Article Text

PDF

Absence of spatial variation in rates of the common mental disorders
  1. Scott Weich
  1. Correspondence to:
 Professor S Weich
 Division of Health in the Community, Warwick Medical School, Medical School Building, University of Warwick, Coventry CV4 7AL, UK; s.weichwarwick.ac.uk

Statistics from Altmetric.com

Place may still matter—but not in the ways that have been studied to date.

The intuitive importance of location as a determinant of life chances1,2 contrasts with growing evidence of little or no variation in the prevalence of the most common mental disorders (CMD), anxiety and depression, across small and mid-sized areas—particularly after adjusting for the characteristics of individual residents.3–9 The study by Henderson and her colleagues,10 based on an urban US sample confirms this. By contrast, larger area level effects are found for psychotic illnesses (such as schizophrenia) and more severe forms of depression.11–15 Should we conclude that place doesn’t matter for the most CMD, or are there alternative explanations for these negative findings?

ARE WE STUDYING THE WRONG SPATIAL SCALE?

The spatial scale at which contextual factors might have an impact on mental health remains unknown. Most studies have used data collected within administrative boundaries.2,16 Studies of large areas, such as UK regions (with hundreds of thousands of residents), are difficult to interpret.17,18,19 Recent studies have examined effects over smaller areas, ranging from Amsterdam boroughs (average population 33 000), postcode sectors and neighbourhoods (average population 8000–10 000),19 to UK electoral wards (average population 5500),3,4,7 and US census tracts (average population 4000).9 Effect sizes at these levels are small and rarely statistically significant—percentage of variance in symptoms of anxiety and depression ranges from 0.5% to 4% before adjusting for residents’ characteristics, to less than 1% after doing so.

Wards may be too large and heterogeneous to detect contextual effect, and variance in CMD may be greater over smaller areas.20 The significance of this modest trend remains unclear, and there have been few studies of very small areas. A study using postcode units (average population 150) in South Wales (Glyn Lewis, personal communication) found results that differed little from those across UK electoral wards. “Neighbourhood” remains notoriously difficult to define.21 While some studies have defined neighbourhoods using natural boundaries,22–24 others have used “neighbourhood” to describe administrative units such as US census tracts.25–27 Other studies avoid this altogether, asking survey respondents to make implicit judgements about boundaries of “their neighbourhood” or “their area”.28

One notable finding is substantial between household variation in rates of CMD in Britain, although most studies overlook this as a discrete level.3,5,14 In a national study, over 10% of variance in score on the 12 item general health questionnaire occurred at household level. This finding was not changed, even by adjusting for characteristics of individuals (including marital status, ethnicity, education, employment status, financial strain, and the number of current physical health problems), households (income, car access, housing tenure, social class, composition), or wards.4 This finding remains unexplained, but could be attributable to exposures operating at a spatial level between ward and household.

ARE WE STUDYING THE WRONG EXPOSURES?

As most studies are secondary analyses,2 places are often characterised using census based aggregate measures of the population composition. The most commonly used measures are those reflecting prevailing socioeconomic circumstances, such as (un)employment, housing tenure, and proportion of single adult households, either singly or in combination,5,6,9,10 or income inequality.10a None of these is associated with the prevalence of CMD to a statistically significant degree after adjusting for individual characteristics.

“Contextual” measures reflect characteristics of places rather than residents. The latter are rare in the literature, notable exceptions being studies with independent observations of the built environment.24,26,27 The distinction from compositional indices may not be clear cut,2 as variables measured at the individual level (for example, educational attainment, employment) may be influenced by area characteristics (for example, quality of local schools, job opportunities, public transport). Unfortunately, the dearth of validated contextual measures precludes empirical investigation of this hypothesis. The work of MacIntyre and her colleagues, who are developing measures of place based on needs, represents a major empirical advance.28

Summary statistics from surveys of residents’ perceptions of their local environment25,29 are one alternative to census based descriptors. This approach is common in social capital research, which has been criticised theoretically and empirically (Health Development Agency, 2004). Another approach has been to classify small areas according to a particular characteristic. Several UK studies have found a higher prevalence of CMD among those living in “urban” (compared with “rural” or “suburban”) areas.30–33 Suicide rates are also higher in urban than rural areas of Britain,34 although this gradient may be falling.35 In Sweden, a study of the entire population aged 25–64 found a statistically significant linear association between increasing population density and rates of first admission for depression.15 By contrast, studies in New Zealand,36 USA,37 Scandinavia,33 and Canada38 found no evidence of statistically significant urban-rural differences in CMD prevalence. These inconsistencies may be partly methodological, especially given varying definitions of “urban” and “rural”. Cross national comparisons are difficult given historic, socioeconomic, and ethnic differences in rural and urban populations in different countries.39

Population density (for example, Sundquist et al15 and Wang38) may fail to capture aspects such as geographical remoteness,35 and some researchers have resorted to subjective or impressionistic definitions.31,33 The assumption that rural residents are less deprived and healthier than their urban counterparts has also been challenged statistically. Rural wards (in the UK) are smaller, and have greater internal (between individual) variability with respect to deprivation than urban wards. While rural areas are more internally heterogeneous, even over areas smaller than wards, there is less variation in deprivation between rural areas than their urban counterparts.40 Associations between area level socioeconomic deprivation and worse health emerge for rural areas when wards are aggregated to approximate the greater size of urban wards.40

ARE WE STUDYING THE WRONG OUTCOMES?

Reliance on secondary analysis and the need for large samples means that most studies use self report measures of anxiety and depression symptoms. In studies of individual socioeconomic status and CMD, larger effect sizes are observed in studies that use standardised clinical interviews.41,42 The same may also be true for place and CMD.14 Standardised clinical interviews might capture more severe episodes of disorder, and be less prone to “false positives” arising from mild or transient disturbance, or physical ill health. However, traditional objections to findings not based on clinical diagnostic categories are lessened by evidence that CMD are most validly represented as a single dimension encompassing comorbid anxiety and depression.43–45 One important problem is that measures such as the general health questionnaire may be prone to socioeconomic response bias, with those in lower occupational grades underreporting symptoms.46

The dearth of prospective studies is striking. This is especially problematic in studies of chronic or recurrent disorders such as anxiety and depression, because cross sectional studies may conceal associations between risk factors and either the onset or outcome of episodes of disorder. Evidence that socioeconomic adversity is associated with longer episodes of the CMD but not episode onset47–49 suggests that episode duration should be longer in areas with the highest levels of socioeconomic deprivation.

ARE OUR MODELS MISSPECIFIED?

Substantial variance in CMD at household level is little changed by controlling for a host of household and individual level characteristics. As this variance remains unexplained, our models remain incomplete. The findings at household level are consistent with spousal similarity in depressive symptoms,50 and intra-household processes warrant scrutiny.

The effects of place may also vary with individual and households characteristics.2 This is reflected in an urban excess of CMD only among those who were economically inactive,51 variation in suicide rates with area level and individual socioeconomic factors, particularly unemployment,52,53 and interaction between ethnicity and urban/rural location in the association with depressive symptoms among those living in poverty in the USA.54 Thus, place may only affect those with specific vulnerabilities.13

ARE WE ASKING THE WRONG QUESTIONS?

It is inconceivable that the effects of place on mental health are instantaneous, and cross sectional studies are arguably the least informative. The most potent risk factors may be those operating during childhood.55 Educational and employment opportunities vary considerably between places. Adjusting for the socioeconomic characteristics of residents overlooks the fact that these are likely to be determined in part by where they live (or have lived). We know that deprived people live in deprived places and are less healthy than those living in affluent areas. We need to know much more about residential mobility, who moves between areas, why they move, and what effect this has on their health. The health effects of residential mobility (or lack thereof)—like those of place more generally—may vary with individual circumstances, including health.56

CONCLUSIONS

There is little cross sectional variance in the prevalence of the CMD between areas with populations of 5000–8000 in the UK, Netherlands, and USA. Such areas may be too large to observe effects at a very localised level. Substantial variance at the household level may provide partial support for this view.

Place may still matter—but not in ways that have been studied to date. Anxiety and depression are important public health problems in their own right, and their prevalence is not declining. These conditions are also associated with mortality and physical morbidity, particularly cardiovascular disease. As acute and chronic environmental stressors are potent drivers of onset and outcome, living in places with fewer amenities, or where personal safety is less secure, might lead to higher rates of psychiatric morbidity. Alternatively, risk may be confined to those with specific vulnerabilities, or those exposed to such environments at developmentally critical times. If this is so, modifying the physical or social environments could lead to substantial reductions in rates of the most common mental disorders.

Most existing studies are limited by reliance on secondary and cross sectional data sources, administrative rather than natural geographical boundaries, and compositional measures of place. There is a need for more hypothesis driven primary research, and more measures of place that do not rely on residents’ characteristics or perceptions of their locale, and that describe places at different spatial scales. Research needs to be longitudinal, based on population samples large enough to test hypotheses about interactions between people and places, and inclusive of household level exposures and outcomes.

Place may still matter—but not in the ways that have been studied to date.

REFERENCES

  1. 1.
  2. 2.
  3. 3.
  4. 4.
  5. 5.
  6. 6.
  7. 7.
  8. 8.
  9. 9.
  10. 10.
  11. 10a.
  12. 11.
  13. 12.
  14. 13.
  15. 14.
  16. 15.
  17. 16.
  18. 17.
  19. 18.
  20. 19.
  21. 20.
  22. 21.
  23. 22.
  24. 23.
  25. 24.
  26. 25.
  27. 26.
  28. 27.
  29. 28.
  30. 29.
  31. 30.
  32. 31.
  33. 32.
  34. 33.
  35. 34.
  36. 35.
  37. 36.
  38. 37.
  39. 38.
  40. 39.
  41. 40.
  42. 41.
  43. 42.
  44. 43.
  45. 44.
  46. 45.
  47. 46.
  48. 47.
  49. 48.
  50. 49.
  51. 50.
  52. 51.
  53. 52.
  54. 53.
  55. 54.
  56. 55.
  57. 56.
View Abstract

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Linked Articles

  • In this issue
    Carlos Alvarez-Dardet John R Ashton