Abstract
The Lundby Study has a unique place in psychiatric epidemiology. It is the longest comprehensive prospective study of an entire community. The work began when Essen-Möller had the idea of observing the entire population of a community in the south of Sweden, notionally called Lundby. His aim was to study individual traits and morbidity in a general population, not in patients. In 1947 he and three other psychiatrists examined all but 1% of the 2550 adult inhabitants of Lundby aged ≥15 years [1]. They would not have known that their respondents were to be assessed a further three times over the next 50 years. In 1957 Hagnell re-examined 98% of the same people, irrespective of domicile, together with 1013 newcomers who had moved into the area [2]. In 1972 and again in 1997, the investigating team examined the survivors of the original 1957 cohort [3]. In addition to the psychiatric interview, and unlike most other surveys, they used multiple sources of information: face-to-face interviews, informants and community nurses, general practitioners, death registers, the Swedish psychiatric register, the national hospital inpatient register and the local outpatient register. The community was ethnically homogeneous. The exceptional duration of the study means that the cohort has been exposed to significant social and economic changes.
The Lundby work has been conducted throughout by research psychiatrists. It yields data about incidence, not just prevalence. It has maintained exceptionally high contact rates throughout, even when people had moved out of the area. Information on those who had died was obtained from informants and registers. In 1997, 50 years after the start, the investigators claim that they were able to obtain sufficient information to reach a diagnosis, where indicated, on 99% for persons in the first 25 year period, and 94% in the second [4].
These advantages are offset by some significant deficiencies. In the first place, for estimates of the incidence of specific diagnoses, the numbers are inevitably small, too small at times to examine rates by age group and gender. Next, morbidity is examined only categorically rather than on a continuum. No data are presented from symptom scales, although we note that the Hopkins Symptom Checklist was administered in the 1997 follow up [3]. Then there is uncertainty about the validity of diagnoses. These were reached by the psychiatrists themselves, who were free to explore the respondent's symptoms and behaviour at interview, using their own ‘Lundby diagnostic system’. They also used the interview data to reach consensus diagnoses by DSM or ICD according to which version was in place at the time. The conditions that allowed such a research style included relative freedom of the investigator from administrative chores and from pressure to publish; circumstances not widely prevalent at present. The Lundby strategy is the antithesis of what is done in today's large scale surveys. There, information is obtained by lay interviewers, symptom by symptom, following a tightly scripted text [5] from which they must not deviate. It is then assembled by computer algorithm to reach what is accepted by the investigators as a diagnosis. The World Mental Health Surveys in 15 countries, the large US surveys and our own national surveys in Australia and New Zealand have been based on this method. But in our view, no one yet knows which method gets closer to the clinical truth across all the main diagnostic groups.
Next, when a change in incidence is found, there is the possibility that it might not be real, but attributable to differences that have taken place between one wave and the next, in how diagnoses were reached or in non-random attrition. If a change is accepted as a valid observation, there are always many candidate variables in the environment to account for it. These caveats need to be borne in mind in interpreting the changes in incidence reported from Lundby.
In previous publications the Lundby team have reported a number of changes over the 50 year span. In a celebrated paper, ‘Are we entering an age of melancholy?’, Hagnell et al. presented evidence for an increase in depressive disorder between 1947 and 1972 [6]. The paper attracted a great deal of attention, with much effort by others to find if the increase was to be found elsewhere. Using the same 1972 data, Rorsman et al. found the cumulative incidence of depression up to the age of 70 to be 27% in men and 45% in women, thereby showing how episodes of this disorder are so common in the general community [7]. After the study had run another 25 years, Mattisson et al. found in the 1997 survey that the increase in depression had ended [8]. Congruent with this, Bogren et al. found a drop in the incidence of neurosis [9]. An interesting observation was made by Nettelbladt et al., who found that after the introduction of antidepressants in 1962, the suicide rate in depressed persons decreased, but the overall rate increased [10]. In the period 1977–1992 other diagnoses, including alcohol abuse, were more often associated with suicide than was depression.
To determine if personality traits confer vulnerability to mental illness, which was Essen-Möller's original aim, one essential requirement is that these be assessed prior to the onset of symptoms. This had been done in the 1947 wave using the dimensional trait model proposed by Sjöbring [11]. Mattisson et al. found that the personality trait of being nervous or tense predisposed to depression for both men and women [12]. For men, the personality trait of having low energy and having had nervous symptoms as a child were risk factors. For women, being easily hurt or introverted was associated with subsequent depressive disorders. Of some significance, they found that the sexes shared only some risk factors, indicating possibly different pathways to depression.
Another problem addressed systematically by Essen-Möller in the original survey was the association between neurotic and physical morbidity. He found that the age-specific point prevalence of combined physical and psychiatric morbidity was 159 per 1000 for age groups 15–39, 312 per 1000 for age groups 40–59 and 413 per 1000 for age groups ≥60 [1]. Such data, bridging the traditional split between medicine and psychiatry, were also highly relevant to the planning of primary health care.
For the psychoses, Bogren et al. reported an increase between 1947 and 1997, but only in men [9]. For psychoses with severe impairment, the age-standardized incidence for men increased from 0.8 per 1000 years at risk to 1.4, but there was no rise in women. For organic states including the dementias, they found a decrease in both sexes during the 50 years. In this issue Bogren et al. now report the incidence of psychosis in 3498 persons during a period of 50 years, analysed by sex and age group [4]. In that period, 18 men and 24 women developed a non-affective psychosis. Of these, there were nine men and nine women with schizophrenia. Such small numbers preclude any meaningful age- or sex-specific analyses. Accepting that, they note an earlier age of onset of psychoses in men, but all cases of late-onset psychosis were in women. Substance-induced psychotic disorder was higher in men. The World Health Organization's 10-country study of schizophrenia found incidence rates similar to Lundby, despite major differences in methodology [13]. For affective psychoses, the age of onset and overall incidence are again close to other studies.
Finally, the repeat census survey of the Lundby population enabled Hagnell and Öjesjö to examine the total mental morbidity in the community and, by cutting across nosological boundaries, to look for natural groupings of incident disorders that may not be apparent if only the rates for specific disorders were added together [14]. They identified two major clusters of illnesses, a middle-years-maximum group that included mainly neuroses, and a rising-with-age group consisting predominantly of organic brain conditions. These two groups, sufficiently important because of their size, also raise the question of whether the clustering of diverse illnesses at particular stages of the life cycle may reflect shared risk factors and pathogenetic mechanisms.
Mental illnesses occurring in the same individuals in a single community, across half a century, have never before been examined like this. Because the entire community that Essen-Möller started with was not many times larger, the yield of cases has proved too low for some analyses. Despite that, and the many possible distortions from case definition, attrition and recall bias, the landmark Lundby Study is a unique record of the pattern of mental illnesses as they have occurred in a general population over half a century.