Target populations and sampling frames

Adults aged 18-64 years were the primary survey population. Community study samples were identified through census registers or lists of patients registered with primary care physicians. These sampling frames may be considered equally valid in terms of the reliability of the data sets (Reference Shanks, Kheraj and FishShanks et al, 1995). Census registers may be highly reliable at the time of collection, but suffer from attrition over time, and the quality of electoral registers in Britain has been vitiated by their use for local taxation purposes. Primary care registers, accessed conjointly across a locality, offer a slightly different but equally accurate representation of the population. In the present study, the different research teams involved made the choice to use one sampling procedure or the other — that is, either census records or lists of patients registered with primary care physicians — based on their previous experience in community surveys. Three centres (Oslo, Turku and Santander) had previously achieved high response rates through population register surveys, and therefore used this method for Phase 1 screening in the ODIN study. Patients in Britain and Ireland were identified through primary health care registers, a selection process similar to the one used in the EURODEP study that assessed the prevalence of depression among those aged ≥ 65 years at the Dublin and Liverpool centres (Reference Copeland, Beekman and DeweyCopeland et al, 1999). The Irish research team had to reduce the scope of its intended sampling and interviewing procedures owing to operational problems which arose during the study. At the rural site in Ireland, the registers of five general practitioners were involved, out of a total of 27; in Dublin, the registers of two general practitioners were involved, from a total of 390. At the British rural site, seven of the nine practices that covered the population area took part in the study; in Liverpool, 32 practices of the 106 that covered the population area participated.

In Oslo, Turku and Santander, the sample was randomly drawn from the population registers of the five sites (two in Norway and Finland, one in Spain) involved in the study. In Liverpool, a random set of patient names was obtained from health authorities, and interviewers contacted the practices with which the patients were listed. In northern Wales and Ireland, the procedure was to identify relevant practices, and obtain random sets of names from their patient lists. The entire sample was stratified by gender and age in all the centres.

First phase

The first-phase assessment identified possible cases of depression using the Beck Depression Inventory (BDI) (Reference Beck, Ward and MendelsonBeck et al, 1961), with a threshold score above 12 (Nielsen & Williams, 1980; Reference Lasa, Ayuso-Mateos and Vázquez-BarqueroLasa et al, 2000). The BDI was combined with a questionnaire on social support (details available from the author upon request), the List of Threatening Experiences (Reference Brugha, Bebbington and TennantBrugha et al, 1985) and socio-demographic details. In Santander, the first phase was conducted by home-based personal interview. In all other centres, it was conducted using an initial postal survey, with postal, then telephone, then home-visit follow-ups. All refusals were accepted, and non-responders were contacted up to three times.

Second phase

All of those scoring at or above the BDI threshold and a random 5% of responders were offered detailed interviews with research workers trained in mental health, conducted in the participant's language. To date, most of the epidemiological studies on depression in the general population have used strict definitions of depression, according to DSM-III/IV (American Psychiatric Association, 1980, 1994) and ICD-10 (World Health Organization, 1992) criteria, focusing on the prevalence of depressive episodes or major depression. This may lead to a tendency to consider severe/major depression as the only affective disorder worthy of intervention. In order to overcome such a prejudice, in our study we extended the definition of depressive disorders to include dysthymia and adjustment disorders with depressive mood. The Schedule for Clinical Assessment in Neuropsychiatry (SCAN) Version 2.0 (World Health Organization, 1994) was used to generate diagnoses of depressive disorders on the basis of ICD-10 and DSM-IV categories. For ICD-10, these include single and recurrent depressive episodes (F32, F33), bipolar and persistent affective disorders (F31, F34) and adjustment disorders with a depressive component (F43.2). For DSM-IV, these include depressive, bipolar and adjustment disorders with a depressive component (codes 293.83, 296.xx, 300.4, 309.xx, 311, V62.82). Other instruments administered at this phase have been described elsewhere (Reference Dowrick, Casey and DalgardDowrick et al, 1998).

Training and quality control

The diagnostic interviewers were psychiatrists, general practitioners or psychologists. All interviewers received an initial weeklong training course at an approved SCAN training centre, and subsequently practised the full diagnostic interview schedule on at least 10 volunteers. Interrater reliability over time was monitored by means of assessment and feedback using a standardised videotaped consultation. A videotape including a full SCAN interview was used for this exercise, supplied by the WHO-approved SCAN training centre that trained ODIN's first-phase diagnostic interviewers. Each diagnostic interviewer was asked to rate and score the interview, then send their score sheets to a central analysis centre (Liverpool). Scores were compared with the ‘official’ set of ratings which accompanied the video. The videotaped interview contained 113 questions that could be rated, and all 13 of the interviewers were included in this exercise. A 100% agreement was reached for overall diagnosis (moderate depressive episode) and for diagnostic category (F32.1). There was 70% interrater agreement on scores for individual questions.

Statistical analysis

Routine data management and description of the results were carried out using SPSS 7.5 for Windows (SPSS Corporation, 1997). Prevalence estimates were carried out using STATA Release 6.0 (Stata Corporation, 1999) after allowing for both the two-phase sampling procedure and different response rates across sites through the use of weights (Reference Pickles, Dunn and Vázquez-BarqueroPickles et al, 1995; Reference Dunn, Pickles and TansellaDunn et al, 1999). Information arising from the Phase 1 screening results and the Phase 2 sampling mechanism was processed by assigning a ‘sampling weight’ to each individual participant, given by the inverse of the Phase 2 sampling fraction. The sampling weight is an indicator of how many participants in Phase 1 are ‘represented by’ each of the Phase 2 records. We used adjustment weights in order to standardise the overall prevalence estimates at each centre for the age and gender distribution of the population of Santander, which served as our reference. Prevalence estimates were obtained via a logistic model, producing a symmetrical confidence interval for the regression coefficient, and then reversing the logistic transformation to produce the corresponding interval for the prevalence itself. Prevalence estimates were calculated separately for each of the study sites. We also carried out a meta-analysis of the pooled data from the nine centres to obtain an overall weighted prevalence.

Prevalence estimates

The methodology used in this project, two-phase sampling, is a type of stratified design that has been proposed in psychiatric research as an efficient way of estimating prevalence of psychopathology in large epidemiological surveys. The diagnostic instrument used in the second phase of the study is the latest version of the SCAN, which has been presented by some authors (Reference Brugha, Bebbington and JenkinsBrugha et al, 1999) as closely approximating a ‘clinical gold standard’.

The major finding of the present study is the wide difference in the prevalence of depressive disorders found across the study sites and between urban and rural centres. Taking the genders together, the centres fall into three categories: high prevalence (urban Ireland and urban UK: 12.8-17.1% respectively), low prevalence (urban Spain: 2.6%) and medium prevalence (the rest of the sites: 6-9.3%). The study found high proportions of depression among survey responders in some centres, particularly among the female population in urban areas. At seven of the nine study sites, the prevalence of depressive disorders was higher among women than among men, confirming the results of several previous studies (Reference Bebbington, Sturt and TenantBebbington et al, 1984; Reference Weissman, Bland and CaninoWeissman et al, 1996). Over the past few decades, there has been growing evidence of significant intergender differences in the rates of specific mental disorders (Reference Lehtinen, Londholm and VeijolaLehtinen et al, 1990). A variety of social and medical factors have been considered in an attempt to explain the higher rate of depressive disorders in women (Reference Bebbington, Sturt and TenantBebbington et al, 1984; Reference Vázquez-Barquero, Diez-Manrique and GaiteVázquez-Barquero, 1987). Further analysis of the data collected in the epidemiological arm of the ODIN study will enable us to test whether some of the gender differences in the depressive disorders prevalence estimates across sites could be explained by different levels of exposure to life events and other social factors.

When DSM-IV and ICD-10 criteria were compared, there were notable similarities in the global prevalence figures for all diagnostic categories; however, at some centres a discordance was found between ICD-10 and DSM-IV diagnoses. Previous studies have also documented discordance between these two major psychiatric classification systems, for the diagnosis of cases in community surveys (Reference Andrews, Slade and PetersAndrews et al, 1999). This could be due to the ICD-10's lower threshold in the number of symptoms required for the diagnosis of a depressive episode.

Urban-rural differences

Most research on the epidemiology of depression in Europe has been conducted in urban settings. The few available studies assessing differences in prevalence of depression between urban and rural areas vary widely in their findings (Reference Brown and PrudoBrown & Prudo, 1981; Reference Vázquez-Barquero, Diez-Manrique and GaiteVázquez-Barquero et al, 1987; Reference Sievewright, Tyrer and CaseySievewright et al, 1991). This variety of outcomes could be attributable to differences in the measures used, and in the selection and sampling of rural and urban areas. In the ODIN study we tried to overcome these problems by using the same methodology in the urban and rural sites at each centre. We found that the weighted prevalence of depressive disorders among responders in the four rural communities was relatively uniform, ranging between 6.5% and 9.3%. However, the weighted prevalence in the five urban communities varied markedly, from 2.6% in Santander to 17.1% in Liverpool and 12.8% in Dublin. These differences in the prevalence figures suggest that there are cultural differences or different risk-factor profiles across countries and sites which may affect the expression of the disorder. The role of psychosocial factors (mainly life events and social support) in explaining these wide differences in the prevalence of depressive disorder across study sites will also be examined in later publications. Our finding of lower prevalence of depressive disorders in rural areas than in urban ones, in three of the four countries where urban and rural samples were studied, agrees with the results of other epidemiological studies (Reference Crowell, George and BlazerCrowell et al, 1986). In Britain, the National Psychiatric Morbidity Survey reported higher rates of depression in urban areas than in rural ones, but relied on the interviewers' opinion of whether subjects lived in an urban, semi-rural or rural areas (Reference Jenkins, Lewis and BebbingtonJenkins et al, 1997). The Netherlands Mental Health Survey and Incidence Study also found that people living in rural regions showed lower prevalences of mood disorders (Reference Bijl, Ravelli and van ZessenBijl et al, 1998).

In the ODIN study, not only did the rural communities show a lower prevalence of depressive disorders than the urban ones, but the prevalence figures in rural areas remained strikingly similar across the different sites, whereas the urban figures varied markedly from one country to another. A methodological factor in our design could have influenced this uniform distribution of depressive disorders among rural communities in Europe found in the ODIN study. In our methodology, ‘rural’ was defined according to a similar socio-economic definition: rural areas were those having no centre of population greater than 15 000 people, with at least 20% of economically active citizens engaged in occupations directly related to agriculture, fishing or forestry. On the contrary, ‘urban’ was defined solely in terms of population density criteria. Thus, the uniform distribution of depressive disorders across the rural sites could be merely a reflection of a similar distribution of socio-economic factors. On the contrary, the heterogeneity of the prevalence figures in the urban sites could be related in part to the heterogeneous socio-economic circumstances across the study sites. Further analysis of the ODIN sample will enable us to study to what extent these urban-rural differences in the prevalence of depressive disorders may be related to differential exposure to life events or differential levels of social support networks, as has been recently proposed by Paykel et al (Reference Paykel, Abbot and Jenkins2000).

Implications

Although the representation of European countries is incomplete, the centres in this study are spread between northern and southern Europe, representing different religious denominations and covering mainland areas and islands. The information gathered in the ODIN study could be used to evaluate needs for treatment and the allocation of health resources. However, some authors have raised concerns about the health policy implications of high prevalence rates of psychiatric disorders for determining treatment needs (Reference Regier, Kaelber and RaeRegier et al, 1998; Reference SpitzerSpitzer, 1998). In particular, they question whether making a psychiatric disorder diagnosis can be equated with demonstrating a treatment need. It is still unclear to what extent the psychiatric disorders, as defined by rigorous criteria such as the ICD-10 or DSM-IV used in our project, identified in community populations are equivalent to those identified with the same criteria in clinical settings, or whether they have the same clinical significance and response to treatment. The design of the ODIN enables us to introduce new elements of discussion into this debate, and adds fresh evidence that validates the ability of epidemiological studies to identify subjects with depressive disorders who might benefit from therapeutic interventions. As we have described elsewhere (Reference Dowrick, Casey and DalgardDowrick et al, 1998), this is the first population-based study that incorporated into its design a randomised controlled trial of individual problem-solving treatment and a group psychoeducation programme. Participants identified as having depression in the epidemiological phase were offered the chance to take part in the controlled trial, the results of which (Reference Dowrick, Dunn and Ayuso-MateosDowrick et al, 2000) suggest that psychological interventions are effective in reducing depressive caseness, symptoms and personal disability in the short term. Thus, our prevalence estimates identified a segment of the population with a depressive disorder that could benefit from an intervention.

We believe that the data in the present study confirm that depressive disorder is a highly prevalent condition among working-age adults in Europe, particularly in urban centres, and that this epidemiological information should be used to inform and implement equitable and effective health policies across the continent in order to address this public health challenge.