Abstract
Importance:
Although improved outcomes for children treated at pediatric (PTC) compared to adult trauma centers (ATC) have been observed, differences in mortality have not been consistently observed for adolescents using data obtained from regional trauma systems. Because trauma is the leading cause of death and acquired disability among adolescents, it is important to better define differences in outcome among injured adolescents using national data.
Objective:
To use a national dataset to compare mortality of injured adolescents at different trauma center types: ATC, PTC, or mixed trauma centers (MTC) that treat both pediatric and adult trauma patients.
Design:
Multilevel models accounting for center-specific effects were used to evaluate the association of center characteristics (PTC, ATC or mixed [MTC]) on mortality, controlling for gender, mechanism of injury (blunt vs. penetrating), injuries sustained based on Abbreviated Injury Scale (post-dot values <3 or ≥3 by body region), initial systolic blood pressure, and Glasgow coma scale. Missing data was managed using multiple imputation, accounting for multilevel data structure.
Setting:
Level I or level II trauma centers participating in the National Trauma Data Bank
Participants:
Patients ages 15 to 19 years old who were treated for a blunt or penetrating injury were included
Main Outcome and Measure:
Mortality at each center type
Results:
Among 29,613 injured adolescents, most were treated at ATCs (68.9%), with the remainder at MTCs (25.6%) or PTCs (5.5%). Mortality was higher among adolescents treated at ATCs and MTCs than those at PTCs (3.2% and 3.5% vs. 0.4%, p<0.001). The adjusted odds of mortality was higher at ATCs (OR 4.19, 95% CI: 1.30, 13.51, p<0.001) and MTCs (OR 6.68, 95% CI: 2.03, 21.99, p=0.002) compared to PTCs.
Conclusion and Relevance:
Mortality among injured adolescents was lower in PTCs, compared to those treated at ATCs and MTCs. Defining resource and patient features that account for these observed differences is needed to optimize adolescent outcomes after injury.
INTRODUCTION
As adolescents transition into adulthood, their medical care transitions from delivery by pediatric providers to adult providers. This transition also occurs for their hospital-based treatment after injury, with more trauma adolescent patients being treated at centers primarily treating injured adults. Pediatric trauma centers (PTCs) are unique in that these centers provide care that addresses the specific physiological, anatomic, and social needs of children.1 As adolescents move into adulthood, however, the perceived need for these resources decreases. The largest proportion of patients seen at trauma centers are young adults, a population with physiological, anatomic and social needs similar to older adolescents.2 Because the resources of PTC are comparatively scarce and as they may not be needed for older adolescents, injured adolescents may also be treated either in an ATC or MTC rather than a PTC.
Trauma systems in which PTC and either ATC or MTC are found typically define a cutoff age for triage to either a pediatric or primarily adult trauma center. This cutoff age varies in different trauma systems based on local preferences and practice patterns and is usually not based on outcome differences at different center types. Previous studies examining the optimal trauma center type for children have mostly shown an outcome advantage for children treated at a PTC or MTC compared to an ATC.3–9 Although a consensus is forming for the optimal treatment center for younger children, controversy still remains about the optimal location for treating injured adolescents.10–14
Few studies have evaluated the outcome for adolescents at different trauma center types after injury.7,13,14 Studies evaluating outcome among adolescents treated at different types of trauma centers have shown conflicting results, with some showing different frequencies of interventions (e.g., laparotomy and blood transfusion) and no difference in mortality with others showing improved mortality.7,13,14 These previous studies, however, have used regional data obtained from mature trauma systems and have not addressed this issue in a larger national cohort in which more diversity in care and resource availability may exist. Although patient demographics and injury patterns have been proposed as explanations for mortality differences, these differences were not consistently observed even after controlling for patient and injury features.3,4,6,8,9,15,16 The objectives of this paper are to determine whether type of trauma center designation (pediatric, mixed or adults) is related to mortality among adolescent trauma patients in a sample obtained from the National Trauma Data Bank and to determine differences among patients presenting to different center types. A second objective was to determine the final discharge disposition of survivors at different center types.
METHODS
Data Source
This study was approved by the Institutional Review Board at Children’s National Medical Center as an exempt study. Data were derived from the 2010 National Trauma Data Bank (NTDB). The NTDB is a database maintained by the American College of Surgeons that contains records of injured children and adults treated at reporting trauma centers. Patients who were 15 to 19 years old treated at a level I or level II trauma center for a blunt or penetrating injury were included. Patients who were transferred to a short-term general stay hospital were excluded because their outcome was unknown. Of the 29,658 records obtained, 45 were missing data for the primary outcome (mortality) and were excluded for a final analysis dataset of 29,613 patient records.
Data Analysis
We performed multilevel logistic regression to identify the impact of trauma center type and trauma center level on mortality and discharge disposition among survivors. We separately evaluated overall mortality that included deaths that occurred in the emergency department and in the hospital after admission as well as in-hospital mortality. We defined discharge to rehabilitation as disposition to an intermediate, long-term or skilled nursing facility. Patient factors modeled included gender, mechanism of injury, transfer status (transported directly from scene versus seen at another hospital before arrival), initial systolic blood pressure <90 mm Hg, the initial motor component of the Glasgow coma score (modeled as five binary variables representing six values) and abbreviated injury scale (AIS) post-dot values for each of nine body regions (treated as binary variables for values <3 or ≥3 within each of nine body regions). Hospital-level variables included the type of center (PTC, ATC or MTC) and trauma center verification level (I or II).
The amount of missingness among the covariates modeled was between 0% and 4.5%. To manage missingness in the covariates, we performed multiple imputation using a procedure that accounted for the hierarchical structure of the dataset, creating 10 imputed datasets (Appendix). The 10 multiple imputation datasets were analyzed using Mplus 7.3.17 Parameter estimates were averaged across the 10 analyses, and their standard errors were computed using Rubin’s method.18,19 Multilevel logistic regression were used for model estimation with mortality or discharge to rehabilitation as the dependent variable. Standard errors were computed using a robust sandwich estimator. Models with a random intercept were used to account for the hierarchical data structure (patients nested within hospitals). Comparisons of patient and center characteristics between the three types of trauma centers were performed using either one-way ANOVA or chi-squared test as appropriate using imputed data to obtain estimates for variables with missing values (SAS version 9.3, SAS Institute Inc., Cary, NC, USA). Statistical significance was defined a p<0.05. Model diagnostics were performed measuring discrimination with area under the receiver operator curve (AUC) and measuring calibration with the Hosmer-Lemeshow statistic, dividing records in deciles of estimated probability of mortality.
RESULTS
Among the 29,613 patients, most were treated at an ATC (n=20,402, 68.9%). The remainder were treated at either an MTC (n=7,572, 25.6%) or a PTC (n=1,636, 5.5%). More adolescents were treated at a level I trauma center (63.3%) than at a level II trauma center (36.7%). Treatment at a level I center was more common among patients at a MTC compared to those at a PTC and was least common among patients treated at an ATC (Table 1). Patients treated at a PTC were less likely to arrive directly from the injury scene than transferred from another hospital compared to those treated at either an ATC or MTC. Adolescents treated at ATC and MTC were older and more severely injured than those treated at PTC, including more frequent hypotension, lower GCS motor scores, higher ISS and more frequent severe head, chest and upper extremity injuries (Table 1). Patients treated at a PTC were less commonly discharged to rehabilitation than adolescents treated at either ATC and MTC (Table 1).
Table 1.
Patient and Center Characteristics
| All (n=29,613) | PTC (n=1,639) | ATC (n=20,402) | MTC (n=7,572) | p-value | |
|---|---|---|---|---|---|
| Age (yrs, mean±SD) | 17.3 (±1.4) | 16.0 (±1.0) | 17.4 (±1.3) | 17.1 (±1.4) | <.001 |
| Male (%) | 72.7 | 75.4 | 72.7 | 71.3 | <.001 |
| Transferred (%) | 25.6 | 44.2 | 22.1 | 30.9 | <.001 |
| Initial blood pressure <90 mm Hg | 2.9 | 1.4 | 3.2 | 2.3 | <.001 |
| GCS motor<6 (%) | 10.7 | 6.4 | 11.2 | 10.5 | <.001 |
| ISS>15 (%) | 19.3 | 11.1 | 19.7 | 20.3 | <.001 |
| AIS post-dot value ≥3 (%) | |||||
| Head | 14.0 | 11.3 | 15.2 | 15.6 | <.001 |
| Face | 0.1 | 0.1 | 0.1 | 0.1 | 0.68 |
| Neck | 0.2 | 0.1 | 0.2 | 0.2 | 0.30 |
| Thorax | 14.0 | 7.9 | 17.1 | 17.2 | <.001 |
| Abdomen | 5.0 | 4.8 | 5.1 | 5.3 | 0.63 |
| Spine | 1.6 | 1.6 | 1.4 | 1.8 | 0.04 |
| Upper extremity | 1.6 | 1.0 | 2.1 | 1.8 | 0.007 |
| Lower extremity | 10.2 | 9.3 | 10.8 | 10.5 | 0.14 |
| External | 0.0 | 0.0 | 0.0 | 0.0 | 0.81 |
| Level I trauma center (%) | 63.3 | 76.1 | 55.3 | 81.9 | <.001 |
| Discharge to rehabilitation center | 5.1 | 3.3 | 5.0 | 6.0 | <.001 |
| Emergency room mortality (%) | |||||
| Arriving from scene | 1.4 | 0.3 | 1.5 | 1.4 | 0.01 |
| Transferred | 0.3 | 0.1 | 0.3 | 0.4 | 0.60 |
| Overall | 1.1 | 0.2 | 1.2 | 1.0 | <.001 |
| Overall mortality (%) | 3.1 | 0.4 | 3.2 | 3.5 | <.001 |
PTC, pediatric trauma center; ATC, adult trauma center; MTC, mixed trauma center; SD, standard deviation; GCS, Glasgow coma scale; ISS, injury severity score; AIS, Abbreviated Injury Scale
Adolescents treated at PTC were more likely to be injured by a blunt than penetrating injury mechanism compared to those treated at ATC or MTC. Unintentional injuries were most common at all center types but were less frequent among patients treated at ATC and MTC. Injury by assault was less frequent at PTC. Patients injured in a fall or when struck were most common at PTC, while patients injured as motor vehicle occupants were most common at ATC and MTC. Penetrating injury mechanisms (firearm and cut/pierce) and motorcyclist injuries were more common at ATC and MTC than at PTC (Table 2).
Table 2.
Mechanism of Injury by Center Type
| All (n=29,613) | PTC (n=1,639) | ATC (n=20,402) | MTC (n=7,572) | p-value | |
|---|---|---|---|---|---|
| Blunt injury type (%) | 82.1 | 91.4 | 80.4 | 84.6 | <.001 |
| Injury Intent (%) | |||||
| Unintentional | 77.7 | 86.5 | 75.6 | 81.4 | <.001 |
| Assault | 19.6 | 11.5 | 21.5 | 16.2 | |
| Self-inflicted | 1.5 | 1.0 | 1.3 | 1.9 | |
| Other | 0.6 | 0.9 | 0.6 | 0.5 | |
| Mechanism of Injury (%) | |||||
| Motor vehicle occupant | 32.2 | 18.5 | 32.6 | 34.3 | <.001 |
| Fall | 14.3 | 25.9 | 12.9 | 15.1 | |
| Struck | 13.3 | 26.1 | 12.3 | 13.2 | |
| Firearm | 10.7 | 4.5 | 12.0 | 8.3 | |
| Transport | 9.2 | 9.9 | 9.0 | 9.7 | |
| Cut/pierce | 7.1 | 4.1 | 7.3 | 7.1 | |
| Pedestrian | 4.7 | 3.7 | 5.0 | 4.2 | |
| Pedal cyclist | 3.9 | 4.9 | 3.8 | 3.9 | |
| Motorcyclist | 2.8 | 1.2 | 3.0 | 2.7 | |
| Other | 2.0 | 1.2 | 2.3 | 1.5 |
PTC, pediatric trauma center; ATC, adult trauma center; MTC, mixed trauma center
Unadjusted emergency department and overall mortality was higher among adolescents treated at ATC and MTC compared to those treated at PTC (Table 1). The emergency department mortality of adolescents arriving from the injury scene was higher among adolescents treated at ATC and MTC than among those treated at PTC but was similar among those transferred from another hospital at each center type (Table 1). After adjusting for patient and center level variables, overall mortality remained higher at ATC (OR 4.19, 95% CI: 1.30, 13.51, p<0.001) and MTC (OR 6.68, 95% CI: 2.03, 21.99, p=0.002) compared to PTC but was not observed to be different between level I and II centers (OR 0.76, 95% CI: 0.59, 0.99, p=0.07) (Table 3). The model derived for these results show excellent discrimination and was well calibrated, overall and among subgroups of patients (Table 4). Similar significant differences in mortality between adolescents treated at PTC and other center types was observed in the subgroup of adolescents who were admitted to the hospital (i.e., did not die in emergency department) and in the subgroup of adolescents sustaining only penetrating injuries (Tables 3 and 5). After adjusting for patient and center variables, no difference in disposition to rehabilitation was observed among the three center types (Table 3).
Table 3.
Odds of Death and Discharge to Rehabilitation by Patient and Center Characteristics
| Variable | OR of Death (95% CI) | OR of In-Hospital Death (95% CI) | OR of Discharge to Rehabilitation (95% CI) |
|---|---|---|---|
| ATC‡ | 4.19 (1.30, 13.51) | 6.61 (2.51, 17.36) | 1.31 (0.64, 2.66) |
| MTC‡ | 6.68 (2.03, 21.99) | 10.34 (3.82, 27.99) | 1.51 (0.72, 3.20) |
| Level II | 0.76 (0.59, 0.99) | 0.80 (0.61, 1.05) | 1.24 (0.91, 1.69) |
| Male | 1.03 (0.80, 1.33) | 1.04 (0.79, 1.37) | 0.78 (0.66, 0.91) |
| Transferred | 0.41(0.32, 0.53) | 0.51 (0.40, 0.65) | 0.84 (0.73, 0.98) |
| Systolic blood pressure <90 | 10.61(7.74, 14.55) | 5.36 (3.83, 7.51) | 1.08 (0.68, 1.70) |
| GCS | |||
| GCS motor score=1 | 93.1 (58.99, 147.04) | 54.38 (35.33, 83.70) | 10.04 (8.14, 12.39) |
| GCS motor score =2 | 48.47 (24.41, 96.25) | 41.22 (22.06, 77.03) | 8.50 (4.20, 17.21) |
| GCS motor score =3 | 37.41 (17.83, 78.48) | 31.50 (16.05, 61.82) | 11.46 (5.73, 22.94) |
| GCS motor score =4 | 9.88 (4.92, 19.86) | 7.79 (3.92, 15.50) | 10.22 (7.17, 14.57) |
| GCS motor score =5 | 4.08 (1.92, 8.70) | 3.74 (1.75, 8.02) | 4.47 (3.43, 5.82) |
| AIS | |||
| Head AIS ≥3 | 3.67 (2.80, 4.82) | 6.38 (4.68, 8.69) | 4.73 (3.97, 5.64) |
| Face AIS ≥3 | 1.42 (0.30, 6.85) | 1.68 (0.41, 6.90) | 0.909 (0.18, 4.51) |
| Neck AIS ≥3 | 1.27 (0.41, 3.90) | 1.78 (0.62, 5.09) | 0.73 (0.24, 2.22) |
| Thorax AIS ≥3 | 1.81 (1.44, 2.29) | 1.89 (1.49, 2.40) | 2.11 (1.79, 2.48) |
| Abdomen AIS ≥3 | 2.47 (1.82, 3.34) | 3.38 (2.46, 4.65) | 1.24 (0.96, 1.61) |
| Spine AIS ≥3 | 1.68 (0.96, 2.97) | 2.30 (1.26, 4.20) | 53.68 (39.15, 73.59) |
| Upper extremities AIS ≥3 | 0.83 (0.47, 1.45) | 0.66 (0.32, 1.38) | 1.82 (1.24, 2.68) |
| Low extremities AIS ≥3 | 1.08 (0.77, 1.51) | 1.20 (0.84, 1.72) | 4.97 (4.18, 5.92) |
| Blunt injury type | 0.18 (0.80, 1.33) | 0.12 (0.08, 0.18) | 0.45 (0.30, 0.66) |
| Mechanism of injury†† | |||
| Motor vehicle occupant | 1.04 (0.73, 1.48) | 0.93 (0.65, 1.34) | 1.98 (1.61, 2.43) |
| Fall | 0.49 (0.27, 0.88) | 0.49 (0.26, 0.95) | 1.75 (1.37, 2.25) |
| Firearm | 2.15 (0.29, 16.2) | *** | 0.84 (0.56, 1.28) |
| Cut/pierce | 0.39 (0.05, 2.83) | *** | *** |
| Pedestrian | 1.65 (1.03, 2.64) | 1.58 (0.97, 2.59) | 2.31 (1.72, 3.09) |
| Motorcyclist | 1.42 (0.79, 2.56) | 1.31 (0.64, 2.68) | 2.48 (1.77, 3.48) |
OR, Odds Ratio; CI, confidence interval; GCS, Glasgow Coma Score; AIS, Abbreviated Injury Scale; ATC, adult trauma center; PTC, pediatric trauma center; MTC, mixed trauma center
Variable not included because of colinearity with injury type reference group: GCS motor=6; reference group: AIS <3 for that body region;
reference group: penetrating injury;
reference group: all other injuries;
reference group: PTC
Table 4.
Mortality and Model Performance Overall and Among Significant Model Predictors
| Variable | Area under ROC | Hosmer-Lemeshow Statistic | Mortality (%) |
|---|---|---|---|
| All patients | 0.98 | 13.0 | 3.1 |
| Transferred | 0.98 | 16.3 | 1.8 |
| Not transferred | 0.98 | 9.4 | 3.6 |
| ATC† | 0.98 | 7.7 | 3.2 |
| MTC†† | 0.97 | 19.6 | 3.5 |
| PTC‡ | 0.99 | 3.1 | 0.4 |
| GCS motor score <6 | 0.91 | 11.9 | 26.4 |
| GCS motor score =6 | 0.90 | 7.4 | 0.4 |
| Head AIS <3 | 0.97 | 22.3 | 1.5 |
| Head AIS ≥3 | 0.95 | 27.0 | 12.4 |
| Thorax AIS <3 | 0.98 | 30.1 | 1.7 |
| Thorax AIS ≥3 | 0.94 | 48.3 | 10.2 |
| Abdomen AIS <3 | 0.98 | 21.5 | 2.7 |
| Abdomen AIS ≥3 | 0.94 | 25.2 | 11.1 |
| Blunt | 0.97 | 14.4 | 2.1 |
| Penetrating | 0.98 | 11.8 | 7.7 |
GCS, Glasgow Coma Score; AIS, Abbreviated Injury Scale;
ATC, adult trauma center;
MTC, mixed trauma center;
PTC, pediatric trauma center
Table 5.
Odds of Death for Penetrating Injuries by Patient and Center Characteristics
| Variable | OR of Death (95% CI) | OR of In-Hospital Death (95% CI) |
|---|---|---|
| ATC‡ | 1.83 (0.46, 7.24) | 5.58 (1.60, 19.54) |
| MTC‡ | 2.01 (0.44, 9.22) | 5.57 (1.44, 21.55) |
| Level II | 1.02 (0.61, 1.70) | 1.08 (0.64, 1.80) |
| Male | 1.17 (0.54, 2.53) | 1.17 (0.47, 2.92) |
| Transferred | 0.11 (0.06, 0.22) | 0.22 (0.12, 0.42) |
| Systolic blood pressure <90 | 5.78 (3.62, 9.21) | 3.49 (2.00, 6.13) |
| GCS | ||
| GCS motor score=1 | 245.43 (122.87, 490.23) | 83.26 (3.69, 5.15) |
| GCS motor score =2 | 83.60 (24.18, 289.07) | 69.27 (24.66, 194.59) |
| GCS motor score =3 | 57.57 (13.29, 249.41) | 43.34 (11.36, 165.29) |
| GCS motor score =4 | 25.61 (9.56, 68.64) | 23.83 (9.21, 61.66) |
| GCS motor score =5 | 8.31 (2.95, 23.38) | 8.45 (3.09, 23.14) |
| AIS | ||
| Head AIS ≥3 | 11.43 (6.16, 21.19) | 19.14 (11.32, 32.37) |
| Face AIS ≥3 | 0.15 (0.03, 0.65) | 0.26 (0.07, 1.02) |
| Neck AIS ≥3 | 0.59 (0.02, 18.57) | 1.3 (0.07, 26.48) |
| Thorax AIS ≥3 | 1.65 (0.92, 2.96) | 1.50 (0.81, 2.77) |
| Abdomen AIS ≥3 | 3.99 (2.18, 7.28) | 6.92 (3.63, 13.20) |
| Spine AIS ≥3 | 2.60 (0.81, 8.31) | 3.19 (1.14, 8.94) |
| Upper extremities AIS ≥3 | 0.87 (0.26, 2.88) | 0.82 (0.17, 3.94) |
| Low extremities AIS ≥3 | 0.84 (0.32, 2.23) | 0.87 (0.29, 2.55) |
OR, Odds Ratio; CI, confidence interval; GCS, Glasgow Coma Score; AIS, Abbreviated Injury Scale; ATC, adult trauma center; PTC, pediatric trauma center; MTC, mixed trauma center reference group: GCS motor=6; reference group: AIS <3 for that body region;
reference group: penetrating injury;
reference group: all other injuries;
reference group: PTC
DISCUSSION
Injury is the leading cause death and acquired disability among adolescents.20 To reduce the impact of trauma in this age group, it is essential to determine optimal treatment strategies. Because adolescents straddle the gap between pediatric and adult medicine, identifying differences in care among PTC, ATC and MTC will help determine the most appropriate triage strategies or identify practice strategies that can optimize the outcome for this age group. We assessed the populations presenting to PTCs as compared to ATCs and MTCs and evaluated mortality at PTCs and other center types controlling for these differences. We found that trauma center verification level differed among these center types and that the adolescent patients treated at PTCs were different from those treated at ATCs and MTCs based on injury type, mechanism of injury and severity and in demographics.
After controlling for these differences, a lower mortality was observed among adolescents treated at PTCs compared to those treated at ATCs and MTCs. This finding is similar to that of studies that evaluated mortality differences among children of all ages (including adolescents) and among younger children only.3,4,6,7,21 Previous studies that have examined outcomes for only adolescents did not find mortality differences among center types.13,14 Both studies were performed using state registry data (Ohio and Pennsylvania) rather than data obtained from a national sample. The study using data from Ohio used propensity case matching, rather than multivariate regression analysis and combined data from level I and level II PTCs to avoid identification of specific centers. Although propensity case matching is an appropriate method when the ratio between the number of predictors and outcome is low, multivariate regression is preferred when this ratio is higher, as in the cohort that we examined.24 Combining trauma center designation levels could also have influenced the outcome observed in the PTC cohort. The analysis that we used differed from that used in previous studies by using imputation rather than data exclusion to manage missing data and by considering the hierarchical structure of the data (patients nested in hospitals) in model development. Another difference is that we separately evaluated mortality in adolescents surviving to hospital admission.
Adolescent patients treated at ATCs and MTCs may have features different from those treated at PTCs. Patients treated at PTCs were younger, more likely to be transferred from other hospitals, less severely injured, and had different distribution of injury mechanisms. The demographic and injury severity differences are consistent with most previous studies including data from children of all ages3,9,16,23 and those containing only adolescent data.13,14 The patient population differences observed between those treated at PTCs and either ATCs or MTCs may be explained by either first responder triage decision-making or local triage criteria. For example, adolescents injured by firearms may be preferentially brought to ATC or MTC regardless of a given adolescent’s age because this mechanism is more likely to be treated at facilities that also see adult trauma patients.
Although more adolescents treated at PTC were transferred from another hospital, the emergency department mortality of adolescents arriving at PTC was similar to those arriving at ATC and MTC. In addition, differences in mortality between centers were similar when including and excluding patients who died in the emergency department. These results suggest that differences in mortality cannot be explained by the stability of those transferred to different center types. The greater injury severity of children and adolescents seen at ATCs and MTCs as compared to PTCs has been previously observed.6,10,11 Our study further defines this difference by showing that higher injury severity is related to more severe injuries in head, thorax, and upper extremity body regions. Although one previous study suggested improved functional outcome at PTC compared to other center types,5 we observed no differences in disposition to a rehabilitation facility in our study. This finding shows that the observed differences in mortality may not be associated with other outcome measures and warrants separate analysis.
Without finding the cause of the mortality differences between center types, we cannot take appropriate steps to improving outcomes at all centers. Other groups have found that operative intervention, such as spleen and liver procedures, were increased in children of all ages at ATCs and compared to PTCs.4, 24 In the adolescent patient population, it has been noted that overall intervention rates, not limited to operative management of solid organ injury, were higher at ATCs than PTCs.13 Determining treatment practices differences between center types will be needed to show the reasons for differences in mortality rates and may suggest areas in patient management that can be modified to improve outcomes.
This study has several limitations. First, our analysis is based on data from the NTDB that collects data only from participatory centers. Because this dataset does not represent all trauma centers, it may result in response bias. Second, although we observed a difference in mortality between those treated at PTC and other center types, our study does not provide information about what may account for these differences. Although we controlled several potential confounders that may contribute to mortality differences at different center types and observed excellent model fit, other unknown factors that we did not control for may account for these differences. Finally, although we used a large sample from the NTDB, mortality was sparsely represented in each center type. Our findings will need to be evaluated in a larger cohort of patients to further support our findings.
Trauma centers dedicated to the treatment of pediatric patients see a different adolescent population than ATC and MTC. After controlling for these differences, we observed that adolescent trauma patients have lower overall and in-hospital mortality when treated at PTC. The optimal care of adolescents at all center types requires the identification of either additional patient differences or treatment practices that account for this mortality difference. Analysis of the association of specific care processes with mortality at center types will be needed to further clarify the etiology of these mortality differences.
ACKNOWLEDGEMENTS
Research reported in this publication was supported by the National Institute of General Medical Sciences of the National Institutes of Health under award number R01GM087600-03.
Rachel Webman and Randall S. Burd had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
APPENDIX
Imputation Procedure
To account for the hierarchical structure of the data (patients nested within hospitals), we performed multiple imputation using a dataset that included patient variables (age, gender, mechanism of injury, initial systolic blood pressure, initial pulse rate, initial respiratory rate, GCS score motor component, AIS post-dot values for each body region and mortality) and hospital level variables (type of center and center level) as well as variables representing the interaction between each of these variables and those representing each of the 270 treating hospitals. This approach led to an increase in the total number of variables to >24,000 and increased the number of variables with missingness from >7,300.
Standard imputation packages cannot be used when a large number of covariates are included. To overcome this issue, we used the fully conditional specification method to perform this imputation.17 We imputed continuous and binary variables using regularized least squares and regularized logistic regression respectively, using software implementations optimized for high-dimensional data.25 To account for error in predicting the missing values, the final imputed value was selected from a probability distribution based on the regression method used to impute each missing value. For binary variables, we chose the final imputed value from a Bernoulli distribution, with probability equal to the predicted value generated by logistic regression. For continuous predictors, we computed a prediction interval around the predicted value. We then generated the final imputed value randomly from a normal distribution around the predicted value, with the standard deviation equal to the prediction interval. These methods for choosing the missing values from their respective distributions are a standard practice for imputation.26
REFERENCES
- 1.Rotondo MF, Cribari C, Smith RS. Resources for Optimal Care of the Injured Patient. American College of Surgeons;2014. [Google Scholar]
- 2.National Trauma Data Bank 2014, Annual Report. 2014; https://www.facs.org/~/media/files/quality%20programs/trauma/ntdb/ntdb%20annual%20report%202014.ashx. Accessed March 15, 2016.
- 3.Hall JR, Reyes HM, Meller JL, Loeff DS, Dembek R. The outcome for children with blunt trauma is best at a pediatric trauma center. J Pediatr Surg. 1996;31(1):72–76. [DOI] [PubMed] [Google Scholar]
- 4.Potoka DA, Schall LC, Gardner MJ, Stafford PW, Peitzman AB, Ford HR. Impact of pediatric trauma centers on mortality in a statewide system. J Trauma. 2000;49(2):237–245. [DOI] [PubMed] [Google Scholar]
- 5.Potoka DA, Schall LC, Ford HR. Improved functional outcome for severely injured children treated at pediatric trauma centers. J Trauma. 2001;51(5):824–832. [DOI] [PubMed] [Google Scholar]
- 6.Densmore JC, Lim HJ, Oldham KT, Guice KS. Outcomes and delivery of care in pediatric injury. J Pediatr Surg. 2006;41(1):92–98. [DOI] [PubMed] [Google Scholar]
- 7.Pracht EE, Tepas JJ 3rd, Langland-Orban B, Simpson L, Pieper P, Flint LM. Do pediatric patients with trauma in Florida have reduced mortality rates when treated in designated trauma centers? J Pediatr Surg. 2008;43(1):212–221. [DOI] [PubMed] [Google Scholar]
- 8.Oyetunji TA, Haider AH, Downing SR, et al. Treatment outcomes of injured children at adult level 1 trauma centers: are there benefits from added specialized care? Am J Surg. 2011;201(4):445–449. [DOI] [PubMed] [Google Scholar]
- 9.Sathya C, Alali AS, Wales PW, et al. Mortality among injured children treated at different trauma center types. JAMA Surgery. 2015; 150(9):874–881. [DOI] [PubMed] [Google Scholar]
- 10.Junkins EP Jr, O’Connell KJ, Mann NC. Pediatric trauma systems in the United States: do they make a difference? Clin Pediatr Emerg Med. 2006;7(2):76–81. [Google Scholar]
- 11.Petrosyan M, Guner YS, Emami CN, Ford HR. Disparities in the delivery of pediatric trauma care. J Trauma. 2009;67(2 Suppl):S114–119. [DOI] [PubMed] [Google Scholar]
- 12.Little WK. Golden hour or golden opportunity: early management of pediatric trauma. Clin Pediatr Emerg Med.11(1):4–9. [Google Scholar]
- 13.Matsushima K, Schaefer EW, Won EJ, Nichols PA, Frankel HL. Injured adolescents, not just large children: difference in care and outcome between adult and pediatric trauma centers. Am Surg. 2013;79(3):267–273. [PubMed] [Google Scholar]
- 14.Walther AE, Pritts TA, Falcone RA, Hanseman DJ, Robinson BR. Teen trauma without the drama: outcomes of adolescents treated at Ohio adult versus pediatric trauma centers. J Trauma Acute Care Surg. 2014;77(1):109–116 [DOI] [PubMed] [Google Scholar]
- 15.Nakayama DK, Copes WS, Sacco W. Differences in trauma care among pediatric and nonpediatric trauma centers. J Pediatr Surg. 1992;27(4):427–431. [DOI] [PubMed] [Google Scholar]
- 16.Osler TM, Vane DW, Tepas JJ, Rogers FB, Shackford SR, Badger GJ. Do pediatric trauma centers have better survival rates than adult trauma centers? An examination of the National Pediatric Trauma Registry. J Trauma. 2001;50(1):96–101. [DOI] [PubMed] [Google Scholar]
- 17.Muthén L, Muthén B, eds. Mplus User’s Guide Los Angeles, CA: Muthén & Muthén; 1998–2012; No. Seventh. [Google Scholar]
- 18.Rubin DB. Multiple Imputation for Nonresponse in Surveys. Hoboken, NJ: John Wiley & Sons, Inc.; 1987. [Google Scholar]
- 19.Schafer JL. Analysis of Incomplete Multivariate Data. London, UK: Chapman & Hall/CRC; 1997. [Google Scholar]
- 20.10 leading causes of death by age group, United States - 2013. Injury Prevention and Control: Data and statistics (WISQUARSTM). Accessed December 20, 2015.
- 21.Amini R, Lavoie A, Moore L, Sirois MJ, Emond M. Pediatric trauma mortality by type of designated hospital in a mature inclusive trauma system. J Emerg Trauma Shock. 2011;4(1):12–19. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Cepeda MS, Boston R, Farrar JT, Strom BL. Comparison of logistic regression versus propensity score when the number of events is low and there are multiple confounders. Am J Epidemiol. 2003;158(3):280–287. [DOI] [PubMed] [Google Scholar]
- 23.Segui-Gomez M, Chang DC, Paidas CN, Jurkovich GJ, Mackenzie EJ, Rivara FP. Pediatric trauma care: an overview of pediatric trauma systems and their practices in 18 US states. J Pediatr Surg. 2003;38(8):1162–1169. [DOI] [PubMed] [Google Scholar]
- 24.Murphy EE, Murphy SG, Cipolle MD, Tinkoff GH. The pediatric trauma center and the inclusive trauma system: Impact on splenectomy rates. J Trauma Acute Care Surg. 2015;78(5):930–933. [DOI] [PubMed] [Google Scholar]
- 25.Van Buuren S, Brand JPL, Groothuis-Oudshoorn CGM, Rubin DB. Fully conditional specification in multivariate imputation. J Stat Comput Simul. 2006;76(12):1049–1064. [Google Scholar]
- 26.Suchard MA, Simpson SE, Zorych I, Ryan P, Madigan D. Massive parallelization of serial inference algorithms for a complex generalized linear model. ACM Trans Model Comput Simul. 2013;23:1–18. [DOI] [PMC free article] [PubMed] [Google Scholar]