The Prevalence of Major Depressive Episodes Is Higher in Urban Regions of Canada

Abstract

Objective:

Major depressive disorder is an important contributor to disease burden. Anticipation of service needs is important, yet basic information is lacking. For example, there is no consensus as to whether major depressive episodes (MDE) are more or less prevalent in urban or rural areas. The objective of this study was to determine whether a difference exists in Canada.

Method:

A series of 11 Canadian national cross-sectional studies were examined from 2000 to 2014, providing much greater precision than prior analyses. Survey-specific MDE prevalence estimates were synthesized into a pooled odds ratio comparing urban to rural areas using meta-analytic methods.

Results:

Differences in the survey-specific estimates were not in excess of what would be expected due to sampling variability. This suggests that inconsistency in the prior literature is due to inadequate power and precision, an issue addressed by the meta-analytic pooling. The pooled odds ratio for Canada is 1.18 (95% confidence interval, 1.12 to 1.25), indicating that urban regions have higher MDE prevalence than rural regions. However, the difference is very small and of uncertain significance for policy and planning.

Conclusions:

Prevalence of MDE is approximately 18% higher in urban compared to rural regions of Canada. The difference is insufficient to impute differing need for services, but the result resolves an inconsistency in the existing literature and may play a role in future needs assessment.

Major depressive disorder is a substantial contributor to the global burden of disease, estimated to affect approximately 350 million people worldwide.¹ There has been speculation that urban living may contribute to an increased frequency of mental disorders.² However, current literature is inconsistent, with some studies finding no difference,^3–6 with some reporting higher prevalence in rural compared to urban⁷ or vice versa.^8–10 Recently, a national US study indicated no difference,¹¹ a finding that is not consistent with results from other countries.¹² In Canada, an analysis of cross-sectional data from the 1998 longitudinal National Population Health Survey (NPHS) found equivocal results. No overall association was found, but a trend (P = 0.05) was observed when the analysis was restricted to nonimmigrant populations.⁶ An analysis of data from the 2002 Canadian Community Health Survey, Mental Health and Wellbeing also found ambiguous results for major depression (P = 0.06).

There are several likely explanations for this inconsistency. If an existing association is weak, individual surveys lack the sample size (and power) to detect it. Alternatively, an association may not exist, or there may be international differences in the definition of urban and rural status or assessment of MDE. Using meta-analysis techniques, our goal was to decisively determine whether an association exists in Canada and to quantify the strength of any association that may exist.

Methods

Data Sources

The analysis for our study uses a series of cross-sectional data files from the Canadian Community Health Survey (CCHS). The CCHS surveys are conducted regularly by Statistics Canada and use similar methods. Detailed methodology of the CCHS is available from Statistics Canada archives.¹³ In brief, the target population is household residents living in the 10 Canadian provinces, excluding persons on Indian reserves, residents of health institutions, full-time members of the Canadian forces, and residents of remote areas in Ontario and Quebec. Sampling is done using a multistage procedure that begins with forming geographical clusters, followed by a selection of households within each cluster and then by sampling of 1 respondent from each household. The CCHS is a series of general health cross-sectional surveys conducted between 2000 and 2014, as well as 2 mental health surveys in 2002 and 2012. Respondents must be at least 12 years of age for the general health CCHS surveys and 15 years old for the 2 mental health CCHS surveys.

Study Measures

Past-year MDE was assessed using the Composite International Diagnostic Interview (CIDI). The 2 mental health surveys incorporated the full-length CIDI,¹⁴ whereas the general health surveys used an abbreviated version.¹⁵ Trained lay interviewers administered the survey to participants using computer-assisted interview procedures.¹³

Statistics Canada defines urban and rural status based on population density and proximity to a census metropolitan area (CMA) or census agglomeration (CA).¹⁶ Urban regions contain a population of at least 1000 with no fewer than 400 persons per square kilometre and can be further subcategorized into urban core, urban fringe, and urban outside CMA/CA. Urban cores are the regions with at least 10,000 individuals, located inside a CMA or a CA. Urban fringe includes all population centers within a CMA/CA that inhabit fewer than 10,000 people, and urban outside CMAs/CAs are the regions with 1000 to 9999 individuals not residing within a CMA/CA. All remaining areas not considered urban are classified as rural. This includes the regions within and outside CMA/CA that occupy fewer than 1000 individuals or 400 persons per square kilometre.¹⁶

Data Analysis

Survey-specific odds ratio estimates were derived using recommended weighting and variance-estimation procedures. The survey-specific estimates and associated 95% confidence intervals were then assessed for homogeneity graphically (using a forest plot) and statistically using the I² statistic and Q tests. We used a random-effects meta-regression model to explore heterogeneity potentially due to survey year and which of the 2 measurement instruments were used. Subsequent analysis developed odds ratio estimates for urban subcategories (urban core, urban fringe, and urban outside CMA/CA) compared to rural. In the meta-analysis, log odds ratios were pooled rather than the odds ratios themselves due to their more normal sampling distribution. A post hoc analysis assessed whether the association was consistent across Canadian provinces and territories.

All analyses were carried out in the Prairie Regional Research Data Centre (RDC) at the University of Calgary. According to Canadian ethical standards, this analysis did not require ethics review board approval.

Results

Table 1 shows the sample sizes available from each survey in the analysis. Random-effects meta-regression found no effect of survey year (β = 0.004, Wald test z = 0.60, P = 0.570) or measurement instrument (β = 0.078, Wald test z = 0.82, P = 0.442). Weighted survey-specific coefficients of the log odds ratios (ORs) are depicted in Figure 1, along with 95% confidence intervals (CIs). The I² and tau-squared values for heterogeneity were low at 35.2% and 0.0023 (Q = 12.321, df = 8, P = 0.137), indicating that the survey estimates are similar to one another and can be pooled into one estimate. Despite the low I², we made the a priori decision to use a random-effects model in the analysis since they resemble fixed-effects models when heterogeneity is low. The pooled estimate from Figure 1 translates into an odds ratio of 1.18 (95% CI, 1.12 to 1.25), which demonstrates a statistically significant (P < 0.01) higher prevalence of MDE in urban compared to rural regions (Table 2).

Table 1.

Samples Sizes of Surveys Included in the Current Investigation.

Survey	Sample Size
CCHS 1.1a	128,182
CCHS 1.2b	36,789
CCHS 2.1a	50,751
CCHS 3.1a	68,389
CCHS 2007/2008a	46,739
CCHS 2009/2010a	58,128
CCHS 2011/2012a	21,636
CCHS 12MHb	24,954
CCHS 2013/2014a	41,881
Total	477,449

^aCanadian Community Health Survey (CCHS), using the Composite International Diagnostic Interview (CIDI) short form for major depression.

^bMental Health Surveys, incorporating the World Mental Health CIDI for major depression.

Figure 1.

Survey-specific log odds ratio (OR) estimates for major depressive episode (MDE) in urban compared to rural regions, with the overall pooled estimate (I² = 35.2%; P = 0.137). Weights are from a random-effects model. CCHS, Canadian Community Health Survey.

Table 2.

Odds Ratio (OR) Estimates for Major Depressive Episode in Urban vs. Rural Regions, with Meta-Regression Output to Assess Heterogeneity between Survey Estimates.

	OR (95% Confidence Interval)	Meta-Regression Output
		I2 Statistic, %	Q Statistic	df	P Value
All urban	1.18 (1.12 to 1.25)	35.2	12.3	8	0.137
Urban core	1.19 (1.12 to 1.26)	46.5	14.95	8	0.060a
Urban fringe	1.25 (1.11 to 1.40)	0	5.88	8	0.661
Urban outside census metropolitan area	1.14 (1.07 to 1.21)	0	3.43	8	0.905

^aThe meta-regression P value for urban core is verging on significance (P = 0.06), which suggests some heterogeneity between estimates. However, the associated tau-squared value is 0.0037, indicating the underlying standard deviation of effects across studies is quite small.

The odds ratio estimates for MDE in subcategorized urban regions are depicted in Table 2. While urban fringe regions demonstrate the highest odds of MDE (OR = 1.25; 95% CI, 1.11 to 1.40), this is very similar to the estimates for urban core (OR = 1.19; 95% CI, 1.12 to 1.26) or urban outside CMA/CA regions (OR = 1.14; 95% CI, 1.07 to 1.21). These results support the decision to combine the urban subcategories into one urban region when assessing the urban-rural differences in prevalence.

The association between urban status and MDE was highest in Newfoundland (OR = 1.52; 95% CI, 1.27 to 1.83) and lowest in British Colombia (OR = 0.94; 95% CI, 0.84 to 1.06). Nonsignificant associations were observed in 4 of the 9 remaining regions, but the confidence intervals were wide. While these province-specific estimates lacked precision to detect the weak effect, the confidence intervals overlapped with the finding from Newfoundland, suggesting the urban-rural difference in MDE was similar across provinces (with the exception of British Colombia). Province-specific results can be found in Table S1.

Discussion

The main finding of this research is that odds of MDE are 18% higher for individuals living in urban compared to rural regions of Canada. This supports the idea that previous studies have lacked sufficient power to detect such a difference. While the data sets analyzed in prior Canadian studies were large (14,781 in the 1998 NPHS⁶ analysis and 31,321 in the 2002 CCHS analysis¹⁰), these studies may nevertheless have lacked power to detect a weak effect.

The main strength of this study is the increase in precision obtained by data pooling. Figure 1 demonstrates that 2 of the 9 individual surveys did not have sufficient power to detect the effect. In the provincial analysis, sample size was reduced further, supporting the use of data synthesis methods. Even with data pooling, there were nonsignificant associations observed in 5 of the 11 provinces. However, the direction of the association (higher in urban regions) was inherent for all provinces except British Colombia. This may result from a change in association across provinces or lack of power to detect the urban-rural difference. Limitations include inaccuracies of measurement inherent in fully structured interviews and the cross-sectional data sets. There is no implication that urban living causes depression, as other factors associated with urban living may be responsible for the difference. Our approach was to focus on burden, not etiology. Covariate adjustments could be used to help estimate the effect of urban living independent of other risk and prognostic factors, but such estimates do not inform an assessment of illness burden. Also, the difference could arise from differential mortality, prognosis, or migration. Longitudinal data are needed to address all of these concerns.

While the large sample size allows small effects to be detected statistically, the effects may be too weak to be important for policy or practice. An 18% increase in prevalence may indicate a need for more services but is likely to be most useful when combined with other determinants in a needs assessment study.