Soc Psychiatry Psychiatr Epidemiol (1995) 30:279-292

9 Springer-Verlag 1995

P. Bebbington. R. Ramana

The epidemiology of bipolar affective disorder

Accepted: 25 November 1994

Abstract This paper reviews the current position of studies on the epidemiology of bipolar affective disorder. A disorder that cannot be recognized until sometime after its onset poses special difficulties for epidemiological study. These are discussed and attempts made to solve them. Community psychiatric surveys suggest a morbid risk of bipolar disorder of around 2-2.5%, but probably include many false-positives. Studies of treated cases indicate a morbid risk of 0.5%, but will miss untreated cases. It is probably reasonable to suggest a compromise value of 1-1.5%; bipolar disorder is thus still a rare condition. It is possible to quantify the unipolar-bipolar conversion rate, which is of the order of 5%, and it is of particular interest that female sufferers have proportionately fewer manic episodes. Age at onset, possible cohort phenomena, comorbidity, and sociodemographic correlates are discussed.

Introduction The last 2 decades have seen an enormous amount of research into affective illness, paralleling the continuing development of specific physical and psychological treatment methods. This has followed the refinement of the concept of polarity, which has both encouraged and been sustained by the search for various biological risk factors, particularly genetic markers. However, there

P. Bebbington ([E~) MRC Social & Community Psychiatry Unit, Institute of Psychiatry, De Crespigny Park, London SE5 8AF,UK R. Ramana Department of Psychiatry, University of Cambridge, Level E4, Addenbrookes Hospital, Cambridge CB2 2QQ, UK

has been a neglect of epidemiological research into bipolar illness, and a dearth of hard data. The acceptance of the concept of polarity has, in any case created considerable difficulties for epidemiological study. It is, after all, difficult to establish the frequency of a disorder that can only be recognized at an unspecified point in its course, i.e. the time at which polarity definitely switches. However, we think that it is timely to bring together the available evidence, and to attempt some kind of integration, partly with a view to defining areas of ignorance where research might be encouraged.

The concept of polarity Falret (1851) 1 was the first to emphasize a differentiation of two forms of affective disorders based on polarity. However, Kraepelin's broad definition of manic-depressive illness (Kraepelin 1921), which did not distinguish depressive and manic subtypes, overshadowed Falret's initiative for almost a century. Leonhard revived the concept in the 1950s (Leonhard 1957-1971) in relation to what he termed the "endogenous affective disorders". He defined a unipolar illness as one with recurrent episodes of either depression or mania, and a bipolar illness as one in which both sorts of episodes occurred. However, it was only in the 1960s, following the comprehensive investigations of Angst (1966) and Perris (1966), that psychiatrists around the world began to accept a division of affective illness that now seems self-evidently sensible. Although consistency demands that if we separate bipolar disorder from unipolar depression we should also distinguish it from unipolar mania (Perris, 1966), most authorities have chosen not to (e.g. Angst, 1966; Dunner et al., 1976).

*translated in Sedler (1983)

280

This is at least pragmatic, as unipolar mania is rare even in comparison with bipolar disorder, and thus very difficult to study on its own. In America, bipolar illness has been sub-divided further (Dunner et al., 1970) into bipolar I (with clear cut episodes of mania and depression) and bipolar II (with depression and short lived episodes of hypomania). American authors remain convinced of the usefulness of the distinction (e.g. Endicott et al., 1985) even though it seems likely to be quantitative rather than qualitative. A category of bipolar III disorder has now suggested to cover so called 'pseudo-unipolar' cases (Akiskal et al., 1989). The concept of polarity entered the American Diagnostic and Statistical Manual from the third edition onwards (APA, 1980). The International Classification of Disease was slower to relinquish the Kraepelinian manic-depressive concept, not doing so until the recently published Tenth Edition moved closer to the American classification (WHO, 1992). In the nineties, the use of the concept of polarity in affective illness has thus become standard, so enhancing the comparability of research throughout the world.

Where should we look for cases of bipolar disorder?

It has been suggested that the proper place to study affective disorders is in the community, as few sufferers are referred to psychiatrists (Dohrenwend and Dohrenwend, 1974). Now that there are agreed and standardised methods for case finding, it is possible to mount reasonably reliable general population psychiatric surveys (Bebbington, 1990). However, they are much less suitable for severe mood disturbances. Bebbington (1988) has argued that a rare disorder like bipolar illness cannot credibly be studied in the majority of sample population urveys as it is very difficult to derive a reliable prevalence, let alone incidence. Data from community surveys based on the Diagnostic Interview Schedule (DIS; Robins et al., 1985; Robins and Regier, 1991) and the Composite International Diagnostic Interview (CIDI; Robins et al. 1990) have now allowed us to derive better estimates of the frequency of the disorder, as the sample numbers are large and similar community surveys have been carried out in various parts of the world. However, hypomanic mood is quite difficult to establish with certainty in the general population, as we ourselves discovered when conducting the Camberwell Community Survey (Bebbington et al. 1981). Sufferers usually lack insight into the abnormality of the mood, and this is a crucial drawback in distinguishing mild cases from normal cheerfulness. The risk of false-positives is also high, as it can be very difficult to distinguish

between ordinary high spirits and abnormal elevation of mood, particularly if hypomanic disorder is to be recognized after a long interval, as in the detection of lifetime prevalence. Considerable clinical experience is thus required if the identification of cases is to be given credence. There must, therefore, be particular reservations about the use of the DIS and the CIDI, which purport to remove the need for clinical judgment through the rigidity of their structure and are designed for use by lay interviewers. Large variances have been noted in the diagnoses of mania made by lay interviewers and psychiatrists (Robins et al. 1985). However, the results retain considerable interest, as they are effectively unique. For most of our information we are thus thrown back on patients identified within specialist medical services. However, due to varying diagnostic practices, case definition, service quality and admission policies, and the risk of duplicating admissions, routinely collected data on first admissions to hospitals are unlikely to be very reliable. Case registers, while offering some advantages, also have their own shortcomings, including referral bias, changing base populations and reliance on diagnoses made by individual psychiatrists. However, they do provide cumulative data and avoid duplicate counts, and thus offer a valuable source of incidence data, particularly for rarer disorders (Der and Bebbington 1987).

Methodological difficulties in the establishment of rates

In new cases of affective disorder, initial episodes may be manic, mixed or depressive, and any one of these presentations may herald a bipolar disorder. Thus, most cases of bipolar disorder cannot be diagnosed at the time of the first episode, and an accurate diagnosis is only possible after a variable interval. If we wish to establish the incidence of bipolar disorder, we are obliged do so retrospectively. We have a problem in deciding how retrospectively, but it certainly implies that we need access to data that can be used to identify individuals over a long period. Additionally, the numerator that would provide the true incidence of bipolar disorder includes an unknown number of as yet unipolar subjects, the bipolar III category of Akiskal and his colleagues (1989). Their number is unknown because we lack the relevant follow-up information. While several researchers have provided data about the rate at which subjects ultimately known to be bipolar become identifiable as such, what we actually require is the survival function in terms of remaining unipolar for all subjects who initially present with monophasic episodes of illness. There are ways of circumventing this requirement. Dunner and his colleagues (1976) several years ago described a method based on an application of Bayes'

281

theorem, 2 given values for (1) the probability of a bipolar subject becoming manic following a given number of episodes of depression, (2) the proportion of a sample of patients identified as affectively ill who are known to have bipolar disorder and (3) the probability of someone with a non-bipolar affective illness having at least a given number of episodes of depression. For first episodes, the last value is 1.0 by definition. Dunner and his colleagues (1976) have suggested that 5% of patients initially hospitalized for a depressive episode will have a bipolar course. Akiskal and his colleagues (1983) found that 20% of their initially unipolar patients converted to bipolarity, but their sample was probably biased. The results from the Epidemiologic Catchment Area ECA study are more consistent, with a 10% conversion rate (Robins and Regier 1991), but we have already expressed our reservations about the accuracy of the DIS in identifying elevated mood. Thus, we think 5% is probably of the right order, although it is likely to differ in male and female subjects as the gender ratio is much closer to unity in bipolar illness. If we decide to subsume unipolar mania under bipolar disorder, the numerator for the incidence of bipolar disorder will thus include all cases whose first affective episode is manic or biphasic, plus say 5% of those whose first episode is depressive. Unfortunately, this does not quite provide us with an unqualified estimate of incidence, because the heterogeneity of course in subjects with first episodes of depression depends crucially and in an unknown manner on the severity of the index depressive episodes. There are other methods of calculation that are basically simpler. The incidence of bipolar disorder is quite closely related to the incidence of mania, provided the latter is strictly interpreted in terms of cases identified for the first time as having mania. This is based on the presumption that since all bipolar cases will eventually suffer an episode of mania they will inevitably appear in the numerator of the incidence of mania at some stage. There are two problems with this. First, there may be a sizeable delay between the onset of affective disorder and the emergence of bipolarity, during which the base population may change. Second, and more importantly, most values for the so-called incidence of mania use as the numerator the number of people who happen to have mania making contact with services for the first time, rather than those making contact for the first time for mania. This means that those with a first episode of mania who have already made contact for some other reason will be excluded.

2 They were concerned with a slightly different problem, namely the estimation of heterogeneity in an apparently homogeneous sample of patients with unipolar depressive episodes, irrespective of whether the episode was their first. In incidence estimations we are obviously concerned with heterogeneity only in subjects with first episodes. This actually makes the estimation more plausible

A more feasible method of estimation is thus to derive empirically the proportion of bipolar subjects whose illness began with a depressive episode and divide the rate of new inceptions with mania by this fraction. This is again based on the assumption that unipolar mania can be subsumed under bipolar disorder. Unfortunately, the values obtained from various studies differ quite a lot, and probably reflect the specialized services (lithium clinics, university research units) where cases are obtained. Thus, Perris (1968) has found that 34% of bipolar cases begin with an episode of mania, while Johnson and Hunt (1979) and RoyByrne et al. (1985) both quote 40%, Winokur et al. (1969), 61% and Dunner and his colleagues (1976), 70%. As an approximation, the incidence of bipolar disorder is probably around twice the first contact rate for those with mania. Given these difficulties, most recent studies have avoided analysing incidence and have studied lifetime prevalence, which is more easily arrived at, and probably a more valid measure for this disorder. It is the proportion of persons in a representative sample of the population who have ever experienced the disorder up to the date of assessment. The period to which this prevalence refers thus differs between subjects (being dependent on their ages), and the value arrived at is consequently a reflection of the age structure of the population. However, there are considerable advantages to the use of lifetime measures, particularly since it is essential in any case to have information about previous episodes in order to make a valid current diagnosis of bipolar illness. Lifetime prevalence identifies a large number of affected cases and is less likely to be affected by the duration of episodes, when compared to period prevalence (Robins et al. 1984). It allows calculation of annual first episode rates by excluding persons who are not at risk of a first episode, but is likely to be reduced artificially by mortality in older persons if the disorder carries an increased death rate (as severe affective disorders do), and in particular by errors of recall (Bromet et al. 1986). Where differences between demographic groups relate more to the frequency than to the fact of episodes, they will be obscured by the use of lifetime prevalence in large field studies like the ECA surveys. Negative findings, therefore, need to be interpreted cautiously. These techniques aim to establish a credible value for the incidence of bipolar disorder. In a crude way, they also permit the prediction of a bipolar course in the individual case of depression, and, therefore, have clear clinical utility. This predictive function can be amplified by relating the chances of a bipolar course to sociodemographic and other variables. Akiskal et al. (1983) have found the following variables useful in predicting a bipolar course: age under 25 years at onset, a family history of bipolar illness, loaded pedigree, precipitation by childbirth, hypersomnia and retardation during depressive episodes, and

282 antidepressant-induced hypomania. The results of W i n o k u r and his colleagues (1993) are essentially in agreement, adding male sex and hyperactive traits in childhood. With further genetic and epidemiological data, estimates of likely polarity m a y perhaps be made more confidently during the first affective episode.

inaccurate diagnosis, selective migration and overrepresentation in recent immigrants of the age group with the highest incidence.

Lifetime prevalence and period prevalence rates Incidence of bipolar illness Values for incidence from various studies are summarized in Table 1. The values quoted by Boyd and Weissman (1982) vary from 9.2 to 15.2 per 105 per year for males and from 7.4 to 32.5 per 105 per year for females. The data from the Camberwell register comprise all first contacts with psychiatric services that are diagnosed as cases of mania. If we follow the rule of t h u m b of doubling these values to obtain an approximation to the rate for bipolar disorder, this would give figures of 9.0 per 105 for men over age 15 years and 9.6 per 105 for women. Alternatively, if we add to the incidence of mania 5 % of the incidence of severe depressive disorder, we obtain somewhat lower values, particularly in men, with rates of 5.9 and 7.4 per 105 per year, respectively. As might be expected, the reported incidences based on first admission rates are lower, ranging from 3.0 to 8.3 per 105 per year for men and from 2.0 to 10.7, for women. Two sets of data from Camberwell in South L o n d o n have shown a particularly high incidence of mania in Caribbean-born males c o m p a r e d to the locally born (Der and Bebbington 1987; Leffer et al. 1976). However, this could arise from a variety of artefacts, including

Table 1 Incidence of bipolar

disorder

Study

Table 2 shows values for lifetime prevalence from the c o m m u n i t y surveys currently available. Apart from the earlier New H a v e n study and the recent National C o m o r b i d i t y Survey, alI have relied on the use of the DIS. When diagnosis of bipolar disorder is based on a history of a manic episode, the rates vary fiom 0.3% in the Hispanic c o m m u n i t y in Los Angeles to 1.1% in New H a v e n and St Louis. Robins and Regier (1991) have reported estimates of the lifetime prevalence of bipolar (rather than manic) disorder in the ECA studies. Values range from 0.4% to 1.2% or, if bipolar II disorders are included, from 0.8% to 1.7%. The value from the National C o m o r b i d i t y Study is at the high end of this range. The rate from Taiwan is low in relation to most of the other studies and also in relation to the period prevalence from the same study, suggesting that the accuracy of lifetime prevalence was particularly p o o r in this investigation. Despite the infrequency of the condition and the difficulty of recognizing it, the values for prevalence are fairly consistent across different sites. The mean lifetime prevalence of bipolar I and II disorders obtained from the ECA studies is 1.3%, with a standard deviation of 1.2. Bland and his colleagues (1988; F o g a r t y e t a l . 1994) used the data from their E d m o n t o n DIS study to calcu-

Location

Period

Incidence per 105 Males

Females

Total

1957-1974 1960-1964 1966-1973 1964-i982

14.1 15.2 9.2 4.5

7.4 17.4 32.5 4.8

10.8 16.3 20.8

Based on first admissions Spicer et al. 197Y England and W a l e s 1965-1966 Left et al. 1976a Camberwell, S. London Aarhus, Denmark Weeke 1979~ Central Danish Registry 1970-1972 Weeke 1979~ Central Danish Registry 1978 Nielsen 1979~ Samso Island, Denmark 1957-1974

3.0 3.1 3.1 8.3 4.8

3.9 2.3 2.0 10.7 4.7 -

3.5 2.6 2.6

Based on all referrals for treatment Nielsen 1979a Samso Island, Denmark Weeke 1979" Aarhus, Denmark Helgason 1979~ Iceland Der and Bebbington 1987b Camberwetl, S. London

6.1

"Quoted as a personal communication by Boyd and Weissman. Details of methodology in Nielsen and Nielsen (1976), Weeke et al. (1975) and Helgason (1977) ball first inceptions who suffered from mania. Based on over 2 million person years at risk CAll first admissions who suffered from manic-depressive psychosis, manic or circular. Cl-year prospective study of all first admissions who suffered from mania

283 Table 2 Lifetime prevalence of bipolar disorder based on c o m m u n i t y psychiatric surveys (ECA, Epidemiologic C a t c h m e n t Area, DIS, Diagnostic Interview Schedule, SADS-L, Schedule for Affective Disorders and Schizophrenia lifetime version

Location

Study

Prevalence % Males

Weissman and Myers 1978"

Females

Totals 0.6e/1.2 f

N e w Haven, U S A

ECA studies: Robins et al. 1984 b

New Haven, U S A Baltimore St. Louis Los Angeles N o n - H i s p a n i c whites Hispanics N e w Haven Baltimore St. Louis D u r h a m N.C. Los Angeles

K a r n o et al. 1987 b

Robins and Regier 1991

0.9 0.8 1.1

1.3 0.5 1.1

1.1 0.6 1.1

_

_

_

_

1.0 0.3 1.2e/1.9 f 0 . 6 ~ / 1 . 2 f 1.0e/1.5 f

_

_

0 . 4 ~

-

r

0.6a/1.1 e

Total

0.7"/I.1 f

0.9e/1.4 f

0.8e/1.3 r

Puerto Rico Edmonton Taiwan Christchurch, New zealand Seoul Iceland Munich

0.7 0.7 0.14 h 0.5

0.4 0.4 0.9 h 0.9

-

-

0.2 -

0.2/0.5 g

-0.6 1.1 h 0.7 0.4 0.2/0.5 g 0.24

Kessler et al. 1994 b

United States

1.6

1.7

aBased on S A D S - L bManic episode ~Cohort aged 55-57 dSample aged 25-66

eBipolar I fBipolar I & II gAtypical bipolar disorder hEstimated from published data

Other DIS studies: Canino et al. 1987 b Bland et al. 1988 H w u et al. 1989 b Wells et al. 1989 b Lee et al. 1990 Stefansson et al. 1991 Wittchen et al. 1992 a

US National Comorbidity Survey

late morbid r i s k - this is a rate calculated on the basis of correction for the fact that members of the population have lived out different proportions of their life expectancy. They report values for manic episode of 1.4% for men and 0.6% for women, i.e. around twice the lifetime prevalence. Reported period prevalences for manic episodes and bipolar disorder emphasize the problems of using community surveys to establish the rates of rare disorders. Most of the variation in the values obtained even from quite large surveys is probably random. Table 3 gives values for 6-month prevalence rates of manic episodes, ranging from 0.1% to 0.8 % in the various DIS studies. Using a clinical interview, Faravelli and his colleagues (1990) have found a very high rate for bipolar I disorder of 1.7%. This may have been due to vagaries of case identification or to the orientation of Italian mental health policies towards community management. In any case, the rate is much higher than those in the DIS studies, which range from 0.3 to 1.0 for bipolar I, and from 0.6 to 1.4 for bipolar II disorder. Thus, a fair amount of the prevalence of bipolar disorder is accounted for by cases of depression with a previous history of manic disturbance. The values for manic

1.6

episodes are noticeably lower in the non-ECA DIS studies-it is not clear why, although sample sizes were smaller in these surveys. Again the National Comorbidity Survey is at the high end of the range. In the studies that provide separate values for the two sexes, there is again no consistent gender difference. The rate of bipolar disorder is almost equal across the ECA sites (1.2) with lifetime prevalence rates for manic episodes in men and women of 0.7 and 0.9 respectively. Only in the Amish study has the ratio been different, with 58% of bipolar patients being male and 42% female, while the rates for unipolar disorder are almost equal (Egeland and Hostetter 1983). The authors believe that sociopathy and alcoholism are rare and do not mask male affective illness in this group, thus allowing a more accurate estimate of the rate, and that the role of women in Amish culture may lead to an underestimation of female depression. Based on these studies, the incidence of bipolar disorder when cases are defined as any psychiatric referral is probably between 10 and 20 cases per 105, and around 8-10 per 105 when the calculation is based on first admission rates. This suggests bipolar disorder is a rare condition, with a morbid risk of the order of

284 Table 3 Six-month prevalence of mania and bipolar disorder based on community psychiatric surveys

Location

Study

Prevalence % Manic episode

BipoIar 1.7 c

Florence

Faravelli et al. 1990

Bipolar I & II

I

ECA studies: (Robins and Regier 1991 Myers et al. 1984 Burnham et al. 1987)

New Haven Baltimore St. Louis Durham N.C. Los Angeles

0.8 0.4 0.7

Total

Other DIS studies: Canino et al. 1987 Bland et al. 1988 Hwu et al. 1989 Oakley-Brown et al. 1989 Wittchen et al. 1992

Puerto Rico Edmonton Taiwan Christchurch Munich

0.3 0.1 0.1 c,d 0.1 0.2

US Na~ioT~al Comorbidity Survey: Kessler et al. 1994

United States

1.46

"Based on SADS-L bpoint prevalence ~One-year prevalence

0.2

1.0 0.5 1.0 0.3 0.4

1.2 0.9 1.4 0.6 0.7

0.6

0.7

1.0

1.3c

1.3c

dEstimated fi'om published data

0.5%. Community surveys, as might be expected, give higher values, with a lifetime prevalence of around 1% (lifetime prevalence in this condition would be perhaps half the morbid risk). However, the value for period prevalence of manic episodes in the ECA studies is quite high, approximately 0.5%; this casts some doubt on the values for lifetime prevalence, particularly in the light of what we have said already about the difficulty of recognizing hypomanic mood in general populations,

-o

1.0 g

1.0

o-

.~ ~ o.a

0.a ?'

-~ 0.6

0.6

g

9c_

= o

0.4

0.4

o o_

0.2

0.2

0

o 2

3

~-'

4

Number of manic episodes

Patterns of bipolarity

It is possible to shed light upon the nature of bipolar illness by studying both the overall pattern of development of bipolarity and the way it emerges in the individual case. Figures 1 and 2 show the probability of an episode of a given polarity following episodes of the opposite kind. The first is drawn from data published by Perris (1968), while the second is based on summed data from five studies (Perris 1968; Dunner et al. 1976; Johnson and Hunt 1979; Akiskal et al. 1983; Roy-Byrne et al. 1985), which in any case report data of considerable similarity. Both figures are based on cases selected precisely because they eventually become bipolar. It can be seen that the conversion curve following initial mania is very similar to that following initial depression. It looks as though the probability that an episode of depression will follow an episode of mania starts at

Fig. 1 Development of bipolarity following initial mania (taken from Perris 1968)

~ 1.0 -2 g 0.8

1.0

-~ oo-

o.8 ~E~ 0,6

"~ 0.a ~.~ 0.4 -~ E 0.2 g o

0.4

~. o

0.2

3 2,

0 2

3

4

c~ I,

Number of depressive episodes

Fig. 2 Development of bipolarity following initial depression (summed data from five studies)

285

just under 50% and declines somewhat with successive episodes. It would be hard to argue that the probabilities of mania following any episode of depression follow a very different pattern. If this reflects the relative possibilities of mania or depression constituting the first episode of bipolar illness, it suggests that they are about equal and we are thus right in thinking that the actual incidence of bipolar disorder is twice the reported incidence of mania. These results also support Perris' (1966) argument that the error in regarding subjects who have experienced three depressive episodes without mania as unipolar is of an acceptable order, representing as it does a failure to identify one true case of bipolar illness of six. However, the Polish study of Bogdanowicz (1991) has reported that almost half of bipolar cases convert after three or more consecutive depressive episodes. Two studies (Dunner et al. 1976; Johnson and Hunt 1979) also provide information about the development of bipolarity in relation to the overall duration of illness (Fig. 3). After a decade of illness, a fifth of cases still had not revealed their bipolar nature. The patterns of illness in the National Institute of Mental Health (NIMH) study have been analysed in rich detail (Roy-Byrne et al. 1985). This was based on 71 bipolar subjects, with very full recordings available on 46. The authors claim evidence for two patterns of relapse; some subjects have a constant interval between episodes, while others reveal a 'sensitivity pattern' with a progressively shorter interval. However, the earlier intervals in patients displaying the sensitivity pattern were long, and the reduction in length merely brought these patients in line with those who had experienced constant intervals from inception. The overall mean and median lengths of the first interval were 189 and 100 weeks, respectively. This material also suggests that subjects whose first episode is depressive have fewer subsequent episodes of mania. This goes counter to what we have argued above, namely that the chance of any given episode

1.0

0.8 E

l

,~ D u n n e r e t 9

Johnson

al 1 9 7 6

& Hunt 1979

0,6 0.4 0.2 O 1

2

3

4

5

Years after

6

7

8

9

onset

Fig. 3 Development of bipolarity following initial depression

being of mania or depression is roughly equal. The finding could be a structural artefact, as the overall proportion of manic episodes in cases where the first episode is depressive is reduced by virtue of the fact that the number of depressive episodes is constrained to be at least o n e - t h i s is not the case where the first episode is manic, where the proportion of manic episodes is increased for the same reason. Nevertheless, the proposition that the degree of bipolarity varies between subjects is an interesting one. Although the morbid risk of bipolar disorder is not very different in the two sexes, females have proportionally more depressive episodes (cf. Angst 1986); this corresponds to their greater susceptibility to depression generally. These findings have implications for the genetics of bipolar disorder. This is usually studied by contrasting the overall family history with that seen in unipolar disorder, the crudest of comparisons. Detailed models relating the clinical loading of mania to gender and the actual pattern of illness within families might reveal more about the nature of its inheritance.

Age and bipolardisorder The period prevalence of a disorder is based on both incident and recurrent cases, and, as such, says little about age at onset. The highest prevalence in the ECA studies (2.1%) was in the 18-24-year age group in New Haven (Robins and Regier 1991) Lifetime prevalence decreases after the 4th decade, contrary to what one would expect. It might seem reasonable to conclude from this that bipolar disorder arising in old age is uncommon. However, various authors have argued that bipolar disorder is not as rare as presumed in the elderly (Yassa et al. 1988; Shuhnan and Post 1980; Glassner and Rabins 1984; Spicer et al. 1973; Shulman and Post, 1980; Stone 1989; Young and Klerman 1992). Yassa et al. (1988) have found that 9.3% of patients over the age of 60 years with affective disorders have bipolar disorder arising for the first time, and Glassner and Rabins (1984) found that 4.6% of their first admissions to a psychogeriatric ward had mania. Eagles and Whalley (1985), in their study of incidence from first admissions have reported an increase in the incidence of, mania with age, although much less markedly than in depressive psychosis. There appears to be little difference in the clinical presentation of these late onset bipolar disorders, and little evidence that they are more likely to have physical precipitants (Glasner and Rabins 1984). However, a positive family history of bipolar disorder is relatively rare. Sibisi (1990) has argued that the influence of age on the incidence of mania differs between men and women, with women having a higher inception rate than men in the middle years. Nevertheless, there was little variation by age in his United

286 12.5

Table 4 Age at onset of bipolar illness

10.0

Study

Mean age at onset in years (SD)

7.5 s o. o o

~=

50 25 0 15-24

25-34

35-44

45-54

55-64

65-74

75+

W e r t h a m 1929 P e r r i s 1968 W i n o k u r et al. 1969 C a r l s o n et al. 1974 L o r a n g e r a n d L e v i n e 1978 B a r o n et al. 1983 F i e v e et al. 1984

Males

Females

Overall

32.9 (13.9) 29.1 33.0 (13.4) 30.0 (13.5)

30.3 (11.7) 27.1 31.6 (13.9) 30.9 (13.5)

20.7

21.5

32 30 31 a 35 b 26.1 28.3 c 30.8 a 21.2

A g e band

Fig. 4 Incidence of mania by age (from Der and Bebbington 1987)

Kingdom national admission data. Our own data suggest that incidence is independent of age (Der and Bebbington 1987; Fig. 4). The limitations of studies on age at onset have been outlined by Loranger and Levine (1978). These comprise small numbers of cases, failure to distinguish polarity, use of different definitions of onset (first hospitalization, treatment or appearance of symptoms) and varying criteria for defining disorder. Some are retrospective surveys of records, which are more prone to diagnostic errors. Most studies are of inpatient populations, which will include recurrent and more severe forms of the illness. The numbers of affected people in the community who are excluded may be significant. Table 4 lists clinical studies that have reported age at onset in bipolar disorder. Comparison between the studies is difficult due to varying methodology. Moreover, the mere stipulation of a mean or median age at onset is a crude way of presenting such data. Crowe and Smouse (1977), 3 Baron et al. (1983) and Smeraldi et al. (1987) provide more sophisticated data based on actuarial techniques that take the demography of the base population into consideration. Although this is a potential source of bias, their results are not substantially different from those that do not make this correction. It might be expected that a truer picture would be obtained from cross-sectional community surveys. Indeed, the size of the ECA studies has permitted the identification of sufficient cases of bipolar disorder to provide reasonable estimates of age at onset (Burke et al. 1990). When the data are summed, the mean age at onset of 156 cases of bipolar I disorder was 18 years, and of 99 cases of bipolar II disorder, 21.7 years. This compares with 26.5 years for Major (unipolar) depressive disorder. The reported age at onset is less in subjects who are younger at the time of interview (Robins and Regier 1991). These results are consistent with those reported from the Edmonton study (Bland et al.

3Crowe ad Smouse (1977) have reanalysed the data of Winokur et al. (1969)

A k i s k a l et aI. 1983 J o y c e 1984 W e i s s m a n 1988

aBipolar I UBipolar II

CFirst syndrome dFirst hospitalization

1988), where the mean age at onset of bipolar disorder in males was 20.5 and in females, 20.0 years. However, the assessment of bipolar disorder in crosssectional surveys is inevitably retrospective and subject to distortion, since those affected but lost to the study (the 'censored' population) are excluded. Life table methods overcome this by accounting for this censored population in calculating the probability of onset of an illness at a given age, provided the individual has reached that age (Fleiss et al. 1976). A recent study has used life table methods to calculate the age at onset of various disorders from the ECA surveys (Burke et al. 1990). The authors have confirmed that the hazard rate for bipolar disorder is greatest between 15 and 19 years of age, and does not differ significantly between the sexes. Although the available studies give a mean age at onset that varies from 16.6 to 33 years, those on which most reliance can be placed suggest values towards the lower end of this range. The peak period of onset thus appears to be late adolescence or early adulthood, with some cases of childhood onset reported. The early onset of bipolar disorder strengthens the case for separating it clinically from the unipolar form, especially since there may be a link between early age at onset and a family history of affective illness (Gershon et al. 1971).

Possible cohort effects Although lifetime prevalence and age at onset may be acknowledged to differ according to demographic status, they are often treated as fixed in relation to time. However, this may not be the case. A number of authors claim to have discovered cohort effects indicating an increasing frequency of affective disorders with succeeding generations (Klerman et al. 1985; Gershon et al. 1987; Lavori et al. 1987; Joyce et al. 1990; Burke et al. 1991). The combined ECA data are large enough to

287

provide equivalent analyses restricted to bipolar disorder suggesting a similar effect (Fig. 5). If these postulated cohort effects are genuine, they are very interesting. It would be difficult to argue that they arise from changes in gene frequency over so short a time, and the explanation must therefore be environmental, and probably social. However, Gershon (1989) has suggested that the cohort effect for affective disorders is more marked in the relatives of those with affective disorders than in the general population, implying a gene-environment interaction. The American authorities place considerable weight on these findings (Lasch et al. 1990; Robins and Regier 1991, p. 70). In fact, they must be treated with caution. It is very difficult to separate age, period and cohort effects statistically, because fixing any two of these factors constrains the third. Wickramaratne and his colleagues (1989) have attempted to do this by making certain assumptions about potential period and cohort effects in the ECA data set. The model of best fit for females required only age and cohort effects, while for males, it required age, period and cohort effects. The number of cases of bipolar disorder is too small to permit this sort of analytical detail. Other reservations compound these statistical difficulties (Bebbington 1994), of which the most important is probably the failure of recall - older folk may appear to have a lower lifetime prevalence because they are unable to remember episodes occurring many years ago. An excellent account of the potential pitfalls in studies of this type is provided by Klerman and Weissman (1989).

2.5

o

2.0

o ~D ~3.

1.8

"Z,

1.0 0.5

~