CATCH: a clinical decision rule for the use of computed tomography in children with minor head injury

Study setting and population

We conducted a prospective cohort study in 10 Canadian pediatric teaching institutions and enrolled consecutive children (0–16 years of age) if they presented to one of the emergency departments after sustaining an acute minor head injury. Eligibility was based on patients having all of the following: (a) blunt trauma to the head resulting in witnessed loss of consciousness, definite amnesia, witnessed disorientation, persistent vomiting (two or more distinct episodes of vomiting 15 minutes apart) or persistent irritability in the emergency department (for children under two years of age); (b) initial score on the Glasgow Coma Scale in the emergency department of at least 13, as determined by the treating physician; and (c) injury within the past 24 hours. Patients were excluded if they had obvious penetrating skull injury or obvious depressed fracture, acute focal neurologic deficit, chronic generalized developmental delay or head injury secondary to suspected child abuse. Patients who were returning for reassessment of a previously treated head injury and those who were pregnant were also excluded. The research ethics committee of each study hospital approved the study.

Standardized assessment of patients

Staff physicians in the emergency department who were certified in pediatrics, emergency medicine or family medicine or supervised residents (in their second year of training or above) assessed the patients. These physician assessors each underwent a one-hour training session on evaluating patients for 26 standardized clinical findings from the history, general examination and neurologic status. These potential predictor variables had been selected a priori by a team of investigators (M.H.O., T.P.K., A.J., G.J., B.B., L.C.-K., M.P., D.M., C.N.-J., B.T., I.G.S.) on the basis of a review of the existing literature and results of a pilot study. The assessors recorded the findings of the standardized assessment on data collection sheets before any CT. When it was feasible, a second emergency physician independently assessed each patient, to allow determination of interobserver agreement. For patients transferred from a primary care hospital, the study assessments took place after arrival at the study site.

Outcome measures and their assessment

The primary outcome was need for neurologic intervention, and the secondary outcome was brain injury on CT. The need for neurologic intervention was defined as either death within seven days secondary to the head injury or need for any of the following procedures within seven days: craniotomy, elevation of skull fracture, monitoring of intracranial pressure or insertion of an endotracheal tube for the treatment of head injury. Brain injury was defined as any acute intracranial finding revealed on CT that was attributable to acute injury, including closed depressed skull fracture (i.e., depressed past the inner table) and pneumocephalus but excluding nondepressed skull fractures and basilar skull fractures.

After the clinical examination, the treating physician determined whether the patient should undergo CT of the head. Staff radiologists at each site, who were blinded to the content of the data collection form, interpreted the CT scans. If the radiologist raised any uncertainty about whether an acute intracranial injury existed, then another radiologist and a neurosurgeon, both also blinded to the content of the data collection form, reviewed the CT scan. If uncertainty remained, the scan was considered negative.

Because not all patients with minor head injury routinely undergo CT at the study sites, we could not ethically mandate universal CT for all patients included in the study. Patients who did not undergo imaging were classified as having no clinically important brain injury if they met all of the following explicit criteria at 14 days, as determined during a structured interview conducted by telephone: headache absent or mild, complaints of memory or concentration problems absent, seizure or focal motor findings absent and return to usual daily activities (feeding, sleeping, school, play and work). A nurse who was unaware of the patient’s predictor clinical variables assessed these criteria. Patients who did not meet these criteria returned for clinical reassessment and CT. Patients were classified as having brain injury solely on the basis of their CT findings. Patients who did not undergo CT and who could not be reached for follow-up were excluded from the final analysis.

Statistical analysis

We assessed the interobserver agreement for each variable using the kappa statistic and 95% confidence intervals (CIs). We did not calculate kappa values for variables created by cut-point (e.g., amnesia ≥ 30 minutes before injury) or for those collected from the medical record (e.g., age). We used univariable analyses with χ² tests (or, for age, the t test) to determine the strength of the association of these dichotomous variables with the primary outcome. We used recursive partitioning to combine variables that we found to be both reliable (kappa coefficient > 0.5) and strongly associated with the outcome measures (p < 0.05) to find the best combinations of predictor variables that were highly sensitive for detecting the outcome measure while achieving the maximum possible specificity. Recursive partitioning creates a branching decision tree by dividing the patient population into subgroups with and without the outcome of interest according to the contents of predictor variables in the subgroup. We used KnowledgeSEEKER version 6.0 Software (Angoss Software International) for the recursive partitioning. Our previous experience suggested that recursive partitioning may be more suitable than logistic regression when the objective is to correctly classify one outcome group at the expense of the other (i.e., where high sensitivity is more important than overall accuracy).

We assessed the derived decision rule by comparing the classification of each patient with his or her actual status for the primary outcomes, which allowed us to estimate, with 95% CIs, the sensitivity and specificity of the rule. The bootstrapping method 25 was used to evaluate the classification performance of the decision rule and to assess overfitting of the model.

Characteristics of the patients

Between July 2001 and November 2005, we enrolled 3866 patients, all of whom underwent complete assessment of the primary outcome (Table 1, Table 2). Assessment of the secondary outcome, brain injury on CT, reflected CT findings for 2043 (52.8%) of the patients. The remaining 1823 (47.2%) patients, who were all discharged directly from the emergency department, underwent the structured telephone interview with a registered nurse at 14 days after discharge for determination of the proxy outcome measure. Of all patients included in the study, 24 (0.6%) underwent a neurologic intervention. CT revealed a brain injury in 159 (4.1%) of the patients. The study sample included 277 children under two years of age, and 23 of these had brain injury revealed by CT. An additional 245 eligible patients were not included in the final analysis because they did not undergo CT or the telephone interview at 14 days to determine the proxy outcome measure. The characteristics of these patients were similar to those who underwent CT or the telephone interview to determine the proxy outcome measure. Another 2178 eligible patients were seen at the study sites but were not enrolled by the treating physicians. The characteristics of these non-enrolled patients were similar to those of patients who were enrolled, including mean age (8.4 v. 9.2 years), rate of arrival by ambulance (35.3% v. 38.2%), transfer from another hospital (15.9% v. 17.2%) and mechanism of injury.

Predictor variables

The variables with the highest associations with brain injury were those found on physical examination: suspected open or depressed fracture; signs of basal skull fracture; large, boggy hematoma of the scalp; and low or deteriorating Glasgow Coma Scale score (Table 3, Table 4). The interobserver agreement (n = 333 cases) for some of the primary variables is given in Table 3.

Combining variables using recursive partitioning analyses yielded a rule based on seven simple questions stratified as representing high risk and medium risk. The resulting rule (Box 1) is called the CATCH rule, for Canadian Assessment of Tomography for Childhood Head injury. Having any one of the four high-risk factors predicting the primary outcome, need for neurologic intervention, had a sensitivity of 100.0% (95% CI 86.2%–100.0%) and a specificity of 70.2% (95% CI 68.8%–71.6%) and would require that 30.2% of patients with minor head injury undergo CT (Table 5). The presence of any one of the four high-risk or three medium-risk factors in the rule would identify any CT-visible brain injury with a sensitivity of 98.1% (95% CI 94.6%–99.4%) and a specificity of 50.1% (95% CI 48.5%–51.7%) and would require that 51.9% of patients with minor head injury undergo CT (Table 6). The three cases of brain injury that were not identified by this rule were an occipital skull fracture with a small pneumocephalus, mild brain edema and a small extra-axial hemorrhage (probably epidural) with a small cerebral contusion. None of these patients required treatment, and none had neurologic sequelae.

• Download figure
• Open in new tab
• Download powerpoint

Box 1: Canadian Assessment of Tomography for Childhood Head injury: the CATCH rule

According to the bootstrapping results, the classification performance of the CATCH rules was accurate across 1000 bootstrapped test sets and was consistent with the estimation from the original data set. For the four high-risk factors for neurologic intervention, the sensitivity was 97.9% (95% CI 97.8%–97.9%) and the specificity was 70.2% (95% CI 70.1%–70.3%). For all seven factors determining the risk for brain injury, the sensitivity was 98.1% (95% CI 98.0%–98.2%) and the specificity was 50.0% (95% CI 50.0%–50.1%).

Limitations

Our study had potential limitations. For ethical reasons, not all enrolled children with minor head injury underwent CT. Nonetheless, we are confident that the children who did not undergo CT received a full assessment for the primary outcome measure, the need for neurologic intervention. All of the study patients who were not examined by CT did undergo a structured and validated telephone interview at 14 days for determination of the proxy outcome measure. Any patient who could not be completely and adequately followed was excluded from the study analyses. Although not all children with minor head injury seen at the study sites during the study period were enrolled in the study, this situation is not out of the ordinary for a clinical study, and we could not determine any systematic difference between the patients who were enrolled and those who were not enrolled. We enrolled relatively few children under two years of age (n = 277), and we identified only 23 cases of brain injury in this group. Although the CATCH rule correctly identified all these cases of brain injury, further prospective study of this subgroup is required, as children under two years of age may have more subtle presentations of head injury than older children. 28^,35

Some may question the significance of relatively small lesions found on the CT scans. In our study we defined as significant any intracranial bleeding or contusion seen on CT, no matter how small, as well as isolated pneumocephalus seen on CT. We consulted several Canadian academic pediatric neurosurgeons, all of whom felt that any abnormality caused by acute trauma is important in children. However, we recognize that there is no consensus among health care professionals on this issue. Additional studies are needed to evaluate the clinical significance of these very small lesions and to evaluate whether finding them justifies the risk associated with exposing the child to radiation. Finally, before it can be used in clinical practice, this derived rule must be prospectively validated to determine its accuracy, its acceptability to clinicians and its impact on care in a new patient population. 36^,37

Conclusion

The CATCH rule is a sensitive, prospectively derived clinical decision rule that has the potential to both standardize the need for CT and reduce the number of CT scans performed for children with minor head injury. Further studies are required to prospectively validate this rule in other pediatric cohorts.