cardiac operations is associated with improved outcomes, including ICU mortality. It is possible that 24-hour in-ICU attending physician coverage may be a surrogate for other factors that may bias the results. Further study is warranted.