We have successfully derived a highly sensitive clinical decision rule for the prediction of clinically significant intracranial pathology in children according to strict methodological standards, in the world's largest prospective cohort of children with head injuries. This rule, if validated, will enable clinicians to request computed tomography scans for their patients on the basis of strong evidence in children.
If our rule is subsequently validated, we believe that patients regarded as high risk should undergo computed tomography scanning to look for intracranial pathology, whereas those with a normal scan might be regarded as low risk. Children who are regarded as low risk by a validated rule should be carefully counselled so that they understand the high‐risk symptoms for which they should return. Although the CHALICE rule will increase the rate of computed tomography scanning, we envisage that the admission rate could be markedly reduced and thus the cost implications of our rule could be neutral. This remains to be shown in further studies.
Any decision rule that is to be of value to clinicians seeing undifferentiated children with head injury must be applicable to all such children. At study design stage it was important to take a pragmatic approach to this requirement, as it was clear that the rate of significant events was low and that it would be unethical to expose large numbers of children to unnecessary major radiation exposure. A balance needed to be struck between inclusion bias (selecting only children who already fulfilled some existing rule, which indicated that they should undergo computed tomography scanning) and the ethical limits that would be placed on the study. Thus, a composite end point was agreed and ethically approved. This involved the identification of all children who died, had neurosurgical intervention or had abnormalities on a computed tomography scan. This composite end point was reinforced with a prospective, thorough follow‐up strategy, which was designed to ensure that no children who died as a result of their head injury or who had late neurosurgical intervention were missed. Although this approach can be criticised because some children with abnormalities on the computed tomography scan may not have been identified as they did not undergo such scanning, it has considerable strength in that children with both immediate or late significant events are included. Thus, it can be assumed that any undiagnosed abnormalities on the computed tomography scan were clinically non‐significant. We believe that this method is superior to the alternative one of telephoning each patient 2 weeks after admission, and previous studies have been criticised for their inability to contact all patients for follow‐up.9,21
We do not believe that there is considerable circularity between the clinical care drivers during the study and the findings as stated in the CHALICE rule. In particular, the recommendations of the RCS17
differ from those of the CHALICE in that a major indication for computed tomography in a district general hospital as per the RCS guidance is fracture on the SXR. CHALICE has looked critically at both the indications for SXR and the computed tomography scan in detail, and distilled from a long list (table A available at http://www.archdischild.com/supplemental
) those that truly indicate the likelihood of clinically significant intracranial injury (the composite outcome). It is not surprising that some of the CHALICE rules support previous expert opinion. Perhaps it would be more surprising if it did not.
Our study has other limitations. Clinicians were not always blinded to the outcome of the computed tomography scan before completing the proforma. Although clinicians completed most proformas on the first clinical examination, they were reminded at a later date if they did not do this. Our proforma was also the clinical record sheet and, thus, compliance was generally high. However, we do not have data on the number of missed patients that could have been eligible. Finally, our study is only a derivation study and it should now be prospectively validated, with its reproducibility, acceptability, usability and economic effect evaluated across multiple sites.
Few studies have successfully derived a clinical decision rule applicable to all children from a large cohort of patients. Palchak et al23
in 2003 derived a rule on examining 2043 patients from a single hospital, aged <18 years, who had had head trauma and showed positive findings on history or clinical examination such as LOC, amnesia, vomiting or headache. Of nine predictive variables assessed, abnormal mental status, clinical signs of skull fracture, history of vomiting, scalp haematoma (in children aged
2 years) or headache identified 96 of 98 patients with positive intracranial pathology on computed tomography scanning (98% sensitivity, 95% CI 93% to 100%).
Greenes and Schutzman23
conducted a prospective study on 608 patients aged <2 years in a single hospital.23
Their results support our finding that children with suspected non‐accidental injury, history of lethargy or a major scalp haematoma had an increased risk of significant intracranial injury. In addition, they found that LOC, seizures or vomiting alone was not adequate to predict intracranial injury, and that the absence of clinical symptoms or signs did not fully exclude the possibility of uncovering positive pathology. Together with a systematic review by the American Academy of Pediatrics and expert consensus, these authors formalised this study by producing guidelines for head injuries in children <2 years old.24
They allocated patients into four risk groups, with computed tomography scanning recommended in the highest risk group of children who vomited >3 times or had an LOC, a history of lethargy, a high‐risk mechanism or considerable bruising. Although these guidelines agree with many of our recommendations, including the suggestion that patients are safe for discharge after a negative finding on a computed tomography scan, like ours, their guidelines have not yet been validated in other hospitals.
Haydel and Shembekar25
in 2003 assessed the New Orleans criteria13
in children aged >5 years. They assessed 175 children with GCS 15 from a single institution and concluded that the 14 positive computed tomography scans that they found could be identified by their rule, which was derived from and validated in scans of adults.
The National Institute of Clinical Excellence, UK, systematically reviewed all studies on head injury up to 2002. They concluded that no studies on childhood head injury could be used to construct a robust rule, and advocated extrapolating the Canadian CT head rule for children. We assessed the performance of this rule in children,26
extending our previous work to the full database. We found that the sensitivity was 94% (95% CI 91% to 97%) and the specificity was 89% (95% CI 89% to 90%), with a computed tomography ordering rate of 12%. Sixteen patients would have been missed if this rule had been strictly applied to our database. Thus, we conclude that the CHALICE rule is safer than extrapolating an adult head injury guideline to children without a considerably increased computed tomography ordering rate. We also found that the RCS guidelines were being poorly applied to children, with half the radiographs, admissions and computed tomography scans recommended by the RCS guidelines not being carried out.26
Many of the variables identified as significant in our study have also been identified in a meta‐analysis of clinical variables identified from 16 papers in the paediatric literature.27
Focal neurology, seizures, LOC and abnormal GCS were all major predictors, but headache was found not to considerably predict significant intracranial pathology.
Although computed tomography scanning is a safe procedure for those who are able to comply with the investigation, young children may require sedation, which is not without complications.28
Occasionally, the risk of the investigation should be balanced with the possibility of delayed diagnosis with observation alone in children who have been identified as high risk by our rule.
Finally, we do not support the continued use of SXR for children with acute head injury, except for highly selected patients who may have had non‐accidental injury. Although many studies have found that the evidence of a fracture on SXR markedly increases the incidence of intracranial pathology,29,30,31
we agree with other studies that the SXR has a poor sensitivity for identifying patients with intracranial pathology, that fractures are identified on only a small number of radiographs and that fractures are easily missed by those interpreting the radiographs.32
We thus advocate that a clear decision rule (such as the CHALICE rule) that uses clinical variables alone and that identifies children at high risk of significant intracranial injury for computed tomography scanning is the optimal rule for the management of head injuries in children.