METHOD

Criteria for selecting studies

Full papers and not abstracts. Case-control or nested case-control studies on GCS 15 patients (adults or adults plus children) with LOC or PTA, all of whom underwent a head tomogram (abbreviated form CT). The indication for head tomography being LOC or PTA. Documentation of one or more symptoms such that a 2×2 contingency table could be constructed. Five symptoms were defined for the purpose of the study; (1) headache, (2) nausea, (3) vomiting, (4) blurred vision, (5) dizziness. The control group being GCS 15 with LOC/PTA but without the symptom variable.

Search strategy

The following databases were searched Medline, and PubMed, January 1990 to February 2001. The following words or phrases were used in the search: minor head injury OR mild head injury OR Glasgow Coma Scale OR traumatic brain injury. Cross referencing from relevant articles and hand searching of journals was also performed. The following journals were hand searched: Journal of Trauma, Critical Care Medicine, Annals of Emergency Medicine, Emergency Medicine Journal, Neurosurgery, and the British Journal of Neurosurgery. A more focused search was performed on the Medline database using the OVID interface; Craniocerebral trauma/cl, di, ra [Classification, Diagnosis, Radiography] LIMIT to human AND abstracts.

Review procedures

A decision to include or exclude a study was made by the author of this monograph using the inclusion criteria defined above. The suitability of the papers for inclusion was also based upon whether all patients in the cohort underwent head tomography, clearly defined definitions, and the availability of data to enable the odds ratios to be calculated by the author of this monograph.

Statistical methods

A pooled estimate (random effect model) for the common odds ratio was performed using the Mantel-Haenszel test. Before calculating the common odds ratio the Breslow-Day test for homogeneity of the odds ratio was performed. The analysis was performed using SAS version 6.12 (SAS Institute, Inc, Cary, North Carolina).

RESULTS

The Medline search using the OVID interface produced 243 full papers of which two fulfilled the inclusion criteria (table 1). A third paper that fulfilled the inclusion criteria was identified by cross referencing (Jeret et al⁹, table 1). The study by Jeret et al⁹ was a cohort study of 712 consecutive patients over the age of 17 years with PTA or LOC. The study by Miller et al¹⁰ was a cohort study of 2143 consecutive patients of all ages with PTA or LOC. The study by Haydel et al¹¹ was a cohort study of 520 consecutive patients who were at least 3 years of age with GCS 15 and LOC. The study design, the CT abnormality rate and the neurosurgical intervention rate are shown in table 1. The likelihood ratio and the post-test probability (for an abnormal head scan) for each of the symptom variables are shown in table 2. The variables available for analysis (by author) are shown in tables 3. The Breslow-Day test for homogeneity of the odds ratio (for an abnormal head scan) showed that only the headache variable had any significant discordance (table 4). This was attributable to the large difference in the odds ratio between the Miller study and the other two studies (table 5). The Breslow-Day test for the homogeneity of the odds ratio was recalculated for the Jeret and Miller studies only. This result showed no significant discordance between the odds ratios of the two studies. An odds ratio was calculated for the symptom severe headache (Miller study only) and also a common odds ratio for the symptom headache (Jeret and Miller study). Figure 1 shows that only the severe headache, nausea, and vomiting variables resulted in a statistically significant increase in the odds ratios for an abnormal head scan.

DISCUSSION

The three studies included in the meta-analysis showed similar neurosurgical intervention rates but some variation in the overall incidence of abnormal CT scans. Variations in thresholds for inclusion may explain some of the variation in the CT abnormality rates. For instance the Haydel study¹¹ included only patients with LOC in contrast with the Jeret⁹ and Miller¹¹ study that included patients with both LOC and PTA. The main difference between the three studies however was that the Jeret⁹ study included patients who were 18 years or over, while the Miller¹⁰ and Haydel¹¹ studies both included children.

A pooled (random effects model) rather than a fixed effect model was used to calculate the common odds ratio because the three studies were similar in size and the Breslow-Day test showed no significant discordance between the odds ratios. All studies in the meta-analysis were nested case-control type studies and in all three studies some patients had more than one symptom. A fundamental assumption being made is that the distribution of the symptom variables is the same in both groups. A true nested case-control meta-analysis (that is, patients with LOC/PTA and no symptoms (control group) compared with those with LOC/PTA and only one symptom (test group)) was not undertaken by the author because the data were not available in any of the monographs. There is no reason to believe that publication bias is likely to be a significant problem in this meta-analysis although inclusion of full articles only might have resulted in loss of some potentially useful additional data.

None of the three studies distinguished between patients with a single vomit and patients with repeated vomiting. Data from a cohort study of 10 000 head injured patients by Arienta et al¹² showed that patients with a single vomit had a much lower risk of an abnormal CT scan when compared with patients with repeated episodes of vomiting (3% versus 43% (χ² 26.25, p=0.001, the χ² test was calculated by the author of this monograph).

Despite some of the limitations previously discussed the results of this study suggest that the variables severe headache, nausea, and vomiting may be important risk factors in predicting abnormal CT scan in GCS 15 patients with LOC/PTA. Several other authors have looked at the relation between head injury symptoms and abnormal CT rates. Nagy et al¹³ performed a prospective study of 1170 patients with GCS 15 and LOC. A comparison of patients with normal scans to those with abnormal scans showed that the abnormal group was more likely to have headache, nausea, vomiting, dizziness, or seizures (12.5% versus 2.1%; p=0.002; Fisher’s exact χ² test). These four symptoms were analysed as a group rather than in isolation. Hsiang et al⁷ prospectively evaluated 1177 GCS 15 patients and also performed a subset analysis of patients with vomiting. Their results showed that there was no difference between the vomiting group and the non-vomiting group with regard to CT abnormality rates. Unfortunately the study was flawed because only 58% of the GCS 15 patients underwent CT. Lee et al¹⁴ also prospectively evaluated 1812 GCS 15 patients. A CT scan was only performed if the patient deteriorated neurologically. Lee et al¹⁴ found that the relative risk of neurological deterioration was 15.4 times greater in patients with vomiting and 7.3 times greater in patients with headache.

The presence of LOC or PTA is currently used as a means of triaging GCS 15 patients either for inpatient observation or head tomography. Stein et al¹⁵ was one of the first authors to use LOC/PTA as a means of triaging GCS 15 patients based upon a retrospective study of 1538 patients. Miller et al¹⁶ performed a similar study to Stein et al¹⁵ and found that the overall CT abnormality rate was 6.1%, compared with 13.2% in the study by Stein et al.¹⁵ If patients with signs or symptoms of a skull fracture were excluded the CT abnormality rate was only 3% in the Miller study. Miller et al¹⁶ questioned the validity of performing head tomography on all patients in category GCS 15 with only LOC/PTA because of the low rate of abnormal CT scans and zero neurosurgical intervention rate in their study. A follow up study ¹⁰ on an extended version of the same dataset identified four risk factors for abnormal CT scans in GCS 15 patients. The four risk factors being severe headache, nausea, vomiting, and evidence of a depressed skull fracture on clinical examination. Using the same dataset the four risk factors resulted in a sensitivity rate of 65% and a specificity rate of 63%. Haydel et al¹¹ in their prospective study identified seven risk factors for predicting abnormal CT scans in GCS 15 patients (short-term memory deficit, drug, or alcohol intoxication, age >60 years, seizure, headache, vomiting, and evidence of trauma above the clavicles). These seven risk factors were applied to a second dataset of 909 GCS 15 patients. All patients with a positive CT scan (n=57) had a least one of the risk factors. Stiel et al^17,18 have also developed a clinical decision rule for head tomography in patients with minor head injury (defined as GCS 13–15). They also identified repeated vomiting as being an important risk factor. Arienta et al¹² using multivariate analysis identified a variety of signs, symptoms, and conditions as risk factors for identifying abnormal CT scans in head injured patients. They proposed a new classification for head injured patients based upon the combination of identified risk factors and the GCS. Their classification system (groups α, β, γ, δ) although evidenced based does have some limitations. A GCS 15 patient with a generalised headache is not classifiable by their system. Their management strategy based upon this classification system may result in a large number of unnecessary head scans because all patients with LOC, PTA, or a single vomit, undergo head tomography.

The results of this meta-analysis and data from the Arienta study¹² provide a basic framework for stratifying GCS category 15 patients into four groups: GCS 15a, GCS 15b, GCS 15c, and GCS 15d based upon symptom variables previously analysed. The fulcrum for this stratification is based around the odds ratio for a symptom variable in GCS 15 patients with LOC/PTA. Category GCS 15a represents head injured patients who have been asymptomatic up to the time of presentation (very small risk). Head injured patients with (or who have experienced) symptoms of dizziness, blurred vision, or mild headache are classified as GCS 15b (low risk—that is, no significant increase in the odds ratio compared with the GCS 15 with LOC/PTA group). Head injured patients with LOC or PTA are classified as GCS 15c (intermediate risk). Patients with severe headache, persistent nausea, or more than one vomit are classified as GCS 15d (high risk category for abnormal head tomography; significant increase in the odds ratio for an abnormal scan).

An initial management algorithm based upon this framework has been developed (fig 2). The management algorithm also incorporates other non-symptom high risk factors. Patients with high risk factors (fig 2) automatically fall into GCS 15d category. More than one vomit was used as a “high risk factor” based upon the data from the Arienta study¹² as discussed previously. It is not clear at the present time what the relative risk (or odds ratio) is for an abnormal CT scan in patients with a single vomit. The definition of severe headache is the same one as used by Miller et al¹⁰—that is, defined on patient questioning by comparison to a mild or moderate headache. The use of the word “high risk” is a relative one in view of the fact that less than 1% of GCS 15 patients in all three studies required neurosurgical intervention. This fact is also apparent from table 2, which shows that the post-test probability for an abnormal head CT is still low even for “high risk” symptom variables. Rimel et al³ in 1981 found from a cohort study of patients with minor head injury (defined as GCS 13–15) that 59% had memory problems three months after injury. A more recent study by Thornhill et al¹⁹ has also shown that a significant proportion (47%) of patients with minor head injuries (defined as GCS 13–15) had significant disability 12 months after injury.

The question of age (eldery patients) as a risk factor remains unclear. A retrospective study was performed by Nagurney et al²⁰ on 1649 GCS 15 patients all of whom underwent head tomography. Twenty per cent in the eldery group (> 60 years) had an abnormal CT scan compared with 13% in the non-eldery group. The difference was statistically significant. The Nagurney study²⁰ used several criteria for head tomography and so it is not possible to calculate an odds ratio for eldery patients with GCS 15 and LOC/PTA. Age has been advocated as a “triage” risk factor for head tomography in GCS 15 patients by some authors (Haydel et al¹¹) but not by others (Arienta et al¹², Miller et al¹⁰). The Haydel study¹¹ identified increased age as a risk factor using recursive partitioning analysis. The odds ratio from the Haydel study¹¹ for eldery patients suggests however that it may not be a statistically significant risk factor (odds ratio=2.498; 95%CI 0.972 to 6.371; authors’ calculation). Thus although there is some evidence to show that elderly patients are more likely to have abnormal head tomograms after head trauma there is no strong evidence at the present time to suggest that the threshold for head tomography should be any lower in this group of patients. Similarly coagulation problems (for example, warfarin) have also been advocated as a triage risk factor for head tomography by some authors (Arienta et al¹²) but not by others (Miller et al,¹⁰ Haydel et al¹¹). Garra et al²¹ retrospectively reviewed 39 GCS 15 anticoagulated patients without LOC or neurological abnormalities all of whom underwent head tomography. None of the 39 patients was found to have an abnormal tomogram. Garra et al²¹ as a result of this study and a review of paediatric head trauma cases in haemophiliacs concluded that in the absence of LOC or neurological abnormalities head tomography is probably not warranted in this group of patients. There was no discussion however regarding patients with severe headache or vomiting. Brown et al²² retrospectively reviewed 124 GCS 15 anticoagulated patients without LOC all of whom underwent head tomography, 6% of the cases had positive scans. This study also failed to record the number patients with severe headache or vomiting. At the present time evidence that demonstrates the relation between head tomogram abnormality rates and patient symptoms is currently incomplete in this heterogeneous group of patients.

Although two of the studies in this meta-analysis included children the management algorithm has been developed for adults and not for children. The results of several studies^23–27 suggest that headache, nausea, and vomiting are not predictive of intracranial abnormalities in this age group.

CONCLUSION

The common odds ratio based model for the subclassification and management of GCS 15 patients has yet to be fully evaluated. The advantage of this model over the triage risk factor model developed by Miller et al¹⁰ and Haydel et al¹¹ is that it provides a better framework for managing low and intermediate risk patients.

Contributors

Anne McGuinness discussed the core ideas and edited the final draft. The guarantors for the paper are Mr John Batchelor and Miss Anne McGuiness.