Traumatic brain injury (TBI) is a leading cause of childhood death and disability worldwide. In the United States, childhood head trauma results in 3000 deaths, 50 000 hospitalisations, and 650 000 emergency department (ED) visits annually.^1,2 Children with seemingly minor head trauma, defined here by Glasgow Coma Scale (GCS) scores of 14–15, account for 40–59% of those with TBIs,^3–5 and present a perplexing problem to clinicians. Despite the frequency of childhood minor head trauma, there exists no highly accurate and reliable clinical scoring system for separating those children with minor head trauma at negligible risk of TBI from those at greater risk. Children with TBIs who present with signs of seemingly minor head trauma are at risk for delayed diagnosis and treatment, and such unrecognised TBIs are a source of preventable morbidity.^6–8 Nevertheless, although some TBIs in children with minor head trauma are initially unrecognised and therefore neuroimaging is delayed, a great many more children evaluated for minor head trauma receive unnecessary neuroimaging.

More than 50% of children evaluated in EDs for head trauma in the United States undergo cranial computed tomography (CT) scanning,¹ and this percentage doubled between 1996 and 2000.¹ Children with minor head trauma account for 70–85% of children who receive cranial CT scans during their ED evaluation for head trauma in the United States,^4,5 yet only 4–8% of these CT scans show TBI.^1,4,5,9–12 Furthermore, approximately 5% of those with positive CT scans undergo operative intervention.^4,9 Therefore CT scans are used inefficiently.

Although CT scanning is the reference standard for diagnosing TBI and delayed recognition of TBI increases morbidity, overuse of CT scanning has important drawbacks. These include transporting the child away from the supervised setting of the ED, the occasional requirement for pharmacological sedation with its potential for complications, and additional health care costs and resource utilisation. The most important drawback of inappropriate CT utilisation, however, is radiation exposure that may lead to malignancy and other sequelae. CT scanning accounts for most (67%) of the collective effective radiation dosage from diagnostic imaging in the United States, although only 11% of all radiographic studies are CT scans.¹³ The typical radiation dosage from a current paediatric cranial CT scan is 300–600 times greater than that for a routine chest radiograph.¹⁴ The estimates of lifetime attributable risk for fatal cancer from one current generation cranial CT scan ranges from 1 per 2000 scans for young infants to 1 per 5000 for those 10 years old.^15,16 Furthermore, for every case of fatal cancer caused by cranial CT scans, several cases of non-fatal cancer may be induced.^15,17 In addition, it has recently been reported that low doses of ionising radiation in infancy may adversely affect cognitive abilities in young adulthood.¹⁸ Nevertheless, although the risks of radiation induced malignancy can be estimated, they are small for any given individual.

It is apparent that both under-utilisation and over-utilisation of CT scanning have potential ill effects, and the use of CT scanning following blunt head trauma must be optimised. Unfortunately, there have been few published prospective studies on this topic in children, and none with the necessary size and patient diversity to ensure narrow confidence intervals around the estimates of model accuracy, and wide generalisability of the resulting model. As a consequence, there is substantial variation among ED physicians in the use of CT scans for the evaluation of these children,^19–21 and the creation of a decision rule to assist with decision making for CT use is a priority among ED physicians.^21,22

The lack of reliable data on risk factors for TBI in children with minor head trauma motivated Dunning et al to perform a meta-analysis of the literature, which is reported in this issue of Archives.²³ Using an appropriate, detailed search strategy, the authors identified 16 papers meeting criteria for inclusion in the meta-analysis. In the meta-analysis, the authors assessed the predictive effect of six clinical symptoms and signs, in addition to plain radiography, for identifying which children have intracranial haemorrhage after minor head trauma. The authors reported the pooled estimates for the relative risks of these seven individual criteria. They found that the variables with significant point estimates for risk of TBI were focal neurological findings, skull fracture, a reduced level of consciousness, and a history of loss of consciousness (with seizure trending towards significance).

In the absence of rigorous prospectively acquired data, the meta-analysis by Dunning et al helps us focus on the areas of greatest uncertainty and controversy. There is little debate that a child with focal neurological signs or a depressed level of consciousness after head trauma needs emergent neuroimaging. In addition, several prospective paediatric studies have shown that signs of a skull fracture are significantly predictive of intracranial haemorrhage.^4,5 There is much more controversy, however, over the importance of headache, vomiting, and history of loss of consciousness.

Readers must be cautious, however, of translating the findings of Dunning et al into clinical practice. As noted by the authors, the heterogeneous nature of the studies and the wide reported rates of intracranial haemorrhage limit the ability to pool the data and translate the results. Even the definition of “minor head trauma” itself varies between studies. Dunning et al defined minor head trauma as GCS scores of 13–15, whereas others have defined this as GCS scores of 14–15,⁴ and GCS scores of 15.⁹ In clinical practice, there is not substantial controversy over imaging children with GCS scores of 13, whereas there is great controversy over children with GCS scores of 15. In our experience, 26.5% (95% CI 14.9% to 41.1%) of children with GCS scores of 13 have intracranial haemorrhage versus 2.6% (95% CI 1.7% to 3.9%) of those with GCS scores of 15.⁴

Another conundrum is the clinical assessment of pre-verbal children with minor head trauma. As noted by Dunning et al, certain variables cannot be used reliably in the assessment of these children, specifically headache, amnesia, and dizziness. It is unclear how headache was assessed in pre-verbal children in the studies compiled by Dunning et al. Furthermore, scalp haematoma is one of the most important clinical variables in the assessment of infants with minor head trauma,^3,24 and was not assessed in the meta-analysis by Dunning et al. The need for a separate neuroimaging decision rule for pre-verbal children is apparent.

An additional important limitation to the clinical utility of the Dunning et al study is the univariate nature of the meta-analysis. Only by assessing the importance of particular variables in the presence or absence of other variables (through multivariable analysis) can one truly assess the independent predictive effect of any particular variable. Some variables which seem important in univariate analysis may no longer be important in the multivariable analysis. For example, it seems intuitive that a child with a history of loss of consciousness after head trauma has a greater risk of TBI than a child without this history. More controversial, however, is the child with an isolated loss of consciousness (that is, without any other symptoms or signs of TBI). Is this finding in isolation sufficiently predictive of intracranial haemorrhage to require neuroimaging? Several multivariable analyses that assessed this variable, including those in recent large prospective studies, suggest not.^4,5,9,10,25

Similarly, other variables which appear not to be predictive in univariate analyses may in fact be predictive when in the presence of other variables. The variables headache and vomiting are not significant in the univariate meta-analysis of Dunning et al, but were important in recent large prospective multivariable studies on this topic.^4,9 This may be due to the insufficient characterisation of headache and vomiting in previous retrospective studies on the topic, as noted by Dunning et al. It may also be that these variables emerge as important only in the presence or absence of other variables. Finally, in a meta-analysis such as this, one cannot determine whether the absence of certain high risk factors decreases the overall risk of intracranial haemorrhage below the clinical threshold to obtain emergent neuroimaging.

The Dunning et al study increases our knowledge base about prediction of TBI. It is nonetheless apparent, however, that accurate, sufficient and generalisable data on which to derive reliable evidence based guidelines for neuroimaging children with minor head trauma are still lacking. Fortunately, multicentre research networks in the United Kingdom, Canada, and in the United States have large ongoing studies to gather these data. The Pediatric Emergency Care Applied Research Network (PECARN), a research network of 25 hospital EDs in the United States,²⁶ will collect such data on approximately 20 000 children with minor head trauma. Using these types of networks to conduct such studies will ensure adequate sample sizes, standardised data collection, appropriate multivariable analyses, and enrolment of a clinical cohort of suitable diversity to generalise the results. These studies will ultimately affect clinical decision making and improve the care of children with minor blunt head trauma.