Background Few studies have examined the relationship between stressful social relations in private life and all-cause mortality.
Objective To evaluate the association between stressful social relations (with partner, children, other family, friends and neighbours, respectively) and all-cause mortality in a large population-based study of middle-aged men and women. Further, to investigate the possible modification of this association by labour force participation and gender.
Methods We used baseline data (2000) from The Danish Longitudinal Study on Work, Unemployment and Health, including 9875 men and women aged 36–52 years, linked to the Danish Cause of Death Registry for information on all-cause mortality until 31 December 2011. Associations between stressful social relations with partner, children, other family, friends and neighbours, respectively, and all-cause mortality were examined using Cox proportional hazards models adjusted for age, gender, cohabitation status, occupational social class, hospitalisation with chronic disorder 1980–baseline, depressive symptoms and perceived emotional support. Modification by gender and labour force participation was investigated by an additive hazards model.
Results Frequent worries/demands from partner or children were associated with 50–100% increased mortality risk. Frequent conflicts with any type of social relation were associated with 2–3 times increased mortality risk. Interaction between labour force participation and worries/demands (462 additional cases per 100 000 person-years, p=0.05) and conflicts with partner (830 additional cases per 100 000 person-years, p<0.01) was suggested. Being male and experiencing frequent worries/demands from partner produced 135 extra cases per 100 000 person-years, p=0.05 due to interaction.
Conclusions Stressful social relations are associated with increased mortality risk among middle-aged men and women for a variety of different social roles. Those outside the labour force and men seem especially vulnerable to exposure.
- SOCIAL EPIDEMIOLOGY
- Cohort studies
Statistics from Altmetric.com
The health protective effects of social relations on mortality risk are widely recognised and have been contrasted with the effects on health of more well-known risk factors for mortality, such as smoking, alcohol intake and obesity.1 ,2 However, social relations also have stressful aspects such as conflicts, demands, insensitivity and worries, with potential but less studied health damaging effects.3–5 Stressful aspects of social relations have previously been associated with physiological parameters detrimental to health, such as dysregulation of endocrine,6 cardiovascular7 ,8 and immune functioning.9 A higher risk of incident cardiovascular disease (CVD) has also been found among those who have conflicts or worries from social relations,7 ,10 ,11 but little is known about the relationship with all-cause mortality.5 In one of the few studies examining this, no association between adverse exchanges with the closest confidant and risk of all-cause and CVD mortality was found.12
Variabilities in social support, network size and stressful social relations exist, with the lowest support and highest levels of stress found in the lower socioeconomic groups.12 ,13 Furthermore, it has been hypothesized that people lower in the social hierarchy are particularly responsive to stressors as formulated in the theory of differential vulnerability.14 ,15 Differential vulnerability concerns the differential effect of a given stressor across social positions.14 ,16 It is suggested that the increased vulnerability of groups of lower socioeconomic position (SEP) is due to the lack of health-promoting coping strategies among people who have fewer socioeconomic and intrapsychic (eg, self-esteem) and social resources, such as access to guidance and support from others.15 ,17 The theory has been supported by previous studies,15 ,18–21—for example, McLeod and Kessler found higher social vulnerability towards several types of major personal events such as income loss, ill health, divorce and death of a loved one for those disadvantaged by income, education and occupational status.19 However, little is known about the differential effects of social relations on health outcomes across SEP. In one of the few studies (cross-sectional in design) to examine this, the poorest self-reported health outcomes was found among those in low SEP and with low social integration (ie, few social contacts), partly supporting the differential vulnerability hypothesis.22 These findings, however, remain to be replicated in a longitudinal design. In addition, to the best of our knowledge the interactive effect of SEP and stressful social relations on mortality has not previously been studied.
Gender differences are also described, with men generally having smaller and less diverse networks,23 ,24 while women seem more likely to have emotional support than men.25 Little is known about gender differences in stressful social relations. A previous study suggested that women have higher levels of demands/worries and conflicts with family and children than men, while no gender differences in experience of worries/demands and conflicts with partner were identified.11 It has been suggested that women are more vulnerable to strain from the family than men,26 but other studies have not supported such gender differences.27 Gender differences in vulnerability to social stressors may be driven by a combination of physiological, psychological and behavioural components.28 Women's stronger inclination to develop social support networks and their relatively greater dependency on enduring relationships may make them more vulnerable to life events or social stressors with negative interpersonal consequences,29 such as, for example, frequent conflicts or demands/worries. Accordingly, greater vulnerability to stressors has been found among women when mental health has been studied as the outcome.30 On the other hand, men exposed to stressors show greater reactivity to stress, as measured by greater increases in cortisol, than women.31 Such gender differences in social roles and stress reactivity makes it interesting to examine potential gender differences in the health consequences of social relations.5
The aim of this study was to evaluate the association between stressful social relations (with partner, children, other family, friends and neighbours, respectively) and all-cause mortality in a large population-based study of middle-aged men and women. Further, to investigate the possible modification of this association by labour force participation and gender.
In Spring 2000, 11 082 men and women aged either 40 or 50 years by 1 October 1999 were randomly selected from a 10% random sample of the Danish population (cohort 1). For the same study, 4145 individuals aged 36–52 years who had been receiving transfer income (due to unemployment, sickness benefit or disability pension) for more than 70% of the time during the 3 years before baseline in early Spring 2000 were randomly selected from the Danish population (cohort 2). Of these cohorts, 9875 completed a questionnaire (response rate: 69% (cohort 1) and 55% (cohort 2)). Five people were non-traceable in the Danish registries, leaving 9870 for inclusion in the study. Non-responders from both cohorts were more often men, immigrants from the third world or the Balkans, and were more often non-trained, semiskilled or unemployed. There was no difference between responders and non-responders in their contact with a general practitioner. The population has been described in detail previously.32 All individuals were linked to the Nationwide Danish Cause of Death Registry using a unique personal identification number for information on all-cause mortality until 31 December 2011.
For each measure of stressful social relations a subsample was constructed with complete information on the covariates included. For example, respondents without a partner (N=1490) were not included in the analysis of worries/demands from partner. Across the subsamples, 4% had missing measures of stressful social relations and 5% had missing information on occupational social class. The subsample sizes varied between 7393 and 8708 people.
Measures of stressful social relations
Stressful social relations were measured by the following questions ‘In your everyday life, do you experience that any of the following people demand too much of you or seriously worry you?’ and ‘In your everyday life, do you experience conflicts with any of the following people?’ with one item for each of the following social roles: partner, children (your own or partner’s), other family, friends and neighbours. The response key was ‘always’, ‘often’, ‘sometimes’, ‘seldom’, ‘never’, ‘have none’. In the main effects analyses the variable was kept in all response categories or the categories always/often were combined where the number of cases was small (<5). ‘Seldom’ was used as the reference category. Being exposed to stressful social relations ‘always’ or ‘often’ was considered the high risk condition and consequently, for the interaction analyses, the variables were dichotomised into ‘always/often’ versus ‘sometimes/seldom/never’.
The questions about social relations included in this study (Copenhagen Social Relations Questionnaire (CSRQ)) were developed based on the recommendations by Glass et al33 of including several social roles. The instrument covers both structural (eg, contact frequency and diversity) and functional aspects of social relations (eg, emotional support, instrumental support and demands/worries and conflicts). Content validity of the CSRQ was tested by focus group discussions and individual interviews with 31 informants. Another 94 men and women took part in an 8-day test–retest analysis. Informants generally stated that the questions and response categories were relevant and easy to understand. Furthermore, themes on the structure of social relations, social support and, of special relevance to this study, the stressful aspects of social relations, emerged clearly from the template analysis of the interviews. The reliability test showed moderate to excellent agreement for all items concerning stressful social relations (κ=0.41–0.82). In general, agreement was better for items dealing with the closest social relations such as partner and children (κ=0.74–82).34 In conclusion, the CSRQ shows satisfactory content and face validity as well as reliability and is suitable for measuring structure and function of social relations, including the negative aspects.34
Potential confounders for the analyses were identified based on prior knowledge and the method of directed acyclic graphs.35 This method provides a graphical tool for identification and selection of relevant confounders in epidemiological studies and thereby reduces the risk of biased results. The selected covariates included age, gender, occupational social class, cohabitation status, prior hospitalisation with chronic physical and mental disorders, depressive symptoms and emotional social support. Baseline social class was measured by occupational position and coded into classes I–V in accordance with the standards of the Danish National Institute of Social Research, which is similar to the British Registrar General's Classification I–V. We added social class VI representing people receiving transfer income, and for the descriptive table the variable was recoded into the following three categories: I–III (predominantly non-manual occupations); IV–V (predominantly skilled and unskilled manual occupations) and VI (transfer income due to unemployment, sickness leave or disability pension).
In the main effects analyses, the variable was included with all its original six categories. For the interaction analyses this variable was dichotomised into social classes I–V (employed) and social class VI (transfer income). Baseline cohabitation status was measured by the question: ‘Do you live alone?’, (yes/no). Hospitalisation with chronic diseases (yes/no) was included as a proxy for severe health problems before baseline. A participant’s personal identification number was linked to the Danish National Patient Registry for information on hospitalisation between 1 January 1980 and 30 June 2000 with any of the following diagnoses of chronic diseases: cancer, diabetes, CVD, chronic lung disease, chronic renal disease and mental disorders.
Depression was measured by the Major Depression Inventory based on 12 items of depressive symptoms (score range 0–50), which has been shown to have good validity.36 We included baseline depressive symptoms as a continuous variable in the multivariable analyses. In the descriptive analyses we categorised depressive symptoms into no (0–2), mild (3–8), moderate (9–19) and severe symptoms (20–50).37
Perceived emotional support was measured by the following question: ‘Can you talk with any of the following people, if you need support?’, including one item for each of the following social roles: partner, other family and friends. The response key was ‘always’, ‘often’, ‘sometimes’, ‘seldom’, ‘never’, ‘have none’. A summary index indicating the number of roles in the close (adult) social network who always/often provided emotional support was developed for the role’s partner, other family, friends; the index ranged from 0 to 3 for the number of social roles from whom emotional support is expected ‘always’ or ‘often’.
Depressive symptoms and emotional support may act both as confounders and mediators for the relation between stressful social relations and mortality and were consequently added to separate models.
To examine the effects on mortality of the combination of stress from partner and labour force participation new composite variables were created: (1) sometimes/seldom/never conflicts with partner and employed (I–V)(reference group); (2) sometimes/seldom/never conflicts with partner and receiving transfer income; (3) always/often conflicts with partner and employed; (4) always/often conflicts with partner and receiving transfer income. Similar variables were constructed for stress from partner and gender. Women with low levels of stress from social relations were used as the reference category.
Cox proportional hazards models with age as the underlying time variable were used to analyse the data. All variables met the proportional hazards assumption. Main effects of stressful social relations on all-cause mortality were initially estimated by HRs with 95% CIs. Model 1 included age and gender; model 2 was further adjusted for baseline occupational social class, cohabitation status and hospitalisation with chronic diseases before baseline; model 3 added depressive symptoms and model 4 further included perceived emotional support.
To assess the absolute associations of the combined variables with mortality, we used the additive hazards model, which is a flexible semiparametric model for survival outcomes.38 In that model, the hazard is modelled as a linear function of the explanatory variables and the estimates can therefore be directly interpreted as the number of additional cases associated with the explanatory variables—that is, the extra cases that exceed the combination of their individual effects. The underlying assumption of age-invariant associations was tested and no indication of violation was identified except for the analyses including worries/demands from partner where maximum follow-up time was limited to 64 years of age. There is continuing discussion about how to estimate interaction, but most health researchers would agree that interventions or preventive strategies should be aimed at those subgroups where most cases could be prevented.39 To identify such subgroups, deviation from additivity of absolute effects is the relevant measure of interest. When applying additive hazards models, the explanatory models are fitted to an additive scale, and product terms to assess deviation from additivity can be included and used to directly obtain the number of additional deaths due to interaction between labour force participation and gender, on the one hand, and stressful social relations, on the other.40 Departure from multiplicativity in the interaction analyses was tested by including a product term between stressful social relations and either gender or labour market attachment in separate models.
Additive hazards models were fitted using the software package R (through the package ‘timereg’) and all other analyses were conducted using PROC PHREG, SAS V.9.2
During 11 years of follow-up, 196 women (4%) and 226 men (6%) died. The major causes of death were cancer (47%), CVD (14%), liver disease due to alcohol abuse (8%) and violent deaths (accidents, suicide) (7%). Nine per cent of the population reported always/often experiencing demands/worries from their partner, 10% from children, 6% from family and 2% from friends. Six per cent of the population always or often experienced conflicts with their partner, 6% with their children, 2% with their family and 1% with friends. Table 1 shows the distribution of conflicts with partner by covariates. Always or often experiencing conflicts with their partner was associated with lower occupational social class, with living alone, with higher depressive symptoms and with lower access to emotional support. These patterns were roughly similar across different types of exposures—that is, demands/worries and conflicts with other parts of the social network (data not shown).
Worries and demands
Those who ‘always’ (HR=1.93; 95% CI 1.02 to 3.65) or ‘often’ (HR=1.81; 1.23 to 2.67) experienced worries and demands from partner had a higher mortality risk after adjustment for gender, social class, cohabitation status and prior hospitalisation than those who ‘seldom’ had this experience (Ptrend = 0.002). Always and often experiencing worries and demands from children was also associated with higher mortality risk (HR=1.55; 1.08 to 2.20), whereas worries and demands from other family, friends and neighbours were not (table 2).
Conflicts with any type of social relation were associated with higher mortality risk (Ptrend=0.001–0.04), and those who always or often experienced conflicts with their social relations were at markedly higher risk of premature death (table 3). For example, those who always or often experienced conflicts with their partner (HR=2.19; 95% CI 1.49 to 3.21) or friends (HR=2.63; 1.16 to 5.93) or who always experienced conflicts with neighbours (HR=3.07; 1.49 to 6.32) had higher mortality risk than those who seldom had such conflicts (table 3).
Generally, adjustment for depressive symptoms and emotional support at baseline only attenuated the associations slightly and did not change the overall conclusions. These findings suggest that none of these factors seem to strongly confound or mediate the associations. ‘Never’ experiencing social negativity from social relations was associated with a slightly increased mortality than those who ‘seldom’ have this experience except for never having conflicts with friends.
Modification by gender and labour force participation
Figure 1 shows the associations of the combined variable for stressful relations with partner and labour market participation with mortality. There is some support for the vulnerability hypothesis, as those who were jointly exposed to worries/demands (HR=3.38: 2.24 to 5.12) or conflicts (HR=4.52: 2.87 to 7.12) and not participating in the labour force were at markedly higher risk of all-cause mortality than those participating in the labour force with low stress from the partner. Further, the additive hazards models estimate that the joint exposure to demands/worries from the partner and being outside the labour force produces 462 additional cases/100 000 person-years owing to interaction (95% CI −6 to 930), p=0.05, meaning that those living on transfer income and who have frequent demands/worries from their partner have a higher mortality risk than expected from the individual effects of each of the exposures. Likewise, exposure both to frequent conflicts with the partner and being outside the labour force produces 830 additional cases/100 000 person-years (95% CI 166 to 1494), p<0.01.
Worries/demands from the partner also appeared to be modified by gender as men with many worries/demands from their partner seemed to have a higher mortality risk than expected from the individual effects of being a man and being exposed to worries/demands from the partner (HR=2.50, 95% CI 1.65 to 5.12) with 315 additional cases per 100 000 person-years (95% CI −28 to 658), p=0.05. The association between conflicts with partner and mortality was not modified by gender (figure 2). There was no support for multiplicative interaction in any of the analyses.
This study suggests that stressful social relations, ranging from partner to neighbours, are associated with mortality risk among middle-aged men and women. Conflicts, especially, were associated with higher mortality risk regardless of whom was the source of the conflict. Worries and demands were only associated with mortality risk if they were related to partner or children. Simultaneous exposure to high levels of stress from the partner and being outside the labour force appeared to be associated with an increased mortality risk, which lends some support for the theory of differential vulnerability.14 The finding is line with empirical studies suggesting an amplified effect of stressors among the most disadvantaged,19 ,20 ,22 and adds to the literature by focusing on stressful social relations in private life and all-cause mortality.
In this study, we found that men were especially vulnerable to frequent worries/demands from their partner, contradicting earlier findings suggesting that women were more vulnerable to stressful social relations,26 but in line with others which suggest that men respond to stressors with increased levels of cortisol, which may increase their risk of adverse health outcomes.31 In contrast, no gender differences in vulnerability to conflicts with the partner were identified in this study, as both men and women seemed to have increased mortality risk with high levels of conflicts. Further research into the possible gender differences in vulnerability to stressful social relations is needed in order to fully understand these mechanisms. Our findings are in contrast to previous findings which showed no association between adverse exchange with closest confidant and mortality.12 Discrepancy in findings may partly be explained by differences in categorisation of the exposure variable (dichotomous12 vs 4–5 levels) as dichotomisation might have blurred differences across levels. In line with our study, a higher risk of ischaemic heart disease-related outcomes has been suggested among people with high levels of demands/worries from partner, children and other family members and among those with high levels of conflict with partner and children.7 ,11
Confounding due to personality factors may be of concern. Personality has been shown to influence social relationships41 ,42 and mortality.43 Individuals scoring high on agreeableness have fewer conflicts, whereas those with high levels of neuroticism have a greater frequency of conflicts.41 ,42 High scores on neuroticism have been linked to increased mortality in several studies, whereas the link between agreeableness and mortality is less clear.43 If personality plays an important part in the way in which we perceive our social relations, certain personality traits may promote the reporting of any social relation as stressful, and therefore strong correlations between measures of stressful social relations would be expected. In this study only weak to moderate correlations were found (Pearson r between 0.14 and 0.54), which reduces the concern of bias. However, part of the association between stressful social relations and mortality might be ascribed to personality factors, which were unavailable.
Three main pathways have been suggested as leading from poor social relations to health outcomes: (1) health behaviour; (2) psychological well-being and (3) physiological reactions. Changes in health-related behaviour have been shown to partly explain the association between social relations and health outcomes.44 It has been more difficult to identify potential psychological pathways.45 Accordingly, we found that depressive symptoms did not explain any substantial part of the association between stressful social relations and mortality. There is relatively strong evidence for the association of social support with known physiological risk factors for increased mortality such as atherosclerosis and hypertension.44 The association between stressful relations and physiological pathways is less well-described in population-based studies. Stressful relations have been associated with heightened proinflammatory cytokine activity, which points to involvement of the immune system,9 and to poorer diurnal cortisol regulation,6 suggesting a possible pathway via hypothalamus pituitary adrenal axis involvement. It therefore seems plausible that at least part of the association between stressful relations and health might be mediated by stress-responsive systems.5
Based on prior knowledge and the methods of directed acyclic graphs, depressive symptoms and emotional support were identified as potential confounders of the association between stressful social relations and mortality. However, none of these factors seem to importantly confound the estimates. Since very few studies have investigated this association we were unable to compare these findings with earlier studies.
Strengths of this study include a large sample of middle-aged men and women with almost complete follow-up, and the use of validated measures of stressful social relations across a number of social roles, which allowed us to examine whether the association differed according to type and social role. Limitations include the rough dichotomisation of social class into those in and out of the labour force as well as items for stressful relations for the interaction analyses, which might have blurred possible interaction across social class groups and stress from social relations. Owing to non-response, the study population included fewer men, respondents with lower vocational training and unemployed people than the source population. Therefore caution should be taken in generalisation of our results to the general population.
In conclusion, stressful social relations are associated with increased mortality risk among middle-aged men and women. Demands and worries from the closest social relations such as partner and children seem more strongly related to mortality than worries and demands from more distant relations. Conflicts across all social roles from partner to neighbours were associated with higher mortality risk. Furthermore, being simultaneously exposed to both stress from the partner and being outside the labour force was associated with a markedly higher mortality risk. Likewise men seemed more vulnerable than women to worries and demands from their partner. Skills in handling worries and demands from close social relations as well as conflict management within couples and families and also in local communities may be important strategies for reducing premature deaths.
What is already known on this subject
The health protective effects of high social integration and access to social support on mortality risk are widely recognised and have been contrasted with the effects of well-known risk factors for mortality risk, such as smoking, alcohol intake and obesity.
Less is known about the health consequences of stressful aspects of social relations, such as conflicts, worries and demands.
Previously, a higher risk of incident cardiovascular outcomes has been found among those who have frequent conflicts or worries from social relations, but little is known about the relationship with all-cause mortality.
What this study adds
This study suggests that stressful social relations, ranging from partner to neighbours, are associated with mortality risk among middle-aged men and women.
Frequent conflicts were associated with higher mortality risk regardless of the person who was the source of the conflict. Worries and demands concerning partner and children were also associated with increased mortality risk.
Skills in handling worries and demands from close social relations as well as conflict management within couples and families and also in local communities may be considered important strategies for reducing premature deaths.
Contributors RL conceived the idea for the study, performed statistical analyses and wrote the first draft. UC contributed to the text and helped in the original data collection. CJN commented on the text and helped to interpret the results. MK performed the analyses for additive interaction in R and helped to interpret the results. NHR commented on the text and helped in interpretation of the analyses. She particularly contributed to the interpretation and analytical choices for the interaction analyses.
Funding This work was supported by The Danish Research Council grant number  and The Nordea Denmark Foundation grant number [02-2010-0385].
Competing interests None.
Provenance and peer review Not commissioned; externally peer reviewed.
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.