Cervical Spine Injury Risk Factors in Children With Blunt Trauma

Between March 2014 and November 2016, we conducted a prospective observational study of consecutive children evaluated at 4 tertiary care children’s hospitals. Participating hospitals were academic freestanding children’s hospitals, American College of Surgeons Pediatric Level 1–verified, and PECARN members.

Inclusion criteria consisted of children younger than 18 years who presented to the emergency department (ED) for blunt trauma and were transported from scene of injury by emergency medical services (EMS) in SMR, underwent trauma team evaluation, and/or had cervical spine imaging ordered in the ED. We excluded children whose injury mechanism was solely penetrating trauma, whose legal guardian had a substantial language barrier, who were in state’s custody, or who were transferred from the study site for definitive care.

Data collection methods for this study were previously reported.³³ In brief, trained research personnel administered electronic branch-logic questionnaires to treating ED providers to gather observations regarding CSI risk factors before knowledge of cervical spine imaging results, if ordered or if obtained at a transferring hospital, before knowledge of their institutional radiologist’s interpretation. Risk factors included those previously demonstrated to be associated with CSIs, that had biological and/or anatomic plausibility, and with good interrater reliability.^16,34,–36 These factors included mechanisms of injury and injury biomechanics (high-risk MVC³⁷: passenger compartment intrusion including roof >12 inches at occupant site or >18 inches at any site, partial or complete ejection from vehicle, death of a passenger in the same compartment, and/or vehicle telemetry consistent with high-risk crashes; diving; axial load; or clotheslining: biomechanical force caused by a rope, cable, or similar item exerting traction on the neck while the body is in forward motion), patient history variables (predisposing conditions, loss of consciousness, neck pain, inability to move neck, paresthesia, numbness or weakness), and physical examination findings (altered mental status, intubation, signs of substantial head injury other than altered mental status, signs of basilar skull fracture, posterior midline tenderness to palpation, limited range of neck motion, substantial torso injury, substantial thoracic injury, respiratory distress, decreased oxygen saturation, substantial abdominal injury, substantial pelvic injury, thoracic spine tenderness, lumbar spine tenderness, sacral spine tenderness, focal neurologic deficits [paresthesia, decreased sensation, and weakness]). We defined substantial injuries as life-threatening and warranting surgical intervention or warranting inpatient observation. Respondents could not skip items but could indicate if a risk factor was unknown, unable to be assessed, not assessed, or unknown.

CSI was defined as vertebral fractures, ligamentous injury, intraspinal hemorrhage, or spinal cord injury (either on magnetic resonance imaging or spinal cord injury without radiographic association) involving the cervical region (occiput to seventh cervical vertebra, including ligamentous structures attaching the seventh cervical vertebra to first thoracic vertebra). Study personnel determined CSI diagnosis by reviewing cervical spine imaging reports and, if applicable, spine surgeon consultation notes. If imaging report diagnoses conflicted with spine surgeon consultation, we contacted the treating spine surgeon and/or the study’s spine surgeon for clarification. For children who did not undergo ED cervical spine imaging, the medical record was reviewed 21 days later for subsequent imaging. If no subsequent imaging was noted, a follow-up call with the legal guardian was conducted 21 to 28 days after the ED visit to verify absence of CSI.

The analytic cohort was described by using counts and frequencies. We dichotomized observations of CSI risk factors into 2 categories: present versus all other responses. Relative risks (RRs) of CSI with 95% confidence intervals (CIs) were estimated for all candidate risk factors. Risk factors from the PECARN retrospective case-control model were fit to prospective data by multivariable logistic regression and odds ratios (ORs), and 95% CIs were calculated for each risk factor. In a separate analysis, we derived a de novo CSI risk model using only variables with a bivariable lower 95% confidence limit >1 as candidates for inclusion and multiple stepwise logistic regression. Multivariable ORs and 95% CIs were calculated for variables retained in the model.

To assess model performance characteristics, each model was applied to the data as a decision tool where subjects were considered at risk for CSI if they had at least 1 positive value for any model variable. Sensitivity, specificity, negative predictive value, and positive predictive value for CSI were calculated for each model. Observer bias was assessed by repeating this assessment with the subcohort of children presenting directly from scene of injury (ie, excluding those transferred from referring institutions).

Expected overall cervical spine imaging rates were estimated for each model by assuming that any 1 positive variable would result in imaging. Estimates for CT use were calculated by applying each model as a prediction rule to the subpopulation that underwent CT.

All analyses were conducted by using SAS version 9.4 (SAS Institute, Inc, Cary, NC).

This study was approved by all sites’ institutional review boards to be conducted under waiver of written informed consent. Legal guardians of enrolled subjects were provided an informational handout that contained withdrawal instructions.

Supplemental Fig 1 summarizes study enrollment. In total, we screened 11 809 children who experienced blunt trauma, of whom 5764 were eligible and 4144 were enrolled. Characteristics of the 1620 children missed for enrollment were previously described.³³ Compared with enrolled children, these children were slightly younger (median age 8.5 vs 9.4 years), more likely white (63% vs 60.8%), and less likely involved in MVCs (22.5% vs 27.3%) and all-terrain vehicle, motorcycle, or scooter crashes (5.1% vs 10.0%). Legal guardians withdrew 53 children.

Table 1 presents total cohort and CSI subpopulation characteristics. Of 4091 children, 74 (1.8%) had CSIs. Cohort mean age was 9.4 years, and mean age of those with CSIs was 10.7 years. A total of 39.3% of the cohort was younger than 8 years and 23 (1.4%) of these children had CSIs. The CSI subpopulation included more boys, white children, and non-Hispanic children. The most common mechanisms of injury for both groups were MVC and sports-related mechanisms. Children with CSIs were more likely to undergo trauma team evaluation and less likely to be discharged. Thirteen children died in the ED, none of whom had documented CSIs.

The bivariable RRs for all risk factors and multivariable ORs for risk factors in the PECARN and de novo models are presented in Table 2. Fourteen factors had significant bivariable associations with CSIs in the study cohort: diving, axial load, clotheslining, loss of consciousness, neck pain, reported inability to move the neck, altered mental status, signs of basilar skull fracture, substantial torso injury, substantial thoracic injury, intubation, respiratory distress, decreased oxygen saturation, and focal neurologic deficits on examination.

Most PECARN model risk factors maintained independent associations with CSIs in the multivariable analysis (diving, neck pain, reported inability to move neck, altered mental status, substantial torso trauma, and focal neurologic deficits); however, 3 variables did not (high-risk MVC, predisposing conditions, and limited neck range of motion on exam). The de novo model for CSI risk included 7 variables: diving mechanism, axial load, neck pain, reported inability to move neck, altered mental status, respiratory distress, and intubation.

In Table 3, we report test accuracies for each model. Compared with the PECARN model, the de novo model had marginally better sensitivity and specificity (90.5% and 45.6% vs 91.9% and 50.3%, respectively) and negative and positive predictive values (99.6% and 3.0% vs 99.7% and 3.3%, respectively). When the analysis was repeated for the 3138 children who presented directly to the study sites, including 43 with CSIs, the results were similar. The sensitivity and specificity of the PECARN model in this subcohort were 93.0% (95% CI 85.4%–100%) and 42.1% (95% CI 40.3%–43.8%), respectively. The sensitivity and specificity for the de novo model in the subcohort were 95.3% (95% CI 89.1%–100%) and 45.9% (95% CI 44.1%–47.7%), respectively.

In Table 4, we present extrapolated imaging rates if these CSI risk models guided decision-making. Overall imaging rates would decrease by more than a third, from an observed rate of 78.2% to 55.1% or 50.5%, in the PECARN and de novo models, respectively. CT scan use would decrease by more than half, from 15.8% to 7.5% or 7.1%. However, both models would have missed children with CSIs; 7 with the PECARN model and 6 with the de novo model.

In Table 5, we present characteristics of the 7 children with CSIs missed by ED providers’ assessments of PECARN risk factors. On retrospective chart review, all but 1 child had PECARN risk factors: altered mental status (3), neck pain (3), and focal neurologic deficit (1). The child that lacked a PECARN risk factor reported back pain and severe upper thoracic tenderness and had a C7 burst fracture. None of the missed children required surgical intervention; however, 1 required a brace and another required a rigid cervical collar.

We included children transferred from other hospitals with imaging. Knowledge of imaging results may have influenced ED providers’ reported risk factor observations. This potential bias was mitigated by efforts to capture ED providers’ observations before radiologists overread outside imaging. Including children transferred from outlying hospitals was important to ensure that the study population was representative of the entire injured population. Importantly, a subanalysis that included only children presenting directly to the study sites yielded similar prediction model sensitivity and specificity estimates. The analyses were underpowered to provide narrow CIs around ORs and RRs point estimates for uncommon risk factors and therefore lacks sufficient power to fully assess model performance characteristics.

In this 4-center prospective observational study of children experiencing blunt trauma, we identified 14 factors that were associated with CSI in bivariable analyses: diving, axial load, clotheslining, loss of consciousness, neck pain, reported inability to move neck, altered mental status, intubation, basilar skull fracture, substantial torso injury, substantial thoracic injury, respiratory distress, decreased oxygen saturation, and focal neurologic deficits. We also demonstrated by multivariable analyses that both the PECARN model (high-risk MVC, diving mechanism, conditions predisposing for CSI, neck pain, reported inability to move neck, altered mental status, limited neck range of motion on exam, substantial torso injury, and focal neurologic deficits) and a de novo model (diving mechanism, axial load biomechanics, neck pain, reported inability to move neck, altered mental status, intubation and respiratory distress) have reasonable test accuracy and the potential to limit imaging if used in clinical decision-making. Collectively, these results support the feasibility of deriving a CSI prediction rule for children.

In this study, we confirm the important association between CSI and head injury. Signs of substantial head injury (basilar skull fracture), signs of traumatic brain injury (altered mental status), respiratory failure and intubation, and head-first impact (diving mechanism and axial load biomechanics) held significant independent associations with CSI. Adult CSI prediction rules for cervical imaging after blunt trauma are contingent on normal mental status for clinical clearance. One argument for this practice is that a patient’s examination is unreliable when mentation is altered. Another compelling argument is that head-first impact predisposes to CSI and that mental status is a good marker for the severity of the head-first, axial load impact and therefore is also a good marker for CSI risk. Future development of a robust pediatric CSI prediction rule should be focused on stratification based on mental status because it may be meaningful in determining which children to triage to CT scan. If the CSI risk based on altered mental status is sufficiently high, this risk would outweigh the risk of CT-related ionizing radiation exposure.

The risk factors identified in this study differ from the adult literature with respect to neck findings. The NEXUS and the Canadian C-spine rules rely on the neck examination, both include posterior midline neck tenderness, and the Canadian C-spine rule also includes lack of normal neck range of motion on examination.^17,18,22 Both the PECARN retrospective case-control study of CSI in children and the current study found self-report of neck pain and difficulty moving the neck to be predictive as opposed to being restricted to the examination findings of posterior midline neck tenderness and limited range of motion.³² One possible reason for this difference is that SMR alters physical examination in the ED. Researchers previously demonstrated increased neck and spine tenderness after supine positioning in a cervical collar on a rigid longboard.^15,38,–40 Another possibility is that ED providers defer aspects of the neck examination for children who arrive in a cervical collar, rendering self-report the only means for assessing these findings. Within this study cohort, ED providers did not assess children transported from the scene of injury by EMS for posterior midline cervical tenderness in 12.3% and neck range of motion in 41.2%.¹⁶ 

Aside from limited range of neck motion on examination, 2 additional PECARN model risk factors did not maintain their associations with CSI in this study: predisposing conditions and high-risk MVC mechanism.³² Both of these risk factors are included in the Canadian C-spine Rule.²² Predisposing conditions in this cohort had low prevalence, occurring in <1% of children. The main aim of this study was to establish the infrastructure for conducting a larger cohort study and is therefore underpowered for rare findings and those with weaker but significant associations. High-risk MVCs, however, were prevalent within the cohort. Although high-risk MVC did not maintain a significant association with CSI, it may be important in prediction rule derivation when there is a sufficient sample size to pair high-risk mechanisms with physical examination findings to determine CSI risk.

The PECARN model had sufficiently accurate performance characteristics in this cohort that use of the model for clinical decision support would significantly reduce cervical spine imaging, although 7 CSIs would have been missed. De novo modeling revealed that there are opportunities to improve the PECARN model. Both models were similar in that axial load biomechanics, mental status assessment, and neck complaints were important factors associated with CSI. However, the de novo model included respiratory distress and intubation, which could reflect substantial head, chest, or spinal cord injury. This compares to focal neurologic deficits and substantial torso injury in the PECARN model. With larger sample sizes, it is likely that all factors that are signs of severe traumatic brain injury and spinal cord injury would be identified as significant in prediction rule derivation.

Clinical prediction rule use has come under scrutiny by some authors using the argument that physician judgement is rarely assessed in studies for comparison; in those studies where it was assessed, prediction rules seldom improve on it.⁴¹ In a separate analysis of this cohort, however, we demonstrated ED provider gestalt for identifying children with CSIs is poor.¹⁶ Twenty percent of the children with CSIs in that analysis were rated by providers as having <1% risk for CSI. This is consistent with investigations of head injury in children after blunt trauma.^42,43 Furthermore, PECARN developed a decision support tool for imaging children after blunt head trauma.^44,–46 Implementation of the PECARN head injury rule has repeatedly demonstrated reduction in CT use during the evaluation of children after blunt head trauma.^46,–50 Developing and implementing clinical prediction rules to aid in identifying children at nonnegligible risk for CSIs after blunt trauma is likely to empower clinical decision-making and appropriate imaging use.

In this prospective cohort of children with blunt trauma, we confirmed that there are risk factors with good test accuracy in identifying CSIs. We also demonstrated that incorporating these risk factors into a clinical prediction rule has the potential to substantially reduce cervical spine imaging during trauma evaluation of children. A future, adequately powered prospective observational study aimed at using these risk factors to construct a definitive pediatric CSI prediction rule is warranted.

We thank the EMS and ED providers, as well as the research staff, whose efforts made this study possible.

CI

confidence interval

CSI

cervical spine injury

CT

computed tomography

ED

emergency department

EMS

emergency medical services

MVC

motor vehicle crash

NEXUS

National Emergency X-Radiography Utilization Study

OR

odds ratio

PECARN

Pediatric Emergency Care Applied Research Network

RR

relative risk

SMR

spinal motion restriction