Results

Table 1 shows the age, sex, duration of stay in hospital, and mortality as a function of diagnostic group. The age at injury increased with the severity of the diagnosis, as did the proportion of males and, as would be expected, the duration of stay in hospital. Similarly, overall mortality was highest among the lesion group as was the percentage of suicide cases. It is noteworthy that, for each diagnosis, the proportion of males among the suicide cases was higher than among all cases. The median times from injury to suicide were about 3–3.5 years for all diagnostic groups and they did not differ significantly (F(2,892)=0.67, p>0.2).

The SMRs for suicide as a function of diagnosis and sex are shown in table 2. The ratios were increased above population rates for all three diagnoses, being roughly tripled for the concussion and fracture groups and quadrupled for the lesion group. The ratios for females were higher than for males in all three groups. All SMR values were highly significant and the 95% confidence intervals (95% CIs) fell below 2.0 only for a single subgroup—males with cranial fractures.

Table 3 shows the SMR values as a function of age at injury. It is noticeable that for all three groups, the ratios were lower for patients injured before the age of 21 or after the age of 60, than among patients injured at ages between 21 and 60. The ratios were highest among patients with lesions injured in that age range, being about five times the population rates. With a single exception SMRs in all groups and at all ages had 95% CIs above unity, the exception being patients with lesions injured before the age of 21, among whom there were only three suicide cases.

Among all 145 440 patients, 8684 (6%) also had a psychiatric diagnosis. In most patients (n=7421; 5%) this concerned substance misuse (ICD 291 alcoholic psychosis, ICD 303 alcoholism, or ICD 304 drug dependence), the most common single diagnosis being inebriation at the time of admission (ICD 303.90; n=5165; 3.6%). Substance misuse was more common among the lesion group (n=681; 5.8%) than among the fracture group (n=312; 4.1%) or the concussion group (n=6428; 5.1%) (χ²(2)=26.3; p<0.001). However, table 4 shows that the presence of a codiagnosis of substance misuse markedly increased the SMRs for suicide in all three groups. The highest SMR, greater than 7, was that for the lesion group, although the small numbers in the fracture and lesion groups were necessarily associated with large 95% CIs.

As shown in table 5 there seemed, among the lesion group, to be a monotonically increasing risk of suicide in proportion to duration of stay in hospital. However, the wide 95% CIs associated with the SMRs makes this interpretation tenuous.

To investigate the time course of suicide after injury we performed a Cox regression analysis in which sex, age at injury, and the presence of a substance misuse codiagnosis were entered in a first block before the inclusion of diagnostic group. In view of its apparently non-linear function, age at injury was entered as a category variable using the groupings shown in table 3. The analysis showed that diagnostic group made a significant improvement (p=0.004) to the overall model after the inclusion of sex, age at injury, and substance misuse. The lesion group had a significantly higher rate of suicides than both the concussion group (hazard ratio=1.42, 95% CI 1.15–1.75) and the fracture group (hazard ratio=1.50, 95%CI 1.06–2.13). The influence of sex, age at injury, and the presence of substance misuse on these ratios was not very substantial; before their inclusion in the analysis the hazard ratio for the lesion group relative to the concussion group was 1.67 ( 95%CI 1.36–2.06) and for the lesion group relative to the fracture group 1.66 (95%CI 1.17–2.35).

The survival functions (specific for suicide) for the concussion and fracture groups were found to be virtually identical. Figure 1illustrates the contrasting functions for the concussion and lesion groups across the maximum of 15 years of follow up; both functions, although differing in slope, were essentially linear. Thus, there was no strong evidence for a specific risk period for suicide after injury in any of the groups.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Cumulative proportional hazards for suicide in the concussion and lesion groups.

The possibility exists, in cases of suicide after a traumatic brain injury, that the injury itself might have been the result of an unsuccessful suicide attempt. We have been able to examine this possibility, as the hospitalisation register has included, from 1987 onwards, a coding of whether the injury was self inflicted. Among all three diagnosis groups combined 48 298 cases occurred in 1987 or later, and of these only 369 (0.8%) were recorded as having been self inflicted injuries. Seven of these 369 (1.9%) were subsequent suicides whereas there were 134 suicides among the remaining 47 929 non-self inflicted cases (0.3%). Although this association is highly significant (odds ratio=6.9, 95%CI 3.2–14.8) it is clear that for only a numerically negligible proportion of cases within our sample could the brain injury have been the result of a failed suicide attempt.

Discussion

The present study, although large scale and nationally representative, has some limitations. Firstly, the hospitalisation register is not completely accurate, particularly effecting the diagnosis grouping we have used here. Thus, from a subsample within the present study, involving review of the original hospital records, we have estimated that about 10% of coded diagnoses of intracranial lesions may be incorrect, especially involving overrecording of the ICD-8 854 diagnosis for “non-specific intra-cranial lesions” where a concussion diagnosis (ICD-8 850) would have been appropriate. Conversely, there may also be some underrecording of cerebral lesions. In a smaller earlier study14 up to one third of patients who had a post-traumatic amnesia exceeding 1 week were found to be recorded with only concussion or cranial fracture diagnoses. The register itself contains no data concerning depth or duration of coma, nor duration of post-traumatic amnesia (we are currently conducting an intensive study of a representative sample of the patients with brain injury in which we are gathering such data directly from the hospital records). The lack of post-traumatic amnesia data is especially unfortunate considering the high predictive value of that measure for outcome after traumatic brain injury.14

Another issue which needs to be considered is the possible underreporting of suicides. Social and religious disapprobation of suicide is common in many cultures.7 20 Although this is perhaps less true of Denmark than elsewhere, it remains the case that within our population some suicides—for example, from drug overdoses, drowning, and falls from high places—could have been erroneously recorded as accidental deaths. It should also be emphasised that it has not been possible for us within the present study to investigate other potentially related factors such as unemployment and marital status.

Within the constraints of these considerations, however, our results do seem to permit some clear conclusions about suicide after a traumatic brain injury, given survival through the acute phase.

Perhaps the most striking finding has been the increased rate of suicides among patients who have had no more than a concussion, leading typically to only a single day in hospital for observation, or a cranial fracture with no identified cerebral lesion. Our findings for the concussion and fracture groups have more similarities than differences and we consider them together, before discussing the findings for the lesion group.

The increased suicide rates for the concussions and fracture groups, by a factor of two or more relative to the general population, were found to a greater or lesser extent for both sexes and for all age groups. Particularly in the case of concussions there can hardly be any direct link between having sustained the injury and the subsequent suicide, and the association must have arisen from concomitant and perhaps premorbid characteristics.

Achté et al 8 noted that some characteristics, such as “broken homes” during childhood and alcohol misuse, occurred with greater frequency among their suicide patients with brain injury. They commented that “there is reason to suppose that a considerable proportion of those who later committed suicide had even prior to the injury manifested deviant psychic features”. This is certainly supported by our finding of the substantial contribution towards suicide risk of a codiagnosis of substance misuse, in most cases no more than a registration that the patient was inebriated at admission. The associations of substance misuse with both traumatic brain injury and with suicide are well documented.7 21

Mortality rates were greater for females than for males, this despite the fact that there were relatively fewer suicides among females than among males in our population. Similarly Hartkoppet al 22 found increased mortality rates for females in a Danish cohort of patients with spinal injury. They speculated that this may be because “women build their self image much more on their physical attractiveness than men” and therefore find it more difficult to cope with the consequences of a spinal cord injury. This explanation will certainly, however, not account for the relatively greater mortality rate among females with a concussion. We suspect that concussion may be more strongly related to psychosocial disadvantage among women than among men, and thereby be more strongly predictive of suicide.

The effect of age may also be related to psychosocial disadvantage. Mortality rates for suicide are greater among patients who have traumatic brain injuries sustained at ages between 20 and 60 than those having such injuries at younger or older ages. We have elsewhere shown that concussions sustained in adolescents are less associated with premorbid cognitive deficit than concussions sustained by young adults,23 a result which we interpreted to suggest that the injuries in the first group were more often associated with play and sports and less associated with alcohol, violence, and other psychosocial factors. The present results are consistent with this view. Similarly milder traumatic brain injuries among older people may also be relatively more often the result of physical frailties, accident proneness, and sheer ill luck—comparatively unrelated to psychosocial disadvantage.

For the lesion group, it is noticeable that mortality rates for suicide were consistently higher than for the concussion and fracture groups, for both sexes and at all ages. To the extent that premorbid psychosocial factors are equated across the three groups, then the higher rate of suicide for the lesion group implicates the direct consequences of the injury itself on the suicide. This interpretation is supported by the clinical evidence stemming from smaller studies of similarly severe traumatic brain injuries.11 12

That the suicide rates within the lesion group tended to increase with the severity of the injury, as gauged by the duration of stay in hospital, is also consistent with this interpretation. We have elsewhere reported duration of hospital stay after traumatic brain injury to correlate consistently with poorer long outcome in terms of education, employment, and disability.24 25 It might have been expected that suicide rate would decline at some level of severity where the person with brain injury would be lacking the cognitive and physical apparatus for the act. That we could not identify such a level within our data is probably due to the limitations of duration of stay in hospital as a measure of severity of injury relative to the generally much more sensitive post-traumatic amnesia.14

From their clinical evidence both Tate et al 12 and Klonoff and Lage11 have suggested that, where relatively severe traumatic brain injury results in subsequent suicide, it is psychologically reactive to the identity crises and generalised social disruption experienced in such cases, rather than any more specific neuropsychological mechanism such as greater impulsivity consequent on frontal lobe injury. Similarly, in their larger study, Achté et al 8 found suicide to be related to the severity of the injury but not its localisation.

Although our evidence supports the view that suicide is indeed more common after a relatively severe traumatic brain injury we could find no evidence of a risk period. It is often suggested that the full impact of the consequences of such an injury is not felt immediately but only after some years of experiencing disappointments and fiascos, of ultimately confronting the fact that improvement, characteristic of early postinjury phases, will not continue to the point where full premorbid functioning returns. However, our present evidence suggests that the risk of suicide is a constant, continuing for at least the maximum of 15 years follow up we are reporting here. It is similarly noteworthy that Achté et al 8found the suicide risk actually to increase to a maximum of 15 years after injury, an interval which, as Tate et al 12 note, was probably more related to the chronological age of his sample 15 years after injury ( people then in their 40s and 50s) rather than the time interval itself.

In summary, we have found increased rates of suicide among patients with relatively severe traumatic brain injury. Present evidence suggests that no small degree of this increase may be due to premorbid psychosocial factors, as the increase is also present in patients with concussion and cranial fractures. The absolute magnitude of the rates is not of alarming proportions, being about 1% over a 15 year period. Nonetheless, as Tate et al 12remark, “the small magnitude in no way corresponds to the seriousness of the problem” and awareness of a suicide risk should be present in the assessment of any traumatic brain injury.