- Email: [email protected]
The depression and marital status relationship is modified by both age and gender
Journal of Affective Disorders 223 (2017) 65–68
Contents lists available at ScienceDirect
Journal of Affective Disorders journal homepage: www.elsevier.com/locate/jad
The depression and marital status relationship is modified by both age and gender
MARK
⁎
Andrew G.M. Bullocha,b,c, , Jeanne V.A. Williamsa, Dina H. Lavoratoa, Scott B. Pattena,b,c a b c
Department of Community Health Sciences, University of Calgary, Calgary, Alberta, Canada Department of Psychiatry, University of Calgary, Calgary, Alberta, Canada Mathison Centre for Mental Health Research & Education, Hotchkiss Brain Institute, University of Calgary, Calgary, Alberta, Canada
A R T I C L E I N F O
A B S T R A C T
Keywords: Major depressive episode Epidemiological studies Cross-sectional studies Meta-analysis Meta-regression
Background: Marital status is associated with major depression prevalence, however, the strength of association may be modified by age and gender. Methods: The data sources were a series of cross sectional national health surveys of the Canadian population carried out by Statistics Canada during 1996–2013. These were cross-sectional files from the National Population Health Survey of 1996, together with the Canadian Community Health Surveys from 2000 to 2013; the respondents were 18 years and older. The data was analyzed with meta-analytic techniques and logistic regression. Results: In terms of gender, the odds ratios of depression were smaller for females (vs males) who were single, widowed or separated compared to married people. Regarding age, the odds ratios for depression showed a steady rise with increasing age for those in single and in common-law relationships compared to married people. In contrast the odds ratios for depression declined with age for those widowed, separated and divorced compared to married people. The strength of the interaction terms used to quantify these moderating effects showed no change from 1996 to 2013. Limitations: Only one member of each household was included, so that relationship issues could not be studied. The generalizability of our findings requires international data. Also the diagnostic interviews used are not as accurate as clinical assessments. Conclusion: Use of large numbers of participants has revealed some robust modifying effects of both gender and age on the depression/marital status relationship. The clinical significance of our findings is that the vulnerability to development of depression is not only related to marital status, but that this relationship is modified by age and gender.
1. Introduction According to the Global Burden of Disease Study 2015 major depression is a leading cause of years lived with disability globally, coming third after lower back and neck pain, and sense organ diseases, respectively (GBD, 2015 Collaborators). Major depression can be the cause or the result of social, psychological and biological factors such as age and sex Studies have repeatedly shown that the prevalence of depression is higher in women than men worldwide (Kessler et al., 2015). In terms of age, the prevalence of depression decreases steadily with advancing age, and although depression is more prevalent in young women compared to men, this prevalence difference decreases with age and in no longer evident in people over 75 years old (Patten et al., 2016). One key social factor that modifies depression is marital status, and
⁎
research has repeatedly shown that married people have better mental health than those who are single, widowed, separated and divorced (Bebbington 1987; Jang et al., 2009; Bulloch et al., 2009; LaPierre, 2009). It has also been shown that the relationship between marital disruption and depression is bidirectional (Bulloch et al., 2009). That being married is protective for depression is not an exclusively Western phenomenon as depression is also relatively high in unmarried Japanese (Inaba et al., 2005). A meta-analysis of people > (or = ) 55 years old showed that being unmarried is a significant risk factor for depression in late life (Yan et al., 2011). However in Koreans aged 75–85 years no relationship between marital status and depression was found for women, whereas divorced and widowed men had higher rates of depressive symptoms than their married counterparts (Jang et al., 2009). Taken together these studies show that there is a complex relationship between age, sex, marital status and depression. In this study
Corresponding author at: Department of Community Health Sciences, TRW 4D67, 3280 Hospital Drive NW, Calgary, Canada T2N 4Z6. E-mail address: [email protected] (A.G.M. Bulloch).
http://dx.doi.org/10.1016/j.jad.2017.06.007 Received 2 September 2016; Received in revised form 20 April 2017; Accepted 11 June 2017 Available online 04 July 2017 0165-0327/ © 2017 Elsevier B.V. All rights reserved.
Journal of Affective Disorders 223 (2017) 65–68
A.G.M. Bulloch et al.
Table 1 Sample sizes in each survey – Age 18 plus.
Table 2 Depression and age-sex interactions.
Surveys
Total N = 878,965
Depressed
Not depressed
NPHS 1996 CCHS 1.1 CCHS 1.2 mental health CCHS 2.1 CCHS 3.1 CCHS 07/08 CCHS 09/10 CCHS 11/12 CCHS 12 mental health CCHS 2013
67,705 115,977 35,216 118,876 118,206 117,894 111,077 112,377 23,846 57,791
2921 8842 1852 2682 3562 2616 2954 1230 1236 1106
62,211 104,131 33,176 40,188 58,515 38,513 48,616 18,024 22,455 17,301
Depression
ONE STEP – LOGISTIC REGRESSION
2 STEP
OR
OR
P-value
95% CI Age (continuous) Sex (female) Age*Sex interaction
0.98 0.98–0.99 2.18 1.99–2.39 0.99 0.99–1.00
P-value
95% CI < 0.001 < 0.001 < 0.001
0.98 0.98–0.99 2.20 2.01–2.41 0.99 0.99–1.00
< 0.001 < 0.001 < 0.001
a
Marital Status Common-law/ partner Single
we examined how age and sex modify modify the association of marital status with major depression as an outcome. Interaction terms in regression models were used to quantify the modification observed, such that the term interaction is used in the remainder of this paper to refer to the resulting subgroup variations. Use of a very large data set enabled us to examine married and common-law categories separately, whereas these are usually combined in existing studies. Similarly we were able to study widowed, separated and divorced categories which are also often combined.
Widowed Separated Divorced Interactions (sex) (female) Single*sex
2. Methods
Widow*sex
This study used the cross-sectional data files collected in an early cycle of the National Population Health Survey (1996), the general health surveys of the Canadian Community Health Survey (2000, 2003, 2005, 2007/2008, 2009/2010, 2011/2012 and 2013), plus the two mental health Canadian Community Health Survey of 2002 and 2012 (Table 1). These surveys used a complex multistage sampling procedure to obtain a representative sample of the Canadian population. First geographical clusters were selected, then households were selected within the clusters and finally one respondent per household was selected. The data analyzed here is for those 18+ years old. In the NPHS survey and in most of the CCHS surveys past year Major Depressive Episode (MDE) was assessed with an abbreviated fully structured diagnostic interview, the Composite International Diagnostic Interview Short form for Major Depression (CIDI-SFMD) (Kessler et al., 1998). The CIDI-SFMD was developed from the National Comorbidity Survey with receiver operator analysis designed to detect a subset of CIDI items most highly predictive of MDE. In addition, the requirement for a total of 5 symptoms, including one of the two cardinal symptoms in the DSM definition provide face validity for the instrument. The two mental health CCHS surveys used a more comprehensive interview that is a Canadian adaptation of the World Health Organization Mental Health (WMH)-CIDI (Kessler and Ustun, 2004). The WMH-CIDI is the most widely used fully structured diagnostic instrument in psychiatric epidemiology and produces similar estimates to the CIDI-SFMD (Patten et al., 2015). We analyzed the data with individual-level meta-analysis using both 1-step and 2-step approaches, using the former approach in the primary analysis and the latter as a means of confirming the robustness of the results. In the 1-step approach a pooled data set was created from all the surveys for direct derivation of estimates. This is called a one-step approach since after pooling the data can be analyzed using non-metaanalytic procedures. In this approach, replicate bootstrap weights were rescaled for use in the pooled data set (Statistics Canada provides sets of 500 such weights for each survey to account for the complex design effects of these surveys). The combined sample size of the surveys was n = 878,965. The replicate bootstrap weights were rescaled by dividing the number of participants in each individual survey by 878,965 and multiplying the weights by this proportion. Weighted logistic regression with bootstrapped derived variance estimation was then performed.
Separated*sex Interactions (age) Common-law*age Single*age Widow*age Separated*age Divorced*age
0.91 0.81–1.02 1.42 1.29–1.58 19.35 12.61–29.69 6.56 5.20–8.28 3.58 3.01–4.25
0.099
0.87 0.80–0.95 0.64 0.53–0.79 0.70 0.58–0.84
0.002
1.01 1.01–1.02 1.02 1.01–1.02 0.96 0.95–0.97 0.99 0.98–0.99 0.99 0.99–1.00
< 0.001
< 0.001 < 0.001 < 0.001 < 0.001
< 0.001 < 0.001
< 0.001 < 0.001 < 0.001 < 0.001
0.93 0.82–1.06 1.43 1.30–1.59 15.84 11.50–21.83 6.44 4.80–8.66 3.75 3.17–4.44
0.258
0.87 0.78–0.96 0.68 0.55–0.84 0.74 0.63–0.87
0.006
1.01 1.01–1.02 1.02 1.01–1.02 0.96 0.96–0.97 0.99 0.98–1.00 0.99 0.99–1.00
< 0.001
< 0.001 < 0.001 < 0.001 < 0.001
< 0.001 < 0.001
< 0.001 < 0.001 0.003 < 0.001
Notes. The table gives the impression that the 2-step estimates are from a single model. Rather these are pooled estimates of model parameters from analyses of the individual surveys. Where the ORs = 1.0, they are < 1.0 to 3 decimal places. a Married is the baseline group, the other marital status categories are dummy coded using 0 or 1 values.
The results are expressed as Odds Ratios (ORs) with 95% Confidence Intervals (CIs). For the two-step approach beta coefficients in logistic regression models were estimated from each survey (step 1) and then pooled (step 2) using meta-analytic methods. Possible interactions were examined by inclusion of cross-product terms of both marital status and sex, and also marital status and age. All models included age as age since 18 years (the minimum age of eligibility in these surveys), in order to create an interpretable intercept term in the models. Random effects meta-regression was used to test for changes over time (i.e., time was a variable included in the models). These analyses used the “metan” command in Stata version 13 [23] and were conducted in the Prairie Regional Data Centre of Statistics Canada at the University of Calgary. This research was approved by the Ethics Review Board of the University of Calgary. 3. Results First we performed a 1-step analysis on the pooled data from all 10 surveys. We were able to include all 5 categories of marital status (Table 2) and we calculated the odds ratios of depression in each category, and estimated the influence of sex and age. A significant 66
Journal of Affective Disorders 223 (2017) 65–68
A.G.M. Bulloch et al.
step analysis were reproduced in the 2-step analysis (Table 2) which helps to confirm their robustness. Since some of the latter have wider CIs they are more conservative and may be considered the principal findings of our study since the replicate sampling weights used in each estimate were unmodified and therefore may more effectively adjust for design effects. In addition, an interaction between age and sex is evident (Table 2). This odds ratio being less than 1.0 mean that the odds ratio of depression in females vs males decreases with age. We previously reported a detailed analysis of this finding recently, showing the sex difference in depression decreases steadily with age and is no longer evident in those over 75 (Patten et al., 2016). We further tested the reliability of our results by running models with age categories instead of age as a continuous variable. Supplementary Tables 1 and 2 confirmed most of the interactions we reported above, and Supplementary Table 3 with supplementary Fig. 1A,B confirmed the findings of this model. Use of a very large database gave us the power to study married and common-law categories separately, which typically have been combined. Likewise we were able to separately study the widowed, separated and divorced categories that are often combined. Our large sample size provided statistical power to detect interactions that may have been missed by prior studies. In terms of sex, we have shown that being female is less strongly associated with MDE in those single, widowed and separated compared to those who are married (Table 2). Females have larger and stronger social support networks than men, whereas men often report their wives to be their chief source of social support (McLaughlin et al., 2010; Pugliesi and Shook, 2009); this may explain why single, widowed and divorced females are less vulnerable to development of depression that males. In a prior study of those 65 plus in Europe a sex by marital status interaction was reported in which the sex difference in depression was the lowest in never married and greatest in married people (Prince et al., 1999). This contrasts with our result for single people (Table 2) in which no interaction occurred. This difference is likely attributable to the age range used in the study by Prince (65 plus) whereas we used 18 plus. Regarding age, the odds ratios of depression increases with age for single people and those in common-law relationships compared to those who are married. It is well known that married people have the lowest prevalence of depression (see Introduction), but to our knowledge it has not been reported before that common-law and single people share an increasing vulnerability to depression with age. It may be that young people are more socialized to unmarried status, i.e., that being young and single is socially acceptable. Age may be protective for depression in those widowed, separated and divorced, perhaps reflecting the reality that older people are more socialized to unmarried status and/or improved coping with losses in older people. Descriptive epidemiological data are valuable for generating hypotheses for research and for quantifying the burden of health conditions in different sub-groups within a population. While we have provided some possible interpretations of our findings in the preceding two paragraphs it should be emphasized that cross-sectional data cannot confirm causality. Also, whereas a strength of our study is its representation of the population, such survey data have limitations. For example, since only one respondent was selected from each household, only one member of a relationship could be studied. Many other modifying factors, e.g. religion, coping strategies, living arrangements were not addressed in our study. One limitation of our findings lies in reliance on diagnostic measures that are not as accurate as a clinical assessment. The CIDI-SFMD in particular is a brief instrument designed for use in large-scale surveys. It does not have the ability to fully assess the characteristics and severity of the symptoms it assesses. As its inaccuracy is not likely to be related to the variables assessed (age, sex, marital status), inaccurate assessment of MDE would be expected to dilute the observed associations, which may therefore be even stronger than reported here. Also the instrument does not differentiate mild from moderate or severe
interaction was found for age and sex (OR = 0.99, 95% CI 0.99–1.00) indicating that the effect of sex became weaker with age. Similarly significant interactions were found for all 5 marital statuses, providing statistical evidence that the effect of marital status differs depending on sex. In terms of sex, males were the referent group and for marital status married was the referent group. No interactions with sex were found for those common-law or divorced, indicating that the effect of these marital conditions is similar in men and women. Significant interactions were found for sex and being single (OR = 0.87, 95% CI 0.80–0.95), being widowed (OR = 0.64, 95% CI 0.53–0.79) and being separated (OR = 0.70, 95% CI 0.58–0.84), i.e., the odds ratios were smaller for females than males in these categories of marital status. In terms of age we used age as a continuous variable in the primary analysis, the odds ratios therefore represent changes over one year of age. Significant ORs greater than 1.0 were observed for interactions of age and those in common-law (OR = 1.01, 95% CI 1.01–1.02) and single (OR = 1.02, 95% CI 1.01–1.02) categories compared to those who were married, i.e. the association of MDE with these marital status categories, as depicted by the odds ratio, increased with age. Significant ORs less than 1.0 were observed for interactions of age and the categories of widowed (OR = 0.96, 95% CI 0.95–0.97), separated (OR = 0.99, 95% CI 0.98–0.99) and divorced (OR = 0.99, 95% CI 0.99–1.0) compared to those who were married, i.e., the strength of association of MDE with these marital status categories decreased with age. To test the robustness of the results from the 1-step method we performed a 2-step analysis as described in the Methods. The same interactions were identified although some small changes were evident for some point estimates and some of the CIs were wider, i.e., these estimates are more conservative (Table 2). The strength of the female sex/marital status interactions were: single (OR = 0.87, 95% CI 0.78–0.96), widowed (OR = 0.68, 95% CI 0.55–0.84) and separated (OR = 0.74, 95% CI 0.63–0.87). For the age/marital status interactions the point estimates from the 2-step model were identical to those of the 1-step model. The strength of these interactions were: common-law by age (OR = 1.01, 95% CI 1.01–1.02), single by age (OR = 1.02, 95% CI 1.01–1.02) widowed by age (OR = 0.96, 95% CI 0.96–0.97), separated by age (OR = 0.99, 95% CI 0.98–0.99) and divorced by age (OR = 0.99, 95% CI 0.99–1.0). When we tested whether the strength of interactions changed over time from 1996 to 2013 by running meta-regression models including time as a variable, no significant changes in any of the interactions over time were observed (data not shown). We further tested the robustness of our results by categorizing age as 18–44, 45–64 and 65 + (referent group) (in contrast to age being a continuous variable as above). Our initial results are shown in Supplementary Table 1 which has the same variables as Table 2. Comparison of these tables shows that most of the interactions remain when age is categorized. The two exceptions are single*sex and common-law*age. When the non-significant interactions were removed from the model a final model was produced that is shown as Supplementary Table 2. To further test the accuracy of our one-step model with age categories we estimated depression prevalence rates and compared them to the fitted values from the model. The estimated prevalence of depression in men and women by age and marital status is shown in Supplementary Table 3. These results highlight the expected higher prevalence in women and in widowed, separated and divorced people. Supplementary Figure 1A,B compares the estimated prevalences with fitted values from the model. A remarkable degree of agreement is apparent in this comparison. The only exception is that the model overestimated the prevalence in widowed men aged 18–44. 4. Discussion Here we report a series of interactions of age and sex with different categories of marital status. All of the interactions found in the initial 167
Journal of Affective Disorders 223 (2017) 65–68
A.G.M. Bulloch et al.
whom had any involvement the planning or execution of the research.
episodes, so differences related to severity could not be assessed. The significance of our findings lies in understanding how the vulnerability to development of depression is not only related to marital status, but that this relationship may be modified by sex and age. The preliminary nature of our data make it more of scientific interest and perhaps less relevant to the practicing physician. However these results raise some interesting hypotheses for future studies. Higher prevalence could be due to an effect on risk or duration of depression, so these results raise a lot of questions – e.g. what are the biological, psychological or social factors that underpin the associations? Now that these interactions have been identified, their origins should be examined by future studies.
Appendix A. Supporting information Supplementary data associated with this article can be found in the online version at http://dx.doi.org/10.1016/j.jad.2017.06.007. References Bebbington, P., 1987. Marital status and depression: a study of English national admission statistics. Acta Psychiatr. Scand. 75, 640–650. Bulloch, A.G.M., Williams, J.V.A., Lavorato, D.H., Patten, S.B., 2009. The relationship between major depression and marital disruption and is bidirectional. Depression Anxiety 26, 1172–1177. GBD 2015 Disease and Injury Incidence and Prevalence Collaborators, 2015. Global, regional, and national incidence, prevalence, and years lived with disability for 310 acute and chronic diseases and injuries in 188 countries, 1990–2015: a systematic analysis for the Global Burden of Disease Study. Lancet 388, 1545–1562. Inaba, A., Thoits, P.A., Ueno, K., Gove, W.R., Evenson, R.J., Sloan, M., 2005. Depression in the United States and Japan: Gender, marital status, and SES patterns. Soc. Sci. Med. 61, 2280–2292. Jang, S.-N., Kawachi, I., Chang, J., Boo, K., Shin, H.-G., Lee, H., Cho, S.-i., 2009. Marital status, gender, and depression: analysis of the baseline survey of the Korean longitudinal study of ageing (KLoSA). Soc. Sci. Med. 69, 1608–1615. Kessler, R.C., Andrews, G., Mroczek, D., Ustun, B., Wittchen, H.U., 1998. The World Health Organization composite international diagnostic interview short-form (CIDISF). Int J. Methods Psychiatr. Res. 7, 171–185. Kessler, R.C., Ustun, T.B., 2004. The World mental health (WMH) survey initiative version of the World Health Organization (WHO) composite international diagnostic interview (CIDI). Int. J. Methods Psychiatr. Res. 13, 83–121. Kessler, R.C., Sampson, N.A., Berglund, P., Gruber, M.J., 2015. Anxious and non-anxious major depressive disorder in the World Health Organization Mental Health Surveys. Epidemiol. Psychiatr. Sci. 24, 210–226. LaPierre, T.A., 2009. Marital status and depressive symptoms over time: age and gender variations. Family Relations 58, 404–416. McLaughlin, D., Vagenas, D., Pachana, N.B., Dobson, A., 2010. Gender differences in social network size and satisfaction in adults in their 70s. J. Health Psychol. 15, 671–679. Patten, S.B., Williams, J.V., Lavorato, D.H., Fiest, K.M., Bulloch, A.G., Wang, J., 2015. The prevalence of major depression is not changing. Can J. Psych. 60, 31–35. Patten, S.B., Williams, J.V.A., Lavorato, D.H., Wang, J.L., Bulloch, A.G.M., Sajobi, T., 2016. The association between major depression and sex becomes weaker with age. Soc. Psychiatry Psychiatr. Epidemiol. 51, 203–210. Prince, M.J., Beekman, A.T.F., Deeg, D.J.H., Fuhrer, R., Kivela, S.-L., Lawlor, B.A., Lobo, A., Magnusson, H., Meller, I., Oyen, H., Van, Reischies, F., Roelands, I., Skoog, I., Turrina, C., Copeland, J.R.M., 1999. Depression symptoms in late life assessed using the EURO-D scale. Effect of age, gender and marital status in 14 European centres. Br. J. Psychiatry 174, 339–345. Pugliesi, K., Shook, S., 2009. Gender, ethnicity and network characteristics: variation in social support resources. Sex. Roles 38, 215–238.
Author disclosures Contributors Authors Williams and Lavorato conducted the analysis of the Statistics Canada data. Authors Bulloch and Patten wrote the manuscript. All authors contributed to and have approved the final manuscript. Role of funding This work was supported by an operating grant from the Canadian Institutes of Health Research (MOP-130415), the Hotchkiss Brain Institute and the Alberta Mental Health and Addictions Strategic Clinical Network, none of whom had any involvement the planning or execution of the research. Acknowledgements The analysis was conducted at the Prairie Regional Data Centre, a part of the Canadian Research Data Centre Network (CRDCN). The services provided by the CRDCN are made possible by the financial or in-kind support of the SSHRC, the CIHR, the CFI, Statistics Canada and participating universities. The views expressed in this paper do not necessarily reflect the views of CRDCN or of its partners. This work was supported by an operating grant from the Canadian Institutes of Health Research (MOP-130415), the Hotchkiss Brain Institute and the Alberta Mental Health and Addictions Strategic Clinical Network, none of
68
Contents lists available at ScienceDirect
Journal of Affective Disorders journal homepage: www.elsevier.com/locate/jad
The depression and marital status relationship is modified by both age and gender
MARK
⁎
Andrew G.M. Bullocha,b,c, , Jeanne V.A. Williamsa, Dina H. Lavoratoa, Scott B. Pattena,b,c a b c
Department of Community Health Sciences, University of Calgary, Calgary, Alberta, Canada Department of Psychiatry, University of Calgary, Calgary, Alberta, Canada Mathison Centre for Mental Health Research & Education, Hotchkiss Brain Institute, University of Calgary, Calgary, Alberta, Canada
A R T I C L E I N F O
A B S T R A C T
Keywords: Major depressive episode Epidemiological studies Cross-sectional studies Meta-analysis Meta-regression
Background: Marital status is associated with major depression prevalence, however, the strength of association may be modified by age and gender. Methods: The data sources were a series of cross sectional national health surveys of the Canadian population carried out by Statistics Canada during 1996–2013. These were cross-sectional files from the National Population Health Survey of 1996, together with the Canadian Community Health Surveys from 2000 to 2013; the respondents were 18 years and older. The data was analyzed with meta-analytic techniques and logistic regression. Results: In terms of gender, the odds ratios of depression were smaller for females (vs males) who were single, widowed or separated compared to married people. Regarding age, the odds ratios for depression showed a steady rise with increasing age for those in single and in common-law relationships compared to married people. In contrast the odds ratios for depression declined with age for those widowed, separated and divorced compared to married people. The strength of the interaction terms used to quantify these moderating effects showed no change from 1996 to 2013. Limitations: Only one member of each household was included, so that relationship issues could not be studied. The generalizability of our findings requires international data. Also the diagnostic interviews used are not as accurate as clinical assessments. Conclusion: Use of large numbers of participants has revealed some robust modifying effects of both gender and age on the depression/marital status relationship. The clinical significance of our findings is that the vulnerability to development of depression is not only related to marital status, but that this relationship is modified by age and gender.
1. Introduction According to the Global Burden of Disease Study 2015 major depression is a leading cause of years lived with disability globally, coming third after lower back and neck pain, and sense organ diseases, respectively (GBD, 2015 Collaborators). Major depression can be the cause or the result of social, psychological and biological factors such as age and sex Studies have repeatedly shown that the prevalence of depression is higher in women than men worldwide (Kessler et al., 2015). In terms of age, the prevalence of depression decreases steadily with advancing age, and although depression is more prevalent in young women compared to men, this prevalence difference decreases with age and in no longer evident in people over 75 years old (Patten et al., 2016). One key social factor that modifies depression is marital status, and
⁎
research has repeatedly shown that married people have better mental health than those who are single, widowed, separated and divorced (Bebbington 1987; Jang et al., 2009; Bulloch et al., 2009; LaPierre, 2009). It has also been shown that the relationship between marital disruption and depression is bidirectional (Bulloch et al., 2009). That being married is protective for depression is not an exclusively Western phenomenon as depression is also relatively high in unmarried Japanese (Inaba et al., 2005). A meta-analysis of people > (or = ) 55 years old showed that being unmarried is a significant risk factor for depression in late life (Yan et al., 2011). However in Koreans aged 75–85 years no relationship between marital status and depression was found for women, whereas divorced and widowed men had higher rates of depressive symptoms than their married counterparts (Jang et al., 2009). Taken together these studies show that there is a complex relationship between age, sex, marital status and depression. In this study
Corresponding author at: Department of Community Health Sciences, TRW 4D67, 3280 Hospital Drive NW, Calgary, Canada T2N 4Z6. E-mail address: [email protected] (A.G.M. Bulloch).
http://dx.doi.org/10.1016/j.jad.2017.06.007 Received 2 September 2016; Received in revised form 20 April 2017; Accepted 11 June 2017 Available online 04 July 2017 0165-0327/ © 2017 Elsevier B.V. All rights reserved.
Journal of Affective Disorders 223 (2017) 65–68
A.G.M. Bulloch et al.
Table 1 Sample sizes in each survey – Age 18 plus.
Table 2 Depression and age-sex interactions.
Surveys
Total N = 878,965
Depressed
Not depressed
NPHS 1996 CCHS 1.1 CCHS 1.2 mental health CCHS 2.1 CCHS 3.1 CCHS 07/08 CCHS 09/10 CCHS 11/12 CCHS 12 mental health CCHS 2013
67,705 115,977 35,216 118,876 118,206 117,894 111,077 112,377 23,846 57,791
2921 8842 1852 2682 3562 2616 2954 1230 1236 1106
62,211 104,131 33,176 40,188 58,515 38,513 48,616 18,024 22,455 17,301
Depression
ONE STEP – LOGISTIC REGRESSION
2 STEP
OR
OR
P-value
95% CI Age (continuous) Sex (female) Age*Sex interaction
0.98 0.98–0.99 2.18 1.99–2.39 0.99 0.99–1.00
P-value
95% CI < 0.001 < 0.001 < 0.001
0.98 0.98–0.99 2.20 2.01–2.41 0.99 0.99–1.00
< 0.001 < 0.001 < 0.001
a
Marital Status Common-law/ partner Single
we examined how age and sex modify modify the association of marital status with major depression as an outcome. Interaction terms in regression models were used to quantify the modification observed, such that the term interaction is used in the remainder of this paper to refer to the resulting subgroup variations. Use of a very large data set enabled us to examine married and common-law categories separately, whereas these are usually combined in existing studies. Similarly we were able to study widowed, separated and divorced categories which are also often combined.
Widowed Separated Divorced Interactions (sex) (female) Single*sex
2. Methods
Widow*sex
This study used the cross-sectional data files collected in an early cycle of the National Population Health Survey (1996), the general health surveys of the Canadian Community Health Survey (2000, 2003, 2005, 2007/2008, 2009/2010, 2011/2012 and 2013), plus the two mental health Canadian Community Health Survey of 2002 and 2012 (Table 1). These surveys used a complex multistage sampling procedure to obtain a representative sample of the Canadian population. First geographical clusters were selected, then households were selected within the clusters and finally one respondent per household was selected. The data analyzed here is for those 18+ years old. In the NPHS survey and in most of the CCHS surveys past year Major Depressive Episode (MDE) was assessed with an abbreviated fully structured diagnostic interview, the Composite International Diagnostic Interview Short form for Major Depression (CIDI-SFMD) (Kessler et al., 1998). The CIDI-SFMD was developed from the National Comorbidity Survey with receiver operator analysis designed to detect a subset of CIDI items most highly predictive of MDE. In addition, the requirement for a total of 5 symptoms, including one of the two cardinal symptoms in the DSM definition provide face validity for the instrument. The two mental health CCHS surveys used a more comprehensive interview that is a Canadian adaptation of the World Health Organization Mental Health (WMH)-CIDI (Kessler and Ustun, 2004). The WMH-CIDI is the most widely used fully structured diagnostic instrument in psychiatric epidemiology and produces similar estimates to the CIDI-SFMD (Patten et al., 2015). We analyzed the data with individual-level meta-analysis using both 1-step and 2-step approaches, using the former approach in the primary analysis and the latter as a means of confirming the robustness of the results. In the 1-step approach a pooled data set was created from all the surveys for direct derivation of estimates. This is called a one-step approach since after pooling the data can be analyzed using non-metaanalytic procedures. In this approach, replicate bootstrap weights were rescaled for use in the pooled data set (Statistics Canada provides sets of 500 such weights for each survey to account for the complex design effects of these surveys). The combined sample size of the surveys was n = 878,965. The replicate bootstrap weights were rescaled by dividing the number of participants in each individual survey by 878,965 and multiplying the weights by this proportion. Weighted logistic regression with bootstrapped derived variance estimation was then performed.
Separated*sex Interactions (age) Common-law*age Single*age Widow*age Separated*age Divorced*age
0.91 0.81–1.02 1.42 1.29–1.58 19.35 12.61–29.69 6.56 5.20–8.28 3.58 3.01–4.25
0.099
0.87 0.80–0.95 0.64 0.53–0.79 0.70 0.58–0.84
0.002
1.01 1.01–1.02 1.02 1.01–1.02 0.96 0.95–0.97 0.99 0.98–0.99 0.99 0.99–1.00
< 0.001
< 0.001 < 0.001 < 0.001 < 0.001
< 0.001 < 0.001
< 0.001 < 0.001 < 0.001 < 0.001
0.93 0.82–1.06 1.43 1.30–1.59 15.84 11.50–21.83 6.44 4.80–8.66 3.75 3.17–4.44
0.258
0.87 0.78–0.96 0.68 0.55–0.84 0.74 0.63–0.87
0.006
1.01 1.01–1.02 1.02 1.01–1.02 0.96 0.96–0.97 0.99 0.98–1.00 0.99 0.99–1.00
< 0.001
< 0.001 < 0.001 < 0.001 < 0.001
< 0.001 < 0.001
< 0.001 < 0.001 0.003 < 0.001
Notes. The table gives the impression that the 2-step estimates are from a single model. Rather these are pooled estimates of model parameters from analyses of the individual surveys. Where the ORs = 1.0, they are < 1.0 to 3 decimal places. a Married is the baseline group, the other marital status categories are dummy coded using 0 or 1 values.
The results are expressed as Odds Ratios (ORs) with 95% Confidence Intervals (CIs). For the two-step approach beta coefficients in logistic regression models were estimated from each survey (step 1) and then pooled (step 2) using meta-analytic methods. Possible interactions were examined by inclusion of cross-product terms of both marital status and sex, and also marital status and age. All models included age as age since 18 years (the minimum age of eligibility in these surveys), in order to create an interpretable intercept term in the models. Random effects meta-regression was used to test for changes over time (i.e., time was a variable included in the models). These analyses used the “metan” command in Stata version 13 [23] and were conducted in the Prairie Regional Data Centre of Statistics Canada at the University of Calgary. This research was approved by the Ethics Review Board of the University of Calgary. 3. Results First we performed a 1-step analysis on the pooled data from all 10 surveys. We were able to include all 5 categories of marital status (Table 2) and we calculated the odds ratios of depression in each category, and estimated the influence of sex and age. A significant 66
Journal of Affective Disorders 223 (2017) 65–68
A.G.M. Bulloch et al.
step analysis were reproduced in the 2-step analysis (Table 2) which helps to confirm their robustness. Since some of the latter have wider CIs they are more conservative and may be considered the principal findings of our study since the replicate sampling weights used in each estimate were unmodified and therefore may more effectively adjust for design effects. In addition, an interaction between age and sex is evident (Table 2). This odds ratio being less than 1.0 mean that the odds ratio of depression in females vs males decreases with age. We previously reported a detailed analysis of this finding recently, showing the sex difference in depression decreases steadily with age and is no longer evident in those over 75 (Patten et al., 2016). We further tested the reliability of our results by running models with age categories instead of age as a continuous variable. Supplementary Tables 1 and 2 confirmed most of the interactions we reported above, and Supplementary Table 3 with supplementary Fig. 1A,B confirmed the findings of this model. Use of a very large database gave us the power to study married and common-law categories separately, which typically have been combined. Likewise we were able to separately study the widowed, separated and divorced categories that are often combined. Our large sample size provided statistical power to detect interactions that may have been missed by prior studies. In terms of sex, we have shown that being female is less strongly associated with MDE in those single, widowed and separated compared to those who are married (Table 2). Females have larger and stronger social support networks than men, whereas men often report their wives to be their chief source of social support (McLaughlin et al., 2010; Pugliesi and Shook, 2009); this may explain why single, widowed and divorced females are less vulnerable to development of depression that males. In a prior study of those 65 plus in Europe a sex by marital status interaction was reported in which the sex difference in depression was the lowest in never married and greatest in married people (Prince et al., 1999). This contrasts with our result for single people (Table 2) in which no interaction occurred. This difference is likely attributable to the age range used in the study by Prince (65 plus) whereas we used 18 plus. Regarding age, the odds ratios of depression increases with age for single people and those in common-law relationships compared to those who are married. It is well known that married people have the lowest prevalence of depression (see Introduction), but to our knowledge it has not been reported before that common-law and single people share an increasing vulnerability to depression with age. It may be that young people are more socialized to unmarried status, i.e., that being young and single is socially acceptable. Age may be protective for depression in those widowed, separated and divorced, perhaps reflecting the reality that older people are more socialized to unmarried status and/or improved coping with losses in older people. Descriptive epidemiological data are valuable for generating hypotheses for research and for quantifying the burden of health conditions in different sub-groups within a population. While we have provided some possible interpretations of our findings in the preceding two paragraphs it should be emphasized that cross-sectional data cannot confirm causality. Also, whereas a strength of our study is its representation of the population, such survey data have limitations. For example, since only one respondent was selected from each household, only one member of a relationship could be studied. Many other modifying factors, e.g. religion, coping strategies, living arrangements were not addressed in our study. One limitation of our findings lies in reliance on diagnostic measures that are not as accurate as a clinical assessment. The CIDI-SFMD in particular is a brief instrument designed for use in large-scale surveys. It does not have the ability to fully assess the characteristics and severity of the symptoms it assesses. As its inaccuracy is not likely to be related to the variables assessed (age, sex, marital status), inaccurate assessment of MDE would be expected to dilute the observed associations, which may therefore be even stronger than reported here. Also the instrument does not differentiate mild from moderate or severe
interaction was found for age and sex (OR = 0.99, 95% CI 0.99–1.00) indicating that the effect of sex became weaker with age. Similarly significant interactions were found for all 5 marital statuses, providing statistical evidence that the effect of marital status differs depending on sex. In terms of sex, males were the referent group and for marital status married was the referent group. No interactions with sex were found for those common-law or divorced, indicating that the effect of these marital conditions is similar in men and women. Significant interactions were found for sex and being single (OR = 0.87, 95% CI 0.80–0.95), being widowed (OR = 0.64, 95% CI 0.53–0.79) and being separated (OR = 0.70, 95% CI 0.58–0.84), i.e., the odds ratios were smaller for females than males in these categories of marital status. In terms of age we used age as a continuous variable in the primary analysis, the odds ratios therefore represent changes over one year of age. Significant ORs greater than 1.0 were observed for interactions of age and those in common-law (OR = 1.01, 95% CI 1.01–1.02) and single (OR = 1.02, 95% CI 1.01–1.02) categories compared to those who were married, i.e. the association of MDE with these marital status categories, as depicted by the odds ratio, increased with age. Significant ORs less than 1.0 were observed for interactions of age and the categories of widowed (OR = 0.96, 95% CI 0.95–0.97), separated (OR = 0.99, 95% CI 0.98–0.99) and divorced (OR = 0.99, 95% CI 0.99–1.0) compared to those who were married, i.e., the strength of association of MDE with these marital status categories decreased with age. To test the robustness of the results from the 1-step method we performed a 2-step analysis as described in the Methods. The same interactions were identified although some small changes were evident for some point estimates and some of the CIs were wider, i.e., these estimates are more conservative (Table 2). The strength of the female sex/marital status interactions were: single (OR = 0.87, 95% CI 0.78–0.96), widowed (OR = 0.68, 95% CI 0.55–0.84) and separated (OR = 0.74, 95% CI 0.63–0.87). For the age/marital status interactions the point estimates from the 2-step model were identical to those of the 1-step model. The strength of these interactions were: common-law by age (OR = 1.01, 95% CI 1.01–1.02), single by age (OR = 1.02, 95% CI 1.01–1.02) widowed by age (OR = 0.96, 95% CI 0.96–0.97), separated by age (OR = 0.99, 95% CI 0.98–0.99) and divorced by age (OR = 0.99, 95% CI 0.99–1.0). When we tested whether the strength of interactions changed over time from 1996 to 2013 by running meta-regression models including time as a variable, no significant changes in any of the interactions over time were observed (data not shown). We further tested the robustness of our results by categorizing age as 18–44, 45–64 and 65 + (referent group) (in contrast to age being a continuous variable as above). Our initial results are shown in Supplementary Table 1 which has the same variables as Table 2. Comparison of these tables shows that most of the interactions remain when age is categorized. The two exceptions are single*sex and common-law*age. When the non-significant interactions were removed from the model a final model was produced that is shown as Supplementary Table 2. To further test the accuracy of our one-step model with age categories we estimated depression prevalence rates and compared them to the fitted values from the model. The estimated prevalence of depression in men and women by age and marital status is shown in Supplementary Table 3. These results highlight the expected higher prevalence in women and in widowed, separated and divorced people. Supplementary Figure 1A,B compares the estimated prevalences with fitted values from the model. A remarkable degree of agreement is apparent in this comparison. The only exception is that the model overestimated the prevalence in widowed men aged 18–44. 4. Discussion Here we report a series of interactions of age and sex with different categories of marital status. All of the interactions found in the initial 167
Journal of Affective Disorders 223 (2017) 65–68
A.G.M. Bulloch et al.
whom had any involvement the planning or execution of the research.
episodes, so differences related to severity could not be assessed. The significance of our findings lies in understanding how the vulnerability to development of depression is not only related to marital status, but that this relationship may be modified by sex and age. The preliminary nature of our data make it more of scientific interest and perhaps less relevant to the practicing physician. However these results raise some interesting hypotheses for future studies. Higher prevalence could be due to an effect on risk or duration of depression, so these results raise a lot of questions – e.g. what are the biological, psychological or social factors that underpin the associations? Now that these interactions have been identified, their origins should be examined by future studies.
Appendix A. Supporting information Supplementary data associated with this article can be found in the online version at http://dx.doi.org/10.1016/j.jad.2017.06.007. References Bebbington, P., 1987. Marital status and depression: a study of English national admission statistics. Acta Psychiatr. Scand. 75, 640–650. Bulloch, A.G.M., Williams, J.V.A., Lavorato, D.H., Patten, S.B., 2009. The relationship between major depression and marital disruption and is bidirectional. Depression Anxiety 26, 1172–1177. GBD 2015 Disease and Injury Incidence and Prevalence Collaborators, 2015. Global, regional, and national incidence, prevalence, and years lived with disability for 310 acute and chronic diseases and injuries in 188 countries, 1990–2015: a systematic analysis for the Global Burden of Disease Study. Lancet 388, 1545–1562. Inaba, A., Thoits, P.A., Ueno, K., Gove, W.R., Evenson, R.J., Sloan, M., 2005. Depression in the United States and Japan: Gender, marital status, and SES patterns. Soc. Sci. Med. 61, 2280–2292. Jang, S.-N., Kawachi, I., Chang, J., Boo, K., Shin, H.-G., Lee, H., Cho, S.-i., 2009. Marital status, gender, and depression: analysis of the baseline survey of the Korean longitudinal study of ageing (KLoSA). Soc. Sci. Med. 69, 1608–1615. Kessler, R.C., Andrews, G., Mroczek, D., Ustun, B., Wittchen, H.U., 1998. The World Health Organization composite international diagnostic interview short-form (CIDISF). Int J. Methods Psychiatr. Res. 7, 171–185. Kessler, R.C., Ustun, T.B., 2004. The World mental health (WMH) survey initiative version of the World Health Organization (WHO) composite international diagnostic interview (CIDI). Int. J. Methods Psychiatr. Res. 13, 83–121. Kessler, R.C., Sampson, N.A., Berglund, P., Gruber, M.J., 2015. Anxious and non-anxious major depressive disorder in the World Health Organization Mental Health Surveys. Epidemiol. Psychiatr. Sci. 24, 210–226. LaPierre, T.A., 2009. Marital status and depressive symptoms over time: age and gender variations. Family Relations 58, 404–416. McLaughlin, D., Vagenas, D., Pachana, N.B., Dobson, A., 2010. Gender differences in social network size and satisfaction in adults in their 70s. J. Health Psychol. 15, 671–679. Patten, S.B., Williams, J.V., Lavorato, D.H., Fiest, K.M., Bulloch, A.G., Wang, J., 2015. The prevalence of major depression is not changing. Can J. Psych. 60, 31–35. Patten, S.B., Williams, J.V.A., Lavorato, D.H., Wang, J.L., Bulloch, A.G.M., Sajobi, T., 2016. The association between major depression and sex becomes weaker with age. Soc. Psychiatry Psychiatr. Epidemiol. 51, 203–210. Prince, M.J., Beekman, A.T.F., Deeg, D.J.H., Fuhrer, R., Kivela, S.-L., Lawlor, B.A., Lobo, A., Magnusson, H., Meller, I., Oyen, H., Van, Reischies, F., Roelands, I., Skoog, I., Turrina, C., Copeland, J.R.M., 1999. Depression symptoms in late life assessed using the EURO-D scale. Effect of age, gender and marital status in 14 European centres. Br. J. Psychiatry 174, 339–345. Pugliesi, K., Shook, S., 2009. Gender, ethnicity and network characteristics: variation in social support resources. Sex. Roles 38, 215–238.
Author disclosures Contributors Authors Williams and Lavorato conducted the analysis of the Statistics Canada data. Authors Bulloch and Patten wrote the manuscript. All authors contributed to and have approved the final manuscript. Role of funding This work was supported by an operating grant from the Canadian Institutes of Health Research (MOP-130415), the Hotchkiss Brain Institute and the Alberta Mental Health and Addictions Strategic Clinical Network, none of whom had any involvement the planning or execution of the research. Acknowledgements The analysis was conducted at the Prairie Regional Data Centre, a part of the Canadian Research Data Centre Network (CRDCN). The services provided by the CRDCN are made possible by the financial or in-kind support of the SSHRC, the CIHR, the CFI, Statistics Canada and participating universities. The views expressed in this paper do not necessarily reflect the views of CRDCN or of its partners. This work was supported by an operating grant from the Canadian Institutes of Health Research (MOP-130415), the Hotchkiss Brain Institute and the Alberta Mental Health and Addictions Strategic Clinical Network, none of
68