Ninety four studies were included in the review: 93 diagnostic accuracy studies and one RCT. Study quality was variable. Patient spectrum was fulfilled by only two of the clinical decision rules studies and one of the individual characteristics studies. Less than 50% of studies fulfilled QUADAS items on reporting of uninterpretable results, avoidance of clinical review bias, blinding of index test and reference standard, avoidance of differential and partial verification bias and use of an appropriate reference standard. Biomarker studies were generally of better quality, but none fulfilled the patient spectrum, uninterpretable results or withdrawals criteria; other criteria were fulfilled by at least 70% of studies. The RCT was considered to have a moderate risk of bias.
Clinical decision rules (adults): Nineteen studies evaluated 25 decision rules. Eleven rules were evaluated in more than one data set and a further rule was evaluated within two cohorts in the same data set. Nine rules had two forms: one to identify patients at medium risk and one to identify patients at high risk.
The Canadian CT Head Rule (CCHR) was validated in six studies. Using the high risk threshold it had a sensitivity of 99% to 100% and a specificity of 48% to 77% for neurosurgical injury; the high risk threshold was not evaluated for the detection of intracranial injury. Using the high or medium risk threshold sensitivity ranged from 99% to 100% for neurosurgical injury and from 80% to 100% for intracranial injury. Corresponding specificities ranged from 37% to 48% and 39% to 50%. Sensitivity for intracranial injury was lower in studies in which all patients had CT compared to those in which some patients received follow-up (80% to 86% compared to 98% to 100%).
The New Orleans Criteria (NOC) rule was validated in four studies. It had a sensitivity of 99% to 100% for neurosurgical lesion and 95% to 100% for intracranial lesions, but specificity was poor at 3% to 33% so that in most cohorts all patients would have undergone a CT scan.
National Institute for Health and Clinical Excellence (NICE) guidelines were based on CCHR and had a sensitivity of 88% to 98% for neurosurgical injury and 67% to 99% for any injury and specificity that ranged from 29% to 70% (three studies).
Other clinical prediction rules were evaluated in two studies or fewer; full results were reported.
Clinical decision rules (children): Fourteen studies evaluated 15 decision rules. Most rules were evaluated in two studies or fewer. The University of California-Davis (UCD) rule was evaluated in three studies. Sensitivity ranged from 91% to 100%. Specificity ranged from 12% to 43%.
Individual clinical characteristics: There were 42 studies in adults and 29 studies in children. No individual clinical or radiological characteristics were found to have sufficient accuracy to be able to rule in or rule out intracranial injury.
Depressed, basal or radiological skull fracture, post-traumatic seizure (adults only) and focal neurological deficit (children only) increased the likelihood of intracranial injury (LR+>10), but skull radiology has been large replaced by CT scanning. Focal neurological deficit (adults only), persistent vomiting (adults only), decrease in Glasgow Coma Scale (adults only), coagulopathy (children only), post-traumatic seizure (children only), and previous neurosurgery were associated with an increased risk of intracranial injury (LR+ range 5 to 10).
Fall from a height (adults only), coagulopathy (adults only), chronic alcohol use (adults only), age more than 60 years, pedestrian motor vehicle accident, cycle motor vehicle accident (children only), any seizure, undefined vomiting, retrograde or anterograde amnesia, Glasgow Coma Scale less than 14 or 15 and severe or persistent headache (children only) were associated with a moderate increase in the risk of intracranial injury (LR+ 2 to 5). Loss of consciousness (adults only) and headache were of little value in diagnosing intracranial injury. Diagnosis of neurosurgical injury was assessed in only a small number of studies.
Biomarkers: There were 11 studies in adults and one study in children.
The protein S100 calcium binding protein-B biomarker was found to have excellent sensitivity with a pooled estimate of 97% (95% CI 94% to 99%) but specificity was less good (43%, 95% CI 31% to 54%) based on nine studies in adults. Estimates from the single study in children were similar with a sensitivity of 100% (95% CI 92% to 100%) and specificity of 42% (95% CI 38% to 43%).
Two studies assessed neuron-specific enolase (NSE) and suggested that it was a poor marker for predicting intracranial injury. A single study assessed various other biomarkers: adrenaline and dopamine were suggested as having good accuracy for ruling out head injury but the findings had not been validated.
Diagnostic management strategies: There was one RCT (2,062 participants). This study randomised patients aged at least six years with minor head injury to immediate CT or observation in hospital. At three months there was no significant difference between the treatment groups in terms of the proportion of patients who had recovered completely; mortality and severe loss of function were also similar between the groups. The authors concluded that use of CT in patients with minor head injury was feasible and led to similar outcomes compared to observation in hospital.