Medical outcome after immediate computed tomography or admission for observation in patients with mild head injury: randomised controlled trial

Definitions

We defined mild head injury as loss of consciousness or amnesia, or both, in patients with normal neurological findings (normal pupillary response, sensibility, reflexes, and response on motor examination) and a Glasgow coma scale (GCS) score of 15 as determined by the attending physician at the patient's arrival in the emergency department after head trauma.19

The results of computed tomography were considered abnormal if there were signs of an acute traumatic injury (haemorrhage, oedema, skull fracture). Scans showing only extracranial injury were not considered pathological nor were findings not correlated with the acute head injury.

A complication was defined as deterioration due to the head injury that necessitated neurosurgical intervention, medical treatment, or intensive care. We also included subsequent readmission because of the head injury.

Eligibility and study setting

Patients aged ≥ 6 with mild head injury who attended for acute care at any of the participating centres were assessed for eligibility (assessment of outcome is difficult in children aged < 6). The eligibility criteria were head trauma within the past 24 hours; confirmed or suspected loss of consciousness or amnesia, or both; normal results on neurological examination; Glasgow coma score of 15, and no associated injuries that required admission. Eligible patients received oral and written information about the study and gave their written consent to participate. For children consent was obtained from a parent or carer. The Swedish national health and pharmaceutical insurance plan covered all patients included in the study.

In Sweden, current practice is to admit patients with mild head injury for observation. Each year the admission rate is around 191/100 000. About a fifth also receive computed tomography.20 Of the 75 emergency departments in Sweden, 39 participated in the study, representing hospitals of all sizes and all parts of the country. A physician and nurse at each participating centre were responsible for the study.

An independent safety committee reviewed data on the main outcome at given intervals and received reports within one week of all deaths and severe complications. The safety committee had the authority to terminate the study early and had decided internally on the rules for termination.

Randomisation

Physicians at the participating emergency departments randomised patients to immediate computed tomography or observation in hospital. Randomisation was carried out by computer generated sequences stratified by centre in blocks of 24 with the restriction that the balance be no worse than in a sequence of a block of six. Assignments were concealed in sequentially numbered, opaque, sealed envelopes kept by the investigators.

A central trial unit continuously checked and monitored all study data. All coded data were checked to ensure consistency with the original documents. Data were stored safely with limited access. We visited participating centres to ensure that randomisation protocols and other study instructions were properly followed.

Interventions

Computed tomography strategy—Patients received computed tomography after randomisation. Scans were reported and interpreted according to local clinical practice. If the scan was interpreted as normal, the patient was discharged home. Attending physicians could admit patients, despite normal findings, for other medical or social reasons. All such events and decisions, including the date and hour they occurred, were noted in the case report form. Local guidelines were used in managing complications.

Observation strategy—Patients were admitted for observation as inpatients according to local guidelines. The attending physician could decide to perform computed tomography if this seemed to be clinically necessary. Results were reported and interpreted according to local clinical practice. All interventions, including the date and hour they occurred, were noted in the case report form. Local guidelines were used in managing complications.

Outcome analysis

We used the extended Glasgow outcome scale (GOS-E), through a postal questionnaire, to assess the outcomes of all randomised patients three months after the injury.²¹²² Patients were asked about their satisfaction with the care received (four alternatives: satisfied, quite satisfied, not satisfied, uncertain). We used reminders and telephone contacts to minimise loss to follow-up. We searched the national population register to rule out complications in patients who could not be traced and hospital care registers of all neurosurgical clinics in Sweden for readmittances and operations.17

End points—Our primary end point was outcome according to a dichotomised extended Glasgow outcome score three months after the injury (8 (fully recovered) v 1-7 (not fully recovered)). To ensure an even distribution among patients with the most severe disabilities, our secondary end points were the same scores dichotomised in six other possible ways (see table 2).

Blinding—Allocation to study group was completely concealed but the patients and study personnel were obviously aware of the allotted strategy. Information used for the Glasgow outcome assessment at three months was self reported by the patients. Two members of the study group and two external reviewers performed a blinded assessment of cause of death and complications. Disagreement was resolved by consensus. During the study, the safety committee had access to unblinded data but had no contact with study participants or the central trial unit.

Sample size

The study was planned as a non-inferiority study. We considered a difference of 5 percentage points to the disadvantage of computed tomography as the limit for acceptability. From previous studies we estimated that about a quarter of the patients would have a Glasgow outcome score 1-7—that is, be less than fully recovered—at the three month follow-up.^23–25 With a sample size of 2000 patients the study would have 80% power to get a one sided 95% confidence interval that excluded the non-inferiority limit—that is, to show that computed tomography is not inferior to observation in hospital if the true probabilities of Glasgow outcome score 1-7 are 0.25 in both groups. We aimed to randomise 2500 patients to allow for losses to follow-up.

Statistical analysis

We analysed all data according to a pre-established analysis plan, which prescribed intention to treat analyses. For the primary analysis we have shown a one sided 95% confidence interval. Thereafter, we used two sided 95% confidence intervals. Although not prescribed in the study plan, we also subjected the Glasgow outcome data to a rank sum test.

Outcome at follow-up

At the three month follow-up we obtained complete data on the extended Glasgow outcome scale for 2523 patients (97.0%) and partial data from 67 (total 2590 (99.6%)). At follow-up, 275 patients (21.4%) in the computed tomography group and 300 (24.2%) in the observation group had not recovered completely. The difference (- 2.8%, one sided 95% confidence interval ≤ 0.03%, two sided confidence interval - 6.1% to ≤ 0.6%) indicates that outcome with computed tomography is not inferior to the outcome with admission for observation. In the secondary analyses we dichotomised the extended Glasgow score in all other possible ways. There were no significant differences between the groups except in one case (1-6 v 7-8), when the computed tomography method was superior (table 2).

As an alternative to the seven analyses of dichotomised data, we also performed a rank sum test. The outcome in the computed tomography group was slightly better than in the observation group (P = 0.062, two sided). The worst outcomes (1-4; death to severe disability) were similarly distributed in the two groups (4% v 4.5%). Two people in the computed tomography group and one in the observation group died as a probable or possible result of the head injury (0.2% v 0.1%). There were four (0.3%) non-fatal complications in the computed tomography group and seven (0.5%) in the observation group (table 3). Tables 4 and 5 describe cases in which patients died or underwent neurosurgical procedures. All three patients in the observation group who needed surgery had a considerable delay in diagnosis and treatment (between 43-74 days after the trauma). Although two of these patients completely recovered, earlier detection of the lesions would have been preferable in each case and possibly could have made an important difference in the one with a poor outcome.

In addition to the prespecified intention to treat analyses, we also performed per protocol analyses excluding patients who did not receive the allocated treatments as well as those who, according to inclusion and exclusion criteria, should not have been randomised. The main result was not changed: 21.3% examined with computed tomography had not recovered completely compared with 24.2% in the observation group. Similar marginal changes occurred in all other analyses.

Among the 1292 patients in the computed tomography group who underwent imaging, 6.3% had abnormal findings (table 6). On average, patients randomised to computed tomography were scanned 5.2 hours after the trauma. Nearly 28% (309/1118) of these investigations were done within three hours of the trauma. None of the patients with normal findings later developed a complication that required admission to hospital or surgery.

Patients' satisfaction was similar in both groups: 92.5% in the computed tomography group and 93.8% in the observation group were satisfied or quite satisfied with the care they had received.

Exceptions to follow-up

Only three patients were completely lost to follow-up. They were foreign citizens and could not be reached despite considerable efforts. None had complications in the acute phase, and all were well at discharge. Seven patients, all well at discharge, withdrew consent to take further part in the study. Nine patients had mild head injury twice and were randomised twice. For seven of them, the second injury occurred more than three months after the first, and these patients were counted as 14 cases. For two patients, however, the two events occurred within less than three months, and we included only the second randomisation in the analyses. Sixty five patients could not be included in the main analysis because we had only partial information on extended Glasgow score: 59 patients could not be reached by regular follow-up (all were registered as alive in the national population register and had no further care in hospital for head trauma within three months), and six submitted incomplete follow-up data and could not be reached to complete the assessment. In 26 cases, the randomised patients were not fully eligible (10 in the computed tomography group and 16 in the observation group). All patients were included in the intention to treat analysis (2590, 99.5%, (figure).

Exceptions to protocol

Two causes contributed to the difference in group sizes (30 cases fewer in the observation group). An error by the statistician in the preparation of the randomisation sequence for one of the centres explains a difference of 12. Furthermore, when all unused envelopes were returned after the end of the study, 55 were missing: 20 from the computed tomography group and 35 from the observation group. The difference was possibly because patients in the observation group are often transferred from the emergency department to an inpatient ward, thereby increasing the risk that the case report form becomes hidden in the case notes. Even if we add 20 not completely recovered patients to the computed tomography group and 35 completely recovered to the observation group, the computed tomography strategy is still not inferior in the sense we have defined.

Wider relevance of findings

To assess generalisability of the findings, we compared the baseline characteristics of the randomised patients with nationwide data from all patients admitted to hospital with mild head injury in Sweden.28 We found no significant difference in age and sex and similar numbers of patients were treated in the various categories of hospital. Furthermore, participating hospitals were from different categories and from different parts of the country. Also, the pragmatic design and large size of the study showed the feasibility of the computed tomography strategy in routine health care.

We also compared our patients and their outcomes with those in other studies of mild head injury (non-randomised clinical trials). Outcome according to the extended Glasgow score at three months' follow-up was consistent with results from previous studies, although direct comparisons are difficult because of differences between the patients.²⁴²⁵²⁹ Our rates of mortality, complications, and abnormal findings on computed tomography were similar to those in recent systematic reviews of mild head injury.³⁰³¹ We therefore believe that the patients in our study were representative of patients with mild head injury, giving the results a high degree of generalisability.

In some hospitals, access to computed tomography is restricted out of office hours. Transportation difficulties and other reasons, such as elderly patients living alone, can also sometimes make observation in hospital the preferred strategy.

Risks of computed tomography

The possible risk of cancer and cognitive dysfunction in adulthood from the low dose radiation with computed tomography has been discussed.³²³³ All use of radiation in diagnostics should be guided by caution, with careful balancing of the risks and benefits. With strict adherence to the definition of mild head injury (including loss of consciousness or amnesia), the benefits probably outweigh the risk of occasional computed tomography.³⁴³⁵ Obviously, efforts should always be made to minimise unnecessary exposure to radiation without compromising image quality.³⁶³⁷

Some guidelines and clinical directives specify the number of hours after the trauma after which it is safe to discharge patients with mild head injury and normal findings on computed tomography.⁷³⁸ There is no evidence in the literature to support such a recommendation.39 Neither do our present results point towards the need for such a limit.