Social relationship correlates of major depressive disorder and depressive symptoms in Switzerland: nationally representative cross sectional study

The Swiss Health Survey is a periodic survey conducted by the Swiss Federal Office of Statistics. We utilized data from the 2007 Swiss Health Survey which included a representative sample of persons living in Switzerland. A random sample from an address pool of phone registered individuals was used for data collection. Four criteria had to be fulfilled to be included in the survey: a) living in Switzerland for more than three months, b) living in a private household, c) being older than 15 years, and d) speaking at least one of the three interview languages (German, French or Italian). Data collection was divided into four assessment periods to control seasonal influence.

Individuals were contacted by telephone and asked to participate in a telephone interview and secondly to complete a written questionnaire which was mailed to them after the initial interview. Most interviews were conducted by telephone although a face-to-face interview was offered in some cases (i.e., age > 74 years, severe somatic disease or accident or language difficulties). The data collection and data storage for the Swiss Health Survey (Schweizerische Gesundheitsbefragung) does not require formal approval by an ethical committee. This data collection is specifically permitted under Swiss law (Verordnung über die Durchführung von statistischen Erhebungen des Bundes vom 30. Juni 1993 (SR 431.012.1) and Verordnung über die eidgenössische Volkszählung vom 19. Dezember 2008 (SR 431.112.1)). Individuals invited to participate received a brief description of the study and could decline to participate or withdraw at any time. Participants’ responses were treated confidentially and aggregated anonymous responses were utilized for analyses presented herein. This secondary analysis of anonymous archival data was approved by the Institutional Review Board of Northern Arizona University.

A total of 18,760 individuals (66% of those contacted) took part in the interview and 77% (14,393) of those interviewed returned the written questionnaire. Census-based weights for the postal survey participants were used to represent the Swiss population. We restricted our analyses to postal survey participants 18 years or older (N = 14,016) who had no missing values for the depressive symptoms questionnaire (N = 12,286).

Major depressive disorder was assessed by using a short version of the Composite International Diagnostic Interview (CIDI-SF). The CIDI-SF is a fully structured diagnostic interview administered over the phone by trained lay interviewers. The CIDI-SF uses DSM-III-R [40] criteria to classify a major depressive episode in the last twelve months. Validation studies report good sensitivity and specificity with the full CIDI (89,6% and 93,9%, respectively) [38].

Depressive symptoms were assessed with the 10-item self-administered Depression Screening Questionnaire (DSQ) [39]. Items assess depressed mood, loss of interest, fatigue, weight change, sleep difficulties, slower movement or speech, decreased sexual desire, feeling worthless or guilty, trouble concentrating and thoughts of death or suicide over a period of the last two weeks using response options that range from no, never (0), to sometimes (1) or on most days (2). Scores can range from 0–20. Cronbach’s alpha for the DSQ in this sample was 0.82. Unlike some depressive symptom measures [41] the DSQ contains no loneliness items and thus is free of potential predictor-criterion overlap.

Social relationships were assessed with measures of social support and social integration. These items were developed specifically for this survey and capture a broad range of social relationship resources. Social support was represented by having a confidant (one item; “Among the people you are close to is there someone with whom you can always talk about really personal problems? No one, one person, many people”), having unmet support needs (one item; “Do you ever miss having someone to talk to about really personal problems?” Yes, no) and having tangible support (one item; “Do you have a family member or a neighbor you can always ask to help or support you with your daily activities?” No one, one person, many people). Social integration was assessed with two variables, marital status (one item; married/cohabiting vs. other) and the frequency of social contacts (five items). Social contact items included the frequency of meeting or phoning friends or relatives and participating in group activities. Respondents were asked: How often do you meet with friends or acquaintances at your home or theirs? How often do you meet with family members at your home or theirs? How often do you call or receive calls from your family members? How often do you call or receive calls from your friends or acquaintances? Response options for these items included Almost never or never (less than once per year); at least once a year; at least once a month; at least once a week; daily or almost daily. Frequency of participation in group activities or clubs (never, rarely, a few times a year, about once a month, about once a week and almost daily) was assessed with the question “How often do you participate in the activities of an association, club, political party or any other type of group?” These ordinal responses were coded 0–4 (0–5 for the group activities item) and summed and recoded into five categories for analysis (summary scores of 0–9 [referent], 10–12, 13–14, 15–16, and 17–21). This ordinal coding is comparable with other studies [19, 26]. The small cell sizes (range 6–50; M = 23) for the lowest contact frequency categories (0–4) precluded reliable comparisons across these groups.

To facilitate comparisons among social relationship regression coefficients we scaled confidant and tangible support to have a 0–1 range, i.e., instead of values of 0, 1 or 2 they were coded 0, 0.5, and 1.0 [42]. Ratio measures of association tend to be larger for binary versus polytomous variables and this scaling provides a more equal context for comparison of social relationship regression coefficients [42].

Among the three social relationship variables with more than two categories (loneliness, confidant, tangible support) loneliness did not meet an interval assumption in bivariate analyses [43]. Therefore we utilized two indicator variables to represent three levels of loneliness.

The distribution of depressive symptoms violated ordinary least squares (OLS) regression assumptions, e.g., symptoms were positively skewed and the variance was greater than the mean. We therefore analyzed the DSQ using negative binomial regression, exponentiating the coefficients to obtain incidence rate ratios (IRR). This approach preserves the interval properties of the scale and accommodates skewness and overdispersion [42]. Untransformed regression coefficients and statistical conclusions from negative binomial regressions were very similar to OLS regressions with square-root transformed DSQ values.

We estimated IRRs for binary major depressive disorder directly using a general linear model with a Poisson distribution and a log link function between covariates and the outcome. Robust standard errors were estimated to provide appropriate coverage for the confidence intervals [44]. The IRR, which reflects the ratio of incidence for the exposed group divided by the unexposed group, is the most general measure of association [45], provides accurate point and interval estimates [46] and is preferred to the odds ratio obtained from logistic regression [47, 48]. The magnitude of the IRR point estimate was generally smaller than the corresponding odds ratio but the direction and significance was almost identical to coefficients obtained from logistic regression (data not shown).

In two parallel sets of analyses we regressed major depressive disorder or depressive symptoms on social relationships. First, we entered every social relationship variable separately. This provided an estimate of the strength of association for each individual social relationship indicator. Second, we simultaneously entered all social relationship variables to determine their independence and to estimate which had the strongest association when adjusting for the others. In the third step we entered indicator variables representing age, gender, education, regional language, and nationality to determine whether these sociodemographic factors altered the association of social relationship variables with the depression outcomes. Analyses were considered statistically significant if the two-tailed p-value was less than or equal to 0.05. All analyses were conducted with Stata 11.2 (Stata Corp., College Station, Texas USA) and variance estimates incorporated the survey design, i.e., sampling strata and weights. Survey-based IRR’s were slightly different from those calculated directly from the unweighted frequencies shown in Table 1.

In bivariate analyses each of the six social relationship variables was associated with major depressive disorder in the expected direction. When all six social relationship variables were entered simultaneously only loneliness, unmet support, and the fourth quintile of social contacts remained significant (Table 3). These patterns persisted when sociodemographic variables were entered, with marital status becoming significant in that model. Loneliness had the strongest association across all models and when we excluded loneliness from the regression the magnitude of association for all other social support variables increased and were statistically significant except for the second and fifth quintiles of the social contacts variable.

All social relationship variables were associated with depressive symptoms when analyzed individually, which is in line with the depressive episode analyses. In contrast to depressive episodes analyses, marital status was not significantly associated with depressive symptoms when all six variables were entered simultaneously. Although loneliness and unmet support had the strongest associations, each social contact frequency category was inversely associated with depressive symptoms, with the benefit of increased social contact frequency leveling off at the two highest quintiles. This pattern of independence among social relationship variables persisted following statistical adjustment for sociodemographic variables (Table 3). When we excluded loneliness from the regression all other social relationship variables, including marital status, were significant and were of somewhat larger magnitude (Table 3).

We conducted exploratory analyses to partition the very large association of loneliness with depressive symptoms. We estimated mean depressive symptoms within the three loneliness categories to allow a more clinically meaningful interpretation. Mean depressive symptoms differed strongly according to loneliness, increasing significantly from never (M = 2.6 [95% CI, 2.6 – 2.7]) to sometimes (M = 4.4 [95% CI, 4.2 – 4.5]) to quite/very often (M = 8.2 [95% CI, 7.6 – 8.9]).

The present study utilized a large, probability sample that represented young adults, a group particularly vulnerable to major depression. We assessed depression with interview and questionnaire methods and we evaluated their association with a large number of variables representing two broad, theoretically important social relationship domains. While not a focus of the present study, we also incorporated “upstream” social and contextual measures in our analyses to more precisely characterize the unique contribution of social relationships [15]. To our knowledge this is one of a handful of nationally representative studies in this area and the first one from Europe including adults 18 years and older. Although Switzerland has a relatively low base rate of loneliness relative to other countries [59] the associations we observed are consistent with prior research utilizing more restricted samples.

This study is also subject to a number of limitations. There are several ways to define and measure social relationships and our study did not assess potentially important domains such as social strain [34], relationship reciprocity or marital satisfaction. The cross-sectional design of our study cannot address the causal direction of the association, i.e., whether lower social support and less social integration cause depression or the other way around. There is evidence for both perspectives. For example, we know that humans are motivated to form and maintain social attachments [14], that social exclusion is painful [60], and that theory predicts that depressive symptoms should be a consequence of low social relationship resources [12, 15]. Longitudinal studies confirm that greater social relationship resources are prospectively associated with fewer depressive symptoms [19, 33, 54, 61], less probable incident depression [32] and better recovery from major depression even when controlling for initial depression levels [20, 62]. Conversely, among adolescents there is evidence that depression is associated with subsequent increases in loneliness [63] and dysfunctional attributions regarding social resources can elicit depression [64] but may resolve after the depressive episode [65]. Low self esteem or low social skills are etiologically relevant for both loneliness and depression [66‐68]. Of course, one can also be depressed without feeling lonely [16] and vice-versa. While acknowledging these reciprocal possibilities it is reasonable to conceptualize low social relationship resources as determinants of depression [69] and as plausible targets for intervention [70]. Future studies with longitudinal measurements of depression and social relationships will help clarify the presence and magnitude of these reciprocal influences (cf. [19, 71]).