Major depression is a serious, recurrent disorder linked to diminished role functioning and quality of life, medical morbidity, and mortality [1, 2]. The World Health Organization ranks depression as the fourth leading cause of disability worldwide [3], and projects that by 2020, it will be the second leading cause [4]. Although direct information on the prevalence of depression does not exist for most countries, the available data indicate wide variability in the prevalence rates. Weissman et al. [5] published the first cross-national comparison of major depression as defined by the Diagnostic and Statistical Manual of Mental Disorders, third edition (DSM-III) from 10 population-based surveys that used the Diagnostic Interview Schedule (DIS) [6]. The lifetime prevalence ranged from 1.5% (Taiwan) to 19.0% (Beirut), with the midpoints at 9.2% (West Germany) and 9.6% (Edmonton, Canada). The 12-month prevalence ranged from 0.8% (Taiwan) to 5.8% (Christchurch, New Zealand), with the midpoints at 3.0% (US) and 4.5% (Paris). A subsequent cross-national comparison [7] included 10 population-based studies that used the World Health Organization (WHO) Composite International Diagnostic Interview (CIDI) for the revised third edition and the fourth edition of the DSM (DSM-III-R) and (DSM-IV) [8]. Consistent with the earlier report [5], the lifetime rates ranged from 1.0% (Czech Republic) to 16.9% (US), with midpoints at 8.3% (Canada) and 9.0% (Chile). The 12-month prevalence ranged from 0.3% (Czech Republic) to 10% (USA), with midpoints at 4.5% (Mexico) and 5.2% (West Germany). Most recently, Moussavi et al. [9] summarized data on depressive episodes as defined by the International Classification of Diseases, 10th revision (ICD-10) in participants in the WHO World Health Survey used in 60 countries, noting that the 1-year prevalence was 3.2% in participants without comorbid physical disease, and 9.3% to 23.0% in participants with chronic conditions.

The wide variability in lifetime and 12-month prevalence estimates of major depression is presumably due to a combination of substantive (genetic vulnerability and environmental risk factors) and measurement (cultural differences in the acceptance and meaning of items, and the psychometric properties of the instruments) factors. Differences in study design might also be involved. That is, apart from administering a common interview schedule, the surveys were not designed as replications with a standard protocol for translation, interviewer training, sampling and quality control. More recently, the WHO World Mental Health (WMH) Survey Initiative conducted a coordinated series of studies using a common protocol and a common instrument, the WHO CIDI, version 3.0 [10], to assess a set of DSM-IV disorders in countries from every continent [11]. The 12-month prevalence of DSM-IV major depressive episode (MDE) in 18 countries ranged from 2.2% (Japan) to 10.4% (Brazil) [12]. The mid-point across all countries was similar to that in previous surveys (5%), as was the weighted average 12-month prevalence for the ten high-income (5.5%) and eight low- to middle-income (5.9%) countries.

Almost all studies find that gender, age and marital status are associated with depression. Women have a twofold increased risk of MDE compared with men [13], people who are separated or divorced have significantly higher rates of depression than the currently married [5, 7], and the rate of depression generally goes down with age [5, 7]. This evidence, however, comes primarily from studies conducted in western countries. The sparse data available from low- to middle-income countries suggest that the age pattern is either not monotonic or that the association is reversed, with depression increasing with age [12, 14]. Other socioeconomic factors have less consistent relationships with depression in different countries [7].

The current report presents data on the prevalence, age of onset and sociodemographic correlates of MDE in 18 countries participating in the WHO WMH Survey Initiative. As noted earlier, each of the WMH surveys used the CIDI for DSM-IV. The CIDI includes a series of diagnostic stem questions to determine which diagnoses are assessed. Unlike previous reports from the WMH or previous surveys, our study used the screening information for MDE in responses to these diagnostic stem questions to conduct an examination of the screen-positive percentages, and of the conditional lifetime and 12-month prevalence of MDE in respondents who endorsed the diagnostic stem questions. This was carried out to investigate the possibility that cross-national differences in prevalence estimates of MDE are due, at least in part, to differences across countries in the optimal threshold of CIDI symptom scores for detecting clinical cases. If such variation exists, we would expect much smaller cross-national differences in endorsement of diagnostic stem questions (which merely ask respondents if they had episodes of several days of being sad or depressed or losing interest in usual activities), than in diagnoses. If this were the case, we would expect the largest cross-national differences in conditional prevalence estimates of MDE to occur in screened positives. If differential variation of this sort exists, it would provide more reason than currently exists to suspect that cross-national differences in optimal diagnostic thresholds of the CIDI symptom scale lead to biased estimates of cross-national differences in prevalence in the WMH data.

A justification for this line of thinking comes from an earlier cross-national WHO study of major depression in primary-care patients, which found strong similarity in the latent structure of depressive symptoms across 14 different countries in different parts of the world, but also found that countries with the highest prevalence estimates generally reported the lowest impairment associated with depression [15]. The authors concluded from these results that although cross-national differences in the estimated prevalence of depression cannot be attributed to differences in the nature or validity of the concept of a depressive episode, it is possible that DSM criteria may define different levels of depression severity in different countries. Our cross-national comparison of responses to diagnostic stem questions, described in the previous paragraph, was designed to shed some light on this possibility. In addition, we carried out a parallel analysis of cross-national differences in impairment associated with MDE.

Results are organized by distinguishing between countries classified by the World Bank [16] as low- or middle-income versus higher-income countries. This distinction was made based on patterns both in the WMH surveys [10] and in other cross-national epidemiologic surveys [7, 9], which raise concerns that MDE prevalence estimates might be artificially lower in low- to middle- than higher-income countries due to methodological differences of the types considered here.

Consistent with previous cross-national reports, the WMH MDE prevalence estimates varied considerably between countries, with the highest prevalence estimates found in some of the wealthiest countries in the world. However, contrary to our initial expectation, we found no evidence that this wide cross-national variation was due as much to cross-national differences in endorsing diagnostic stem questions as to conditional prevalence of MDE among respondents who endorsed a diagnostic stem. The ratio of the highest to lowest screen-positive rates across countries (3.3) was very similar to the ratio of the highest to lowest conditional prevalence rates among screen-positives (3.0). As expected, we also found that MDE was associated with substantial impairment. However, contrary to our initial expectation, we did not find that cross-national differences in prevalence estimates were inversely related to differences in average level of impairment associated with depression; indeed, the opposite pattern was found.

Taken together, these results argue against the suggestion that the wide cross-national variation in depression prevalence estimates in the WMH surveys and previous epidemiologic studies is due to the threshold for defining clinically significant depression in standard diagnostic interviews differing across countries. If that were the case, we would expect that the cases of depression detected in countries with the lowest estimated prevalence of depression would be the most severe cases, resulting in high impairment rates among these cases, whereas the opposite would be true in countries with the highest estimated prevalence of depression. Furthermore, we would expect that reports of core depressive symptoms would be more similar across countries than estimates of disorder prevalence. Neither of these expectations was borne out in the WMH data. A question can be raised by our results regarding why the associations between prevalence and impairment were so different from those reported in the earlier WHO study [15]; however, it is important to bear in mind that this earlier study was based on primary-care samples, for which selection bias regarding seeking help on the basis of either distress or impairment might induce a more negative association between these two variables than exists in the population. The WMH surveys, by contrast, are based on general population samples, for which the selection bias issues that occur in treatment samples do not arise.

Although these results add indirect support to a substantive interpretation of the cross-national differences in MDE found here, they shed no light on why these differences exist. Differences in stress exposure, in reactivity to stress, and in endogenous depression unrelated to environmental provoking factors are all possibilities. On one level, it seems counterintuitive that people in high-income countries should experience more stress than those in low- to middle-income countries. However, it has been suggested that depression is to some extent an illness of affluence [26]. A related argument is that income inequality, which is for the most part greater in high than low- to middle-income countries, promotes a wide variety of chronic conditions that includes depression [27]. Further analyses of the WMH data might be able to shed some light on these perspectives; however such an analysis was beyond the scope of the current report, which focused on the evaluation of a more methodological interpretation of the observed cross-national differences in depression prevalence estimates.

In considering a substantive interpretation of our findings, it is noteworthy that although lifetime prevalence estimates were found to be significantly higher in high than low- to middle-income countries overall, no significant difference in 12-month prevalence was found. The ratio of 12-month to lifetime prevalence estimates, furthermore, was significantly higher in low- to middle-income than in high-income countries. It might be that these results reflect genuinely lower lifetime prevalence but higher persistence of depression in low- to middle-income than high-income countries, but another plausible and more parsimonious explanation is that error in recall of previous lifetime episodes is higher in low- to middle-income than high-income countries. Longitudinal data collection would be required to document such a difference rigorously [28, 29]. Although such data do not exist in all WMH series, it is important to recognize this possibility of cross-national variation in recall error before launching an extensive investigation of substantive explanations. It might be that a fruitful focus of subsequent WMH analysis would be on the youngest respondents, where lifetime recall error might be least pronounced. Alternatively, it might be that the investigation of cross-national differences in lifetime prevalence should be abandoned in favor of a focus on recent prevalence in recognition of the plausibility of significant cross-national variation in recall error of lifetime prevalence.

Another implication of the methodological limitation of the WMH surveys being all cross-sectional is that it made it impossible to determine the temporal direction of the associations examined between depression and the sociodemographic variables. This means that even though variables such as education and marital status were considered predictors of depression, they might actually have been consequences or involved in reciprocal causal relationships with depression. However, within the context of that limitation, the sociodemographic patterns reported here are broadly consistent with those found in previous community epidemiologic surveys of depression [2, 5, 7, 9, 13], adding to confidence in the generalizability of the WMH finding.

The results reported here have several other limitations, relating more generally to the WMH findings [30]. Some of the most important of these issues involve sampling. The response rates varied widely. Although the response rates did not appear to be related to depression prevalence, it is possible that in some settings, particularly those where treatment is unavailable, the most depressed people were unable to participate. Some surveys only included metropolitan areas, whereas others involved national samples. This too may have affected estimates of cross-national variation in prevalence. In addition, the surveys did not include institutionalized patients, people in jails and prisons, people in the military, people who were too intoxicated to be interviewed, or people with severe cognitive or physical disabilities. The samples also reflected survivor bias, which could be of considerable importance for understanding differences between high-income and low- to middle-income countries, given the gap in life expectancy of 10 to 15 years between people in developed and developing countries [31]. Thus, the rates reported here provide conservative estimates of MDE prevalence. A final noteworthy sample bias is that South Africa was the only African country included in this report [32] even though the WMH survey was also conducted in Nigeria [33]. Nigeria was excluded because of the extremely low prevalence of MDE (3.1% lifetime; 1.1% 12-month) and other disorders. These low prevalence estimates raise questions about the willingness of respondents in the Nigerian survey to disclose symptoms to strangers or lay interviewers, and the appropriateness of the CIDI structure for that setting [34]. They also reduced our statistical power to examine the associations of depression considered in the Nigerian data. A similar experience may have occurred in another African population-based survey using the CIDI that was not part of the WMH series. That survey, carried out in Addis Ababa, also found low rates of affective disorders [35]. Given the high level of exposure to trauma in extremely poor countries such as these [36], research is urgently needed to determine the best approaches to study the prevalence of mental disorders in these settings.

The measure of MDE also had inherent limitations. The structure of the CIDI, including the choice of stem questions in the screening section, may have led to underestimates of depression in some settings. As noted above in the section on measurement, the interview translation, back-translation and harmonization process in the WMH surveys included customization within countries of the terms used to describe the core symptoms of depression (that is, sadness, depression, loss of interest) based on clinical experiences of local collaborators and the results of pilot studies [19]. However, no attempt was made to develop distinct cut-off points in the CIDI diagnostic algorithms for different countries or to go beyond the DSM-IV criteria to develop distinct criteria for different countries that might have increased our ability to detect depression or depression-equivalents. It is noteworthy that in the countries for which we carried out blinded clinical reappraisal interviews with subsamples of WMH respondents, we found no evidence for systematic bias in the diagnostic threshold for depression [18], but clinical reappraisal interviews were not carried out in all WMH countries, and it is conceivable that such studies would have found systematic differences in the ability of the CIDI to detect clinical depression across countries.

Despite these limitations, the WMH data provide useful new information about the epidemiology of MDE. We found wide variation not only in the prevalence of MDE but also in the proportion of people who endorsed diagnostic stem questions for MDE, a pattern that has seldom been examined in previous epidemiologic studies [37]. We found cross-national consistency, by contrast, in the impairment associated with MDE. This association has to our knowledge never been considered previously in cross-national community epidemiologic surveys. Our results confirm the public-health importance of major depression as a commonly occurring and seriously impairing condition with a generally early AOO and persistent course in a wide range of countries. In addition, we replicated previous findings on the sociodemographic correlates of MDE. We also documented an intriguing opposite-sign pattern of differences between high and low- to middle-income countries in estimates of lifetime prevalence and persistence of MDE, which might be due to differences in recall error. Future research on cross-national differences in depression needs to take this pattern into consideration, and to develop a workable strategy to deal with the possibility of differential recall error as a plausible contributor to cross-national differences in prevalence estimates.