Pediatric vs Adult or Mixed Trauma Centers in Children Admitted to Hospitals Following Trauma

This systematic review and meta-analysis evaluates whether children admitted to hospitals following trauma who receive definitive treatment in a pediatric trauma center have better outcomes than those treated in adult trauma centers.

There is currently no consensus on age cut-offs for pediatric trauma populations so the World Health Organisation definition of children and adolescents was used (https://www.who.int/health-topics/adolescent-health#tab=tab_1) 2 Studies including patients >19 years of age were included if ≤ 19-year-olds represented more than 80% of the study population.
We applied no restrictions to language or date of publication. Articles in languages other than English were translated. If information to be extracted was unavailable or unclear, we contacted the study authors for clarification. Studies reporting on diverse populations were included if data on pediatric trauma patients could be extracted. We also excluded studies presenting no data on selected outcomes and exposures, conference abstracts, unpublished studies, narrative reviews, case reports, in addition to studies reporting drowning, burns, foreign bodies, poisoning, late effects of injury, cadaver or animal studies.

Information sources
We conducted a systematic search of the Medical Literature Analysis and Retrieval System Online (MEDLINE, via PubMed), Excerpta Medica dataBASE (EMBASE), Web of Science and Cumulative Index to Nursing and Allied Health Literature (CINAHL) from inception to up to February 28 th , 2023. Thesis repositories, healthcare quality websites and references of included studies were also screened.

Search strategy
Our search strategy was designed using a combination of Boolean terms with relevant keywords and subject headings for EMBASE (EMBASE tree; EMTREE) and MEDLINE (Medical Subject Headings; MeSH), and then adapted to the remaining databases (eTable 2). Clinicians practising in pediatric trauma care and information specialists were consulted to refine the search strategy using the Peer Review of Electronic

Selection process
Pairs of reviewers (KM, LM, PAT, AA, TD) independently assessed study eligibility. Inter-reviewer agreement on eligibility was assessed using the first 500 citations. We repeated this process until acceptable inter-rater agreement was attained. Discrepancies between reviewers were resolved by consensus and a third reviewer adjudicated when necessary (LM). Citations were managed using EndNote software (version X9.3.3, New York City: Thomson Reuters, 2018). We managed duplicates via electronic and manual screening. If multiple publications based on the same dataset are identified by crosschecking authors, dates and settings, we selected only one publication for analyses based on study dates (most recent) and sample size (largest).

Data collection process
We developed a standard electronic data extraction form and a detailed instruction manual and piloted it on a representative sample of ten publications. Pairs of reviewers (KM, LM, PAT) independently extracted data including study design, setting (country), patient characteristics (age, injury mechanisms, type and severity of injuries), hospital characteristics (PTC/ATC/CTC, trauma center designation level, and verification organisation), outcome measures, adjustment variables, and measures of association and variation.

Outcomes
Our primary outcomes of interest, defined a priori in collaboration with our advisory committee, were mortality, complications, functional status, discharge destination, and quality of life (eTable 3). Secondary outcomes were resource utilisation and processes of care. For example, use of computed tomography [CT] imaging or operative management of blunt solid organ injuries. Currently no consensus on a definition of complications for pediatric injury populations 2 Does not include interventional radiology

Risk of bias assessment
Three content experts (LM, PAT, JG) independently rated risk of bias using the Risk of Bias in Nonrandomized Studies of Interventions (ROBINS-I) tool. 2 The tool considers bias due to confounding, selection of participants into the study, misclassification of interventions, deviations from intended interventions, missing data, measurement of outcomes, and selection of the reported results. We resolved disagreements via arbitration with a third reviewer (NY). Two content experts (LM, PAT) evaluated the quality of evidence using Grading of Recommendations, Assessment, Development and Evaluation (GRADE) criteria. 3 We assessed publication bias using a contour-enhanced funnel plot and estimated the magnitude of the potential bias with the trim-and-fill method. 4

Effect measures
We summarized data for dichotomous outcomes using odds ratios (ORs) along with 95% confidence intervals (CIs). Mean differences (MDs) or geometric mean ratios (GMRs) were used for continuous outcomes.

Synthesis methods
We restricted data synthesis to studies presenting risk-adjusted measures of association, i.e. measures minimally adjusted for age and injury severity; in the context of a review limited to observational studies with a very high risk of indication bias, unadjusted comparisons were considered not to produce meaningful results. When two studies or more evaluated the same exposure-outcome association, we conducted metaanalyses. We calculated pooled effect estimates and 95% confidence intervals (CIs) using random-effects models with an exact Mantel-Haenszel method, as design-related heterogeneity was present. Due to variations in sample sizes, we estimated between-study variance (τ 2 ) with the restricted maximum likelihood estimator and CIs with the Q-profile method. We applied a Hartung-Knapp correction in order to reflect uncertainty in estimating τ 2 . We presented results using forest plots. We measured the heterogeneity of included studies using the I² statistic and interpreted as low if 0 to 40%, moderate if 30 to 60%, substantial if 50 to 90% and considerable if 75 to 100%. Given that I 2 tends toward 100% as sample sizes increase, we calculated prediction intervals around the pooled effect size as recommended in the Cochrane handbook. 5 These intervals present the range within which we expect the effect sizes of future studies when taking into consideration the current evidence. All analyses were conducted using R Statistical Software

Grading evidence
Two content experts (LM, PAT) independently graded evidence by applying Grading of Recommendations Assessment, Development and Evaluation (GRADE) criteria: 3 1) risk of bias in the individual studies, 2) inconsistency, 3) indirectness, 4) imprecision, and 5) publication bias. In accordance with GRADE guidelines for grading evidence from non-randomized intervention studies, we then graded the certainty of evidence as high, moderate, low, or very low.

Pre-specified subgroup and sensitivity analyses
We conducted subgroup analyses for factors thought to modify the effectiveness of PTCs identified on consultation with our advisory committee: age, type of injury, injury severity, trauma center designation levels and verification body, country, and study period (eTable 4). To assess the impact of risk of bias on effect estimates, we repeated analyses excluding studies with a critical risk of bias. Finally, outliers were identified using a two-step analysis aiming to identify (1) studies whose upper bound of the 95% CI was lower than the lower bound of the pooled effect (extremely small effects) and (2) studies whose lower bound of the 95% CI was higher than the upper bound of the pooled effect (extremely large effects). 6 Insufficient number of studies to conduct analyses for a priori categories spinal cord injury and major orthopedic injury 3 Insifficient number of studies to conduct analyses for other categories of injury severity 4 Insufficient number of studies to look at countries other than the US individually 5 Only one study was moderate and none were low 6 This subgroup analysis was added to assess the impact of what was considered potentially the most important source of bias

Protocol deviations
We were unable to limit our review to major trauma because studies used heterogeneous criteria to define injury severity (ISS, AIS, physiological criteria, LOS, interventions). We thus used 'admission to an acute care hospital following injury' as our inclusion criteria and conducted subgroup analysis for major trauma. We had to group pre-specified subgroup categories, <8 and 9-15 years of age due to the lack of studies on children 9-15 years. We had to extend our pre-planned threshold for older adolescents from 16 to 14 because although 11 studies focussed on adolescents, only one used an age cut-off of 16 years of age or greater. Planned subgroup analyses for country/continent had to be modified to analysis of studies conducted in the USA as only 19% of included studies were conducted in other countries. Due to lack of studies or differences in measures reported, we were unable to conduct meta-analyses for the following preplanned outcomes: functional status, quality of life, discharge destination, hospital and ICU LOS, ventilator days, and costs. We were also unable to conduct meta-analyses for the contrast PTC versus nondesignated hospitals, and subgroup analyses for outcomes other than mortality and operative management.