Abstract
In this paper we will summarize our two recent publications that down-rate the lifetime suicide risk in major depression from the accepted figure of 15% to less than a quarter of this amount [1, 2]. The figure of 15% was first reported by Guze and Robins in 1970 [3] and has been accepted and propagated by prominent psychiatry teaching texts. We will examine the clinical and research applications of our work in re-rating this figure and speculate on the possible implications of our findings with particular attention to the significance of gender in the diagnostic construct of major depression.
A mathematical impossibility
Our research began with a simple observation of an apparent inconsistency: if around 17% of the entire population will suffer from major depression and 15% of them will kill themselves, suicide should be a relatively frequent event. This was our null hypothesis which we then set about disproving. We were able to demonstrate that it was mathematically impossible for this to be true given extant population suicide rates, even allowing for a significant level of official underreporting. To this end we increased the official total suicide rates by 40% to create a population suicide rate of around 17/100 000 annually [1].
An algorithm that made age-specific calculations of suicide and the prevalence of major depression was developed. It allowed for deaths in each age group from causes other than suicide and applied the calculations to the entire population of the USA for the years in question (1983 and 1994). Readers interested in the method of calculation and the parameters used are referred to our two previous papers for details [1, 2].
Down-rating the overall suicide risk
The algorithm was applied to data sets from the USA as this was the only country that had the complete data necessary for the calculations. Using the entire population of the USA gave our calculations significant power. The lifetime suicide rate in major depression was found to be 3.4%: less than one-quarter of the accepted figure of 15%. However, for reasons that will become apparent, even this figure does not accurately reflect the situation when gender factors are considered.
A simple, although less precise, way of understanding our work is the reverse calculation using the hitherto accepted lifetime suicide risk of 15%. If we start with a cohort of 100 000 15-year-olds and follow them for 60 years until age 75, we would expect that around 17 100 of them would develop major depression [4]. If 15% of them were to commit suicide, then, on an average for each of the 60 years, we would expect 43 suicides. However, major depression comprises only between 13 and 70% of all suicides [5]. Even if major depression comprises 70% of all suicides then an annual rate of suicide from all causes of at least 61/100 000 is required to give this expected rate of 43 (per 100 000 per year). This calculated suicide rate of 61/100 000 is around four to five times the actual figure.
By way of example, using a lower figure of suicides attributable to primary major depression (many studies fail to make this distinction), such as that of 31%, as found in the precise Finnish research [6], would require an annual population suicide rate of 138/100 000 to produce the 43/100 000.
In selecting statistics for our calculations we have used figures at the end of the reported spectrums that would make it as difficult as possible for us to disprove our null hypothesis (e.g. the 70% figure giving the 61/100 000 figure rather than the 31% figure giving 138/100 000). Approaching this research from a purely mathematical perspective is novel and highlights the immutable interrelationship between the various statistics. The validity of our data hinges upon a careful evaluation of the validity of underlying data sets and their research methodologies, rather than upon the mathematics. Peer-reviewed publication and presentation of our results in international fora have not identified any methodological flaws in this approach or concerns about the data used. Even the first author of the seminal paper, Samuel Guze, has generously acknowledged the validity of our findings [personal communication 1998]. However, given these considerations, rather than suggesting that our figures apply to the world, we can very simply argue that for populations where the actual suicide rate is 17/100 000 p.a. or less and the lifetime prevalence of depression is 17.1% or more, then the lifetime suicide risk in major depression has to be of the order of 3.4% or less.
Sex and age determinants of suicide risk
Our second paper updated the first both in methodology and the underlying data sets (1994 vs 1983) and re-evaluated the data by incorporating gender and age considerations. The algorithm was refined by including a gender-based differential up-rating factor to allow for the underreporting of official suicide rates. The findings were not only surprising but they illustrated the deceptive nature of a single figure describing suicide risk for both sexes. Our calculated lifetime suicide risk in those with major depression for men is 7% and for women only 1%. The figure of 3.4% was hiding such highly disparate figures that it failed to do justice in conveying the risk in either sex.
As might be expected the greatest gender difference occurred in male youths (under the age of 25), who were 10 times more likely than women to suicide. The available literature, reviewed in our 1999 paper [2], suggests that the gender difference may be partly explained by considering ‘help-accessing’ behaviour and comorbidity. (‘Help-accessing’ is preferred by the authors over the traditional concept of ‘help-seeking’ as the latter term infers a capacity to actively quest for something, a capacity typically compromised by the illness itself. The concept of ‘access’ places greater emphasis on the need for service providers to make care readily ‘accessible’.) Help-accessing has been repeatedly found to be relatively absent in males when compared with females, denying them not only treatment, but possibly equally importantly, an understanding of their illness that offers hope. Comorbidity, especially for substance abuse, tends to be greater in males and is a growing problem in younger males. One study's findings that we should repeat here, are those by Young et al. who found that the effect of substance abuse (most relevant to men) was so great in predicting suicide that once this variable was controlled for, along with having a young child at home (most relevant to women), differences in predicting suicide disappeared [7].
The fundamental flaws
Because every major textbook quotes a suicide risk in major depression of 15% [1], every good psychiatry trainee and, quite reasonably therefore, any speaker who needs to emphasize the seriousness of major depression as a public health concern, uses this figure too. What is probably the most surprising is that a single paper, that by Guze and Robins [3], could be so uncritically accepted and so widely promulgated. Unfortunately, that research was fundamentally flawed in the first instance and then this error was further multiplied by a change in diagnostic classification exactly a decade later.
Guze and Robins' paper combined the findings from 17 studies dating back to 1937. The majority of studies were reported in the 1940s and 1950s. This was an era when study populations were typically identified among inpatients and treatment was primitive. Not only was the sample population undertreated by modern standards, but the authors made no distinction between bipolar and unipolar cases. The assumption that these two conditions share the same suicide risk has been equally propagated by others whenever the suicide risk of 15% has been subsequently applied to unipolar depression alone. We have discussed these latter biases in our 1997 paper so here we will only expand on the more critical issue of patient sampling in the original studies.
The hospitalization bias
Goldberg and Huxley, referring to an era that largely predated community-based care, found that even then 98% of sufferers of mental disorders were not admitted to hospital [8]. As depressive disorders constitute one of the larger subgroupings of mental disorders, and given recent shifts to community treatment, it is likely that only a minority of sufferers of major depression are admitted. If, however, those who are admitted are representative of the greater group when it comes to suicide risk, then this would not present a significant problem. As it happens, suicidality is one of the primary determinants of hospitalization. Fatefully, the variable being studied – suicidality – actually selected those with the higher risk of suicide for inclusion in the studies! We see this sampling bias as the greatest contributor to the spuriously high reported risk of suicide in major depression.
This critical hospitalization bias was unwittingly replicated in later meta-analyses including an exhaustive review by Goodwin and Jamison published in 1990 [9]. While they observed that most studies are done with hospitalized subjects they only commented on this skewing the data towards the more severely ill. Ignoring the significance of this bias they went on to find an even higher suicide risk of 19%, and again they also failed to distinguish between unipolar/bipolar subtypes. With more refined statistical methods and a higher threshold of data completeness Inskip et al. did a recent meta-analysis on 27 studies and arrived at a figure of 6% lifetime risk in ‘affective disorder’ [10]. We would argue that this figure remains too high because of the failure to identify the hospitalization bias and also, possibly, because it includes sufferers of bipolar disorder. As we have reported, studies that are not affected by the hospitalization bias typically report surprisingly low suicide rates in major depression [1].
A study by Simon and VonKorff of ‘depressive disorder’ in a very large sample of over 35 000, found that suicide risk among sufferers declined from 224 per 100 000 person-years for those with a history of hospitalization to 64 among those who received outpatient specialty mental health treatment and to only 43 among those treated in primary care with antidepressants [11]. While this study was complicated by the inclusion of adjustment disorders and a failure to exclude bipolar cases, it clearly demonstrates that inpatients are not representative of outpatients when it comes to suicide as an outcome.
So Guze and Robins' finding now becomes one that relates, at best (if it can be assumed that unipolar depression has the same risk as bipolar illness) only to sufferers of major depression who have a history of requiring hospital admission.
Reassuringly, this factor of the representativeness of the sample in suicide research is more of a concern with major depression than with schizophrenia or mania. In these latter conditions hospitalization is more often influenced by problematic aspects of the illness other than suicidality. The various considerations relevant to exploring the down-rating of suicide risk are summarized in Table 1 (and elaborated on in greater detail in our first paper [1]).
Possible biases in explaining the Guze and Robins lifetime risk of suicide of 15%
Meta-anlyses
The influential Guze and Robins paper was possibly one of the first of what has become the inevitable metaanalysis paper and clearly illustrates some of the risks inherent in trying to achieve research power through sheer numbers of studies presented. It was Newton (in his letter of 1676 to Oldenburg) who said, ‘For it is not number of experiments, but weight to be regarded; as where one will do what need many?’ Indeed each of the handful of later papers (reviewed in our 1997 paper) that studied unselected, combined inpatient and outpatient populations, found much lower rates of suicide consistent with our calculated figures. Most noteworthy was research by Morrison of 12 500 cases identified from private psychiatric practices followed for over 8 years which found that the rate of suicide for unipolar patients was only 1.7 times the local general population whereas that for bipolar patients was 7.5 times the general population [12].
Not only was the seminal meta-analysis-style publication flawed by studies afflicted by the hospitalization bias, but as mentioned above, two further meta-analyses over almost three decades were repeatedly affected by the same unrecognized factor. Eyesenck has argued that a ‘best-evidence synthesis’ which requires the reviewer to exercise expert judgement (and only include studies designed to effectively minimize biases) may often be preferable to meta-analyses [13]. Thompson and Pocock, having highlighted how the value of a meta-analysis is totally dependent on the lack of bias in its component studies, go on to warn that it is ‘unreasonable to interpret simplistically the numerical result of a meta-analysis’ [14]. A further problem for meta-analyses clearly arises when the selected papers cover different nosological eras and different nosologies.
The latter problem is exemplified in the research by Clark and Goebel-Fabbri who utilized a similar methodology to Guze and Robins in a ‘re-examination’ of their research. In expanding the number of studies involved from 17 to 57 without consideration of the biases we have identified, they added little to this field of knowledge largely as a result of a failure to deal with diagnostic heterogeneity [15]. They included an impressive 27 diagnostic categories from different nosologies that included ‘reactive psychosis’, ‘schizoaffective disorder’, ‘manic-depressive illness’ and ‘affective personality disorder’ to find a ‘central tendency’ of around 15% but then say it would be more precise to ‘acknowledge that lifetime suicide mortality has ranged from 5% to 26% in the literature’! We would argue that an awareness of the biases would lead one to say that these results may reflect vastly different standards of methodology in examining cohorts that were variably overrepresentative of high-risk sufferers.
A nosologically sensitive statistic
To further obfuscate the application of Guze and Robins' findings came the advent of DSM-III in 1980. With the shift away from the aetiological conceptualizations of ‘melancholic’ and ‘endogenous’ depression to an atheoretical, dimensional approach came a lowering of the threshold of diagnosing depression. Moreover, large community surveys, such as that which provided our data, identify a further, larger group of cases that typically do not come to medical attention at all.
The net effect has been to greatly increase the number of people diagnosed with major depression. But, do they share the same suicide risk as those who might previously have been admitted in the 1940s and 1950s with a diagnosis of ‘melancholia’ or ‘psychotic depression’? This increase in the denominator (where number of suicides make up the numerator) contributes further to the down-rating of suicide risk in major depression.
While the profession could be forgiven for continuing to promulgate nosologically sensitive statistics after a significant nosological shift occurs, this is a salutary lesson in how such shifts can (further) invalidate certain ‘facts’.
Our much greater calculated suicide risk in males is partly influenced by women being diagnosed with major depression twice as frequently as men. Men who suicide are drawn from a pool that is half the size of their female counterparts. If the depression rates for both sexes were the same, the calculated difference in suicide risk between men and women would be halved, although men would still have a rate that was triple that of women. There is evidence that this in fact may be the case and our failure to recognize it is a diagnostic artefact. Egeland and Hostetter's study of the Amish, where substance abuse and sociopathy are culturally prohibited, revealed that in a sample of 12 500 the rates of unipolar illness were equal for both sexes [16]. Diagnostic criteria that could identify in general populations the extra men that Egeland and Hostetter found would increase the denominator substantially. However, whether this would increase or decrease the overall suicide risk remains unpredictable while we do not know the suicide risk that is attached to this undefined cohort.
A nosology that fails to deal with substance abuse and sociopathic behaviours in diagnosing unipolar depression will skew demographic data through failing to identify all (male) cases of major depression. It may also be that since the threshold of diagnosis for major depression has been lowered this change is more inclusive of women than men: we shall return to this issue in a moment.
When evaluating the relevance of a particular suicide risk study, particularly that from an earlier nosological era, consideration will always have to be given as to whether the diagnostic changes have altered the denominator from which any given figure is calculated.
Let us finally consider recent research that suggests how the diagnosis of depression could be reconsidered in a way that might improve not only diagnostic validity but the prevention of suicide.
In conclusion: is DSM-IV major depression a female-oriented diagnosis?
Our results indicate that while the suicide risk in women with major depression is nothing to be complacent about, their male equivalents are much more likely to suicide. As well as men using more violent and potentially successful means, depressed males are consistently less able to access help than their female counterparts, adolescent males are even less likely to have received care. The factors mentioned above are summarized in Table 2 (with points 1 and 2 elaborated in more detail in our second paper [2]).
Considerations in explaining the higher incidence of suicide in men with major depression
As we discussed in the last section, underdiagnosing depression in men has research ramifications when attempting to establish suicide risk, but it also has ramifications for clinical practice. It may be that the higher rate of suicide risk that we have identified for men is partly the result of them not being diagnosed until the psychological postmortem after they suicide. Clinicians, the various referral agencies and the general public, educated to recognize typical major depression, may not be identifying atypical cases in men and thereby denying them therapy in all its forms.
Rutz et al. have reported compelling research from the Swedish island of Gotland that avoided any sampling biases by studying the island's population in its entirety [17]. When this research is considered alongside Egeland and Hostetter's work the question is squarely raised as to the role of the diagnostic construct in understanding the risk of suicide for men.
In Gotland an intensive general practitioner education programme resulted in a dramatic improvement in a number of depression identification and treatment parameters including an increase in the prescription of antidepressants and a reduction in suicides of 60% – but only in women! Suicide in men was largely unaffected by the educational programme. In explaining their finding, Rutz and coworkers pointed out that, for a start, current DSM-IV diagnostic criteria [18] rely upon reported symptoms, and men seldom voluntarily report depressive symptoms. They go on to postulate a male depressive syndrome that emphasizes features that include stress intolerance, acting out, aggressiveness, low impulse control, indecision, sleep fragmentation, morning anxiety, endorphin-stimulating behaviour and feeling ‘burned out’. The syndrome is distinguished from a personality disorder by identifying a deterioration from a previous level of functioning.
Rutz et al. invoke ethological research that defines two types of breakdown coping strategies: the fight–flight response and the stuporous or catatonic ‘playing dead’ or Totstell reflex. The latter response, more frequently found in the female of a species (unless with young), is a deficit state that aligns closely with the current diagnostic features of depression. Men, who are more likely to adopt the adrenalin–noradrenalin enriched fight–flight response are less likely, in this mode, to align with current diagnostic conceptualizations even though their response may simply be a stereotypical one to a stressor, external or physiologic, identical to that experienced by females. Whatever the explanation, the Gotland experience highlights the fact that our current approach to identifying major depression and encouraging helpaccessing fails men dismally.
Rutz et al.'s proposed syndrome would predict that the emergence of ‘fight’ type symptoms (acting out, aggressiveness, low impulse control) in the context of a deterioration from a previous level of functioning, would respond to antidepressants. There is indeed mounting evidence of a role for selective serotonin re-uptake inhibitors (SSRIs), particularly fluoxetine, in the treatment of anger. A number of studies have now demonstrated a capacity for these medications to decrease aggressive feelings and behaviour in varied animal settings and clinical trials [19]. Other studies have found that a reduction in anger and aggression is independent of changes in ratings of depression, however, these studies included sufferers with personality disorders [20, 21]. Finally, ‘anger attacks’, found in around 40% of those with major depression, respond to fluoxetine in 64–71% of cases and appear to be a variant of depression rather than of panic disorder [22].
The foregoing, taken together, raises the possibility that the nature of disturbed modulation of serotonergic pathways in male depression, or its equivalent, may more often manifest along the anger/aggression spectrum and be equally responsive to modern antidepressant treatment. It may even be that some cases of substance abuse conversely represent the modern equivalent of the ‘flight’ component of the fight– flight response. Men who have decompensated into an anger dyscontrol depressive equivalent, and then selfmedicate through substance abuse, are more likely to be diagnosed with a substance abuse or personality disorder. This is particularly likely if a history of deterioration from a previous level of functioning has not been elucidated.
If two of the key functions of a diagnostic system are to identify those at risk and to dictate effective treatment, then the current diagnostic system can be considered to be working better for women than men. There may be a case to be made for researching gender-specific constellations of symptomatology with a view to reviewing the diagnostic criteria for major depression. At the very least there is a need to evaluate, without prejudice, the possibility that the current diagnostic criteria suffer from a gender bias.
In the interim, given the treatable agony of depression and the availability of safe, well-tolerated antidepressants, there is a compelling argument for lowering our threshold of treatment in men who present with all of three clinical features: (i) a history of anger dyscontrol or substance abuse; (ii) decompensation from previous levels of functioning; and (iii) accompanying features of either typical (DSM-IV) depression or Rutz et al.'s ‘atypical’ male depressive syndrome. It may be necessary to reconceptualize the diagnosis along these lines if we are to effectively combat the alarming increases in male youth suicide rates.
What is less debatable is the need to abandon a combined figure when discussing suicide risk and instead quote risks for each sex. This distinction is a necessary step in heightening clinical concern when dealing with male patients showing either typical or atypical depressive features and will focus future research on the key predictor of gender in understanding suicide risk.
Footnotes
Acknowledgements
We would like to specifically acknowledge the critical contributions made by Margo Eyeson-Annan, statistician, our co-author on both papers, and Chris Cantor, coauthor to the second paper.