Introduction
Mental stress is associated with increased mortality risk (Russ et al. Reference Russ, Stamatakis, Hamer, Starr, Kivimaki and Batty2012; Prior et al. Reference Prior, Fenger-Grøn, Larsen, Larsen, Robinson and Mortensen2016), but it remains unclear whether this association is causal or mediated by what caused the stress. Previous survey-based studies agree that there is a crude association between poor mental well-being and higher risk of death (Chida & Steptoe, Reference Chida and Steptoe2008; Russ et al. Reference Russ, Stamatakis, Hamer, Starr, Kivimaki and Batty2012; Rutters et al. Reference Rutters, Pilz, Koopman, Rauh, Te Velde and Stehouwer2014; Liu et al. Reference Liu, Floud, Pirie, Green, Peto and Beral2016; Prior et al. Reference Prior, Fenger-Grøn, Larsen, Larsen, Robinson and Mortensen2016). However, the association attenuates after adjusting for disease burden, lifestyle, and socioeconomic factors; this suggests that mental stress may simply be a marker of an underlying poor physical health prognosis. Yet, as these studies do not have the optimal study design to control for important confounders and elucidate causality, the conclusions remain controversial (Liu et al. Reference Liu, Floud, Pirie, Green, Peto and Beral2016). A profound understanding of causality is important as only interventions targeting causal risk factors may benefit the patients (Mercer et al. Reference Mercer, Gunn, Bower, Wyke and Guthrie2012).
Randomization to mental stress would reduce the risk of confounding, but it may be unethical and impossible to perform such experiments. However, a ‘natural’ experiment using a well-defined external mental stressor may be conducted; this would allow for estimating the effect of mental stress on physical health with only limited risk of reverse causality. The loss of a spouse is an event of severe mental stress (Holmes & Rahe, Reference Holmes and Rahe1967; Byrne & Raphael, Reference Byrne and Raphael1997; Ryckebosch-Dayez et al. Reference Ryckebosch-Dayez, Zech, Mac Cord and Taverne2016), independently of the health status of the bereaved, and well-established in the literature (Stroebe et al. Reference Stroebe, Schut and Stroebe2007; Shah et al. Reference Shah, Carey, Harris, DeWilde, Victor and Cook2012; Shor et al. Reference Shor, Roelfs, Curreli, Clemow, Burg and Schwartz2012; Moon et al. Reference Moon, Glymour, Vable, Liu and Subramanian2014). So far, no studies have addressed the risk of death after bereavement while taking mental-physical multimorbidity into account.
We aimed to estimate the effect of spousal bereavement as an indicator of mental stress on mortality by utilizing a ‘natural’ experiment design in a large population-based cohort taking mental and physical multimorbidity into account.
Methods
Study population and design
We performed a population-based individually matched cohort study among Danish citizens aged 18 years or older by comparing individuals bereaved (spousal loss) between 1 January 1997 and 1 January 2014 with reference individuals. Prospectively recorded data from nationwide Danish registries were linked at the individual level using a unique personal identification number. All participants were followed until death, emigration, or 1 January 2014 (end of study), whichever came first.
Exposure and matching
We obtained data on age, sex, cohabitation status, and date of spousal loss from the spouse's death certificate stored in the Danish Civil Registration System (Pedersen et al. Reference Pedersen, Gotzsche, Moller and Mortensen2006). On the day of the loss (index date), the bereaved individuals were each matched with five reference individuals on sex and date of birth. Individuals could be selected as references in several matching strata. References had not lost a spouse within the past 5 years and were residing in Denmark on the index date. After 5 years, bereaved individuals could be included as references again to retain a natural background reference population. References were categorized as living as couples (married or cohabitating) or singles, and changes in their cohabitation status were allowed over time. If references were bereaved themselves, their status changed into bereaved and a new matched cohort stratum was created.
Multimorbidity was defined as having two or more chronic conditions (Van Den Akker et al. Reference Van Den Akker, Buntinx and Knottnerus1996) out of 39 selected physical and mental conditions; this information was based on International Classification of Diseases version 10 (ICD-10) discharge diagnoses registered from hospital admissions and out-patient contacts in all Danish hospitals (Carstensen et al. Reference Carstensen, Kristensen, Marcussen and Borch-Johnsen2011; Gjerstorff, Reference Gjerstorff2011; Lynge et al. Reference Lynge, Sandegaard and Rebolj2011; Mors et al. Reference Mors, Perto and Mortensen2011) combined with data on dispensed drug prescriptions in Danish pharmacies (Kildemoes et al. Reference Kildemoes, Sorensen and Hallas2011) from 1995 onwards (online Supplementary Table e1). The algorithm for disease definitions was based on established comorbidity indices and recommendations in systematic reviews; it has previously been described in detail (Prior et al. Reference Prior, Fenger-Grøn, Larsen, Larsen, Robinson and Mortensen2016). The disease status for each condition was assessed on the index date for both bereaved and reference individuals.
Outcomes
Our primary outcome was all-cause mortality obtained from the Danish Civil Registration System (Pedersen et al. Reference Pedersen, Gotzsche, Moller and Mortensen2006). As a secondary outcome, we assessed mortality by natural causes (i.e. due to disease), excluding deaths due to accidents, suicides, and homicide as registered in the Danish Register of Causes of Deaths (Juel & Helweg-Larsen, Reference Juel and Helweg-Larsen1999).
Other covariates
Information on the highest achieved educational level on the index date was obtained from Statistics Denmark (Jensen & Rasmussen, Reference Jensen and Rasmussen2011) and categorized according to the UNESCO classification: <10 years, 10–15 years, or >15 years of education or training (United Nations Educational, Scientific & Cultural Organization 2011).
Statistical analysis
The time axis of the study was time since index date, which was subdivided into time periods (0–1, 1–2, 2–3, 3–6, 6–12 months; 1–2, 2–5, 5–10, ⩾10 years) for period-dependent analyses of the association between bereavement and mortality for the bereaved individuals. Mortality rates (incidence proportions of death) and Kaplan–Meier cumulative incidence proportions (CIPs) for all-cause mortality were calculated for bereaved individuals. We fitted Cox proportional hazards models stratified by matching strata to produce hazard ratios (HRs) for all-cause and cause-specific (natural causes, accidents, or suicides) mortality. In these models, bereaved individuals were compared with reference individuals adjusted for living as a single, thus making individuals living as a couple the reference group. Individual-level cluster robust variance estimation was used to calculate 95% confidence intervals (CIs). The main analyses used complete case analysis because educational level was the only variable with missing information in the registers, whereas the supplementary analyses utilized multiple imputations of missing educational level values in five imputation sets based on our analysis parameters (White & Royston, Reference White and Royston2009). The proportional hazards assumption was evaluated using log [−log (survival)] plots for each time period, and no obvious violations were found.
The primary Cox regression analyses were performed using two adjustment models. Our first model included only adjustment for sex and date of birth, which was intrinsic to the matching and stratification approach. In the second model, we also adjusted for each of the 39 physical and mental conditions as individual indicator variables as well as highest achieved educational level. To assess potential effect measure modification from mental-physical multimorbidity, we stratified the bereaved individuals and matched references by physical disease count (0, 1 to 2, and ⩾3) and by presence of any mental condition (0 or ⩾1).
To compare the mortality of subgroups graphically, the y-axis for the fully adjusted HRs was rescaled by risk-time weighted average mortality HRs for each mental-physical multimorbidity group corrected for age, sex, and educational level. We estimated the excess mortality rate by multiplying the mortality rate among bereaved by the adjusted attributable fraction, (HR-1)/HR (Rothman et al. Reference Rothman, Greenland and Lash2008), with 95% CIs estimated by a normal-based bootstrap procedure.
Sub-analyses
To further assess potential effect measure modifiers, we stratified the cohort by sex, age at bereavement (<65 v. ⩾65 years), and mental condition status (bipolar affective disorders, schizophrenia, or dementia v. other mental condition v. no mental condition). As a measure of the expectedness of spouse death, a Charlson Comorbidity Index (CCI) (Charlson et al. Reference Charlson, Pompei, Ales and MacKenzie1987) score of the deceased spouse one year prior to the index date was compiled using discharge diagnoses (Lynge et al. Reference Lynge, Sandegaard and Rebolj2011) as the CCI score is known to predict the 1-year mortality (Charlson et al. Reference Charlson, Pompei, Ales and MacKenzie1987). To assess the sensitivity of our matching procedure, we reran the model on a dataset in which references had never been bereaved in the study period. All sub-analyses were fully adjusted for the 39 conditions and educational level.
All p values were two-sided with statistical significance set at p < 0.05. Analyses were performed using Stata 13.1 (StataCorp, TX).
Ethical approval
The study was based on anonymized data located at Statistics Denmark and was approved by the Danish Data Protection Agency (ref. no. 2013-41-1719).
Results
We identified 389 316 bereaved individuals of whom 137 247 died during the study period of up to 17 years (median follow-up time: 6.1 years). Two-thirds of the bereaved were women (Table 1). On the index date, 37.5% of reference individuals were single and 62.5% were cohabitating or married.
Matching baseline characteristics
a Mood, stress-related, or anxiety disorders, alcohol or substance abuse, or anorexia/bulimia.
b Bipolar affective disorders, schizophrenia, or dementia.
Bereaved individuals had higher all-cause mortality than non-bereaved individuals living as couples. The HR was particularly high during the first month (HR 2.50; 95% CI 2.37 to 2.63) after the loss when adjusting for sex and date of birth, but it decreased and remained at a stable level of about one third above the reference individuals 1 year after the loss and throughout the study period (Table 2, HR). In absolute numbers, the excess mortality rate associated with bereavement within the first month was 3.12 deaths per 1000 person-months; these are theoretically avoidable deaths if the bereaved individuals had the same mortality risk as the references with comparable morbidity. This number declined, but it remained elevated during follow-up (Table 2, Excess mortality rate). Being single was also associated with higher HR during the study period compared with being part of a couple (HR 1.42; 95% CI 1.41–1.43).
All-cause mortality for bereaved individuals v. couples by time since bereavement
a Matched on sex and date of birth, adjusted for single status.
b Further adjusted for 39 mental and physical conditions, and educational level.
c Cumulative incidence proportion (CIP) for bereaved individuals at the end of the respective time interval per 1000 persons.
d Mortality rate for bereaved individuals in the respective time interval per 1000 person-months.
e Mortality rate for bereaved individuals × (adjusted HR – 1)/adjusted HR – the proportion of deaths that theoretically could be avoided if the risk in the bereavement group equalled to that of the reference group.
The relative all-cause mortality associated with bereavement attenuated with time since the loss across all mental-physical multimorbidity groups (Table 3, Adjusted HR). The excess mortality rate associated with bereavement rose with increasing physical multimorbidity (1.33 v. 7.00 deaths per 1000 person-months within the first month for individuals with 0 v. ⩾3 physical conditions, respectively). Mental comorbidity further added to the excess mortality rate in all groups; the highest bereavement-associated absolute mortality was present in the group with 1–2 physical conditions in addition to mental comorbidity (7.80 deaths per 1000 person-months within the first month; Table 3, Excess mortality rate). Figure 1 visualizes the relative effects of bereavement while incorporating the background risk of each multimorbidity group.
Adjusted all-cause mortality hazard ratios for bereaved individuals v. couples by time since bereavement and mental-physical multimorbidity.
All-cause mortality for bereaved individuals v. couples stratified by time since bereavement and mental-physical multimorbidity
a Cumulative incidence proportion for bereaved individuals at the end of the respective time interval per 1000 persons.
b Mortality rate for bereaved individuals in the respective time interval per 1000 person per month.
c Matched on sex and date of birth, adjusted for single status, 39 mental and physical conditions, and educational level.
d Mortality rate for bereaved individuals × (adjusted HR – 1)/adjusted HR – the proportion of deaths that theoretically could be avoided if the risk in the bereavement group equalled that of the reference group.
Bereavement was associated with increased risk of death due to unnatural causes (e.g. accidents, suicides), particularly within the first few months following spousal death (online Supplementary Table e2). Yet, 96% of deaths in the bereaved individuals had natural causes (97% in reference individuals). When excluding deaths from unnatural causes, the mortality HRs for natural causes were generally similar to the HRs of all-cause mortality in terms of multimorbidity and time since bereavement (online Supplementary Fig. e1). The most common causes of death for bereaved individuals in the study period were cardiovascular disease and chronic obstructive pulmonary disease, even in the first month after bereavement. The causes of death were comparable between bereaved and references.
In subgroup analyses, unexpected loss of a spouse was associated with high mortality early after bereavement (online Supplementary Fig. e2). Bereavement appeared to affect the early HR more in men than in women (online Supplementary Fig. e3) and in those bereaved under age 65 (online Supplementary Fig. e4). Mental illness did not modify the relative risk of death after bereavement in general, but the confidence limits were wide for the group of bereaved individuals with bipolar affective disorders, schizophrenia, or dementia (online Supplementary Fig. e5).
Our sensitivity analyses, which used individuals who had never been bereaved in the study period as references, showed no systematic differences from our primary analysis (online Supplementary Fig. e6). Multiple imputation of missing values for educational level produced a pattern similar to the primary analysis, but we found slightly lower HRs because the oldest individuals tended to have both the most missing data on education and the highest comorbidity burden (online Supplementary Fig. e7).
Discussion
In this large population-based cohort study, mental stress from bereavement was associated with higher mortality. Notably, while the risk of mortality appeared to be particularly high shortly after the loss of a spouse, it remained elevated for at least 17 years even after adjusting for mental and physical comorbidities. Bereavement was associated with higher mortality in individuals with both mental and physical health problems, but the excess mortality was particularly high in those with mental-physical multimorbidity at the time of the loss. Furthermore, most deaths associated with bereavement were due to natural causes. Our ‘natural’ experiment cautiously suggests that a causal link between mental stress and mortality may be present.
The effect of widowhood on mortality has been described previously (Stroebe et al. Reference Stroebe, Schut and Stroebe2007; Moon et al. Reference Moon, Kondo, Glymour and Subramanian2011). To our knowledge, this study is the first to investigate the impact of bereavement or any other stressful life event on mortality in a whole population while taking into account mental-physical multimorbidity.
Strengths and limitations
We obtained information on all spousal losses in Denmark during 17 years with no loss to follow-up, which makes selection bias an unlikely explanation for our results. Information on civil and vital status, diagnoses, and educational level was well-validated, and the data were prospectively recorded in national registries (Pedersen et al. Reference Pedersen, Gotzsche, Moller and Mortensen2006; Carstensen et al. Reference Carstensen, Kristensen, Marcussen and Borch-Johnsen2011; Gjerstorff, Reference Gjerstorff2011; Jensen & Rasmussen, Reference Jensen and Rasmussen2011; Kildemoes et al. Reference Kildemoes, Sorensen and Hallas2011; Lynge et al. Reference Lynge, Sandegaard and Rebolj2011; Mors et al. Reference Mors, Perto and Mortensen2011), which reduced the risk of differential information bias. All information on variables had clear time points, which ensured that outcome did not precede exposure, and no information relied on the reporting from participants or their relatives, which minimized the risk of reverse causality and recall bias and legitimized our ‘natural’ experiment. The large sample size allowed us to evaluate the association between bereavement and death according to time since bereavement in short time spans and in subgroups defined by gender, age, and medical history. The matched cohort design effectively reduced confounding from matching covariates, i.e. sex, age, and calendar period, and we further adjusted the estimates for a broad range of comorbidities and socioeconomic factors. Unfortunately, Danish administrative registers lack information on chronic disease treated only in primary care. To counter this limitation, the disease definitions in our multimorbidity index used information on all filled prescriptions of medications for chronic conditions commonly treated by primary care physicians. We had no data on the personal attachment between the deceased and the bereaved, which is likely to have modified the stress level and the effect of bereavement. Shared health behaviour and lifestyle between spouses could create a selection effect, but selection effects are modest in bereavement studies (Boyle et al. Reference Boyle, Feng and Raab2011) and controlling for diseases and educational level reduces this effect. Furthermore, there was a close temporal relation between spousal death and death of the bereaved individual in our study; chronic diseases were responsible for most deaths of the bereaved individuals, but take years to develop so it is unlikely that e.g. shared lifestyle would be responsible for the death of the bereaved individual shortly following bereavement. Individual characteristics, such as coping strategies, social network, and lifestyle factors like smoking and exercise level, could also be modifying factors. However, lifestyle changes following bereavement would be mediators of the observed associations and adjusting for them would thus be inappropriate (Rothman et al. Reference Rothman, Greenland and Lash2008). We note that observational studies generally carry an inherent risk of residual confounding.
Interpretation
The bi-directional relationship between mental and physical health is well-documented (Katon, Reference Katon2003). Allostatic load theory provides a basis for the relation between psychological and physiological stress; it describes stress as a response mediated by the neurological, endocrine, and immune systems (McEwen, Reference McEwen1998; Buckley et al. Reference Buckley, Sunari, Marshall, Bartrop, McKinley and Tofler2012). Additionally, unhealthy lifestyle choices in relation to the loss of a spouse could lead to worsening of disease or development of new disease (Rutters et al. Reference Rutters, Pilz, Koopman, Rauh, Te Velde and Stehouwer2014). However, this interaction is complex, and it would be almost impossible to fully disentangle the causal mechanisms and the relative impact of the mental component on mortality. Longitudinal studies combining administrative data on mortality and survey data on mental well-being [e.g. psychological distress (Russ et al. Reference Russ, Stamatakis, Hamer, Starr, Kivimaki and Batty2012), perceived stress (Prior et al. Reference Prior, Fenger-Grøn, Larsen, Larsen, Robinson and Mortensen2016), happiness (Liu et al. Reference Liu, Floud, Pirie, Green, Peto and Beral2016), or stressful life events (Rutters et al. Reference Rutters, Pilz, Koopman, Rauh, Te Velde and Stehouwer2014)] have consistently reported that increased mortality is associated with poor mental well-being; most studies report significant associations even after various statistical adjustments. However, it is difficult to interpret causality based on these findings because of potentially insufficient control for disease burden (Russ et al. Reference Russ, Stamatakis, Hamer, Starr, Kivimaki and Batty2012; Rutters et al. Reference Rutters, Pilz, Koopman, Rauh, Te Velde and Stehouwer2014; Liu et al. Reference Liu, Floud, Pirie, Green, Peto and Beral2016). Short-term effects of negative impact may also have been overlooked by separating exposure measurements and outcomes too much in time in an effort to address potential reverse causation, which are inherent to many survey-based studies (Liu et al. Reference Liu, Floud, Pirie, Green, Peto and Beral2016; Prior et al. Reference Prior, Fenger-Grøn, Larsen, Larsen, Robinson and Mortensen2016). Generalizability may be limited by respondent bias and inclusion criteria. In contrast, the design of studies using external stressors (e.g. studies of bereavement or other major life events) is better suited to obtain precise temporal information of the psychological exposure. Therefore, these studies may give better support for examining causality although they rarely account sufficiently for morbidities (Moon et al. Reference Moon, Kondo, Glymour and Subramanian2011). Our study overcame the shortcomings from previous studies by combining exact temporal information on stress exposure and mortality with comprehensive register data on mental and physical conditions and socioeconomic factors.
Other bereavement studies have found increased mortality following bereavement. A meta-analysis showed an overall relative mortality risk of 1.41 during the first 6 months after bereavement; this risk decreased later to 1.14 (Moon et al. Reference Moon, Kondo, Glymour and Subramanian2011). No studies adjusted for mental-physical multimorbidity. Among older bereaved individuals, the confounding or mediating effect of socioeconomic factors and severe physical conditions did not eliminate the association between bereavement and mortality (Shah et al. Reference Shah, Carey, Harris, DeWilde, Victor and Cook2012). This is in line with our detailed findings among subgroups with mental-physical multimorbidity.
The association between bereavement and mental stress has been documented numerous times in the literature (Stroebe et al. Reference Stroebe, Schut and Stroebe2007). The classic study by Holmes and Rahe (Holmes & Rahe, Reference Holmes and Rahe1967) ranked spousal bereavement as the most severe stressor in terms of subsequent life changes when comparing severe life events. However, the stressors of bereavement are diverse and the individual impact depends on personal coping strategies (Byrne & Raphael, Reference Byrne and Raphael1997; Ryckebosch-Dayez et al. Reference Ryckebosch-Dayez, Zech, Mac Cord and Taverne2016). We assumed that the loss of a spouse would be a stressful experience with immediate onset (Shah et al. Reference Shah, Carey, Harris, Dewilde, Victor and Cook2013). Yet, in some cases, elevated stress levels may have been present for a prolonged period of time before the bereavement due to caregiver burden (Nielsen et al. Reference Nielsen, Neergaard, Jensen, Bro and Guldin2016). Although caregiving may increase the levels of stress, it may also increase the preparedness for the impending loss (Schulz et al. Reference Schulz, Boerner, Klinger and Rosen2015) or even develop into relief in the bereaved after the death. However, the analysis of expectedness of death, which was based on the spouse's morbidity 1 year prior to the loss, showed increased mortality in all bereaved individuals regardless of the bereaved individual's health status.
The loss of a spouse can result in complicated grief, depression, and suicide (Stroebe et al. Reference Stroebe, Schut and Stroebe2007; Ajdacic-Gross et al. Reference Ajdacic-Gross, Ring, Gadola, Lauber, Bopp and Gutzwiller2008), but we still lack knowledge on the importance of pre-bereavement mental well-being. Our findings show that bereavement adds to the high underlying risk of dying, which is present in those with mental conditions before bereavement. Even though the relative risk of suicide related to mental comorbidity may be high, the bulk of the widowhood effect in this study was driven by deaths due to medical illnesses. Therefore, it is essential to take into account the pre-bereavement physical health status when investigating the prognosis after bereavement. Extending beyond previous studies, we provided a more accurate estimate of the association between bereavement and mortality by adjusting for pre-bereavement mental-physical health status without adjusting for conditions with onset after bereavement, which could have underestimated the true association (Rothman et al. Reference Rothman, Greenland and Lash2008).
Implications
Our study suggests that bereavement as an indicator of mental stress may play a causal role in excess mortality. This calls for an increased focus on mental well-being and personal coping resources in clinical practice to help alleviate stress and optimize the treatment of chronic conditions, including prevention of treatment failure.
Public health policies should raise the awareness of the consequences of mental stress and allocate resources to health care professionals to ensure a more bio-psycho-social approach to illness. Importantly, our study identifies subgroups of multimorbid and physically fragile patients in whom mental stress is of particular significance. Prioritizing health care resources to favour these groups may prove efficient. While data on evidence-based interventions targeting mental stress is sparse, mindfulness-based stress reduction has shown some promise (Bohlmeijer et al. Reference Bohlmeijer, Prenger, Taal and Cuijpers2010). In younger individuals, the potential bereavement effect appeared to be particularly high, but the absolute number of deaths was low. Most deaths associated with bereavement were due to natural causes, such as cardiovascular events, which may be partly preventable if optimal health care for chronic disease is given.
Conclusion
Mental stress in the form of the loss of a spouse seems to have both short-term and long-term health effects, and these appear to persist for more than 10 years. Pre-bereavement mental-physical multimorbidity appears important in conveying risk of death in bereaved. More research on effective stress-reducing interventions is needed, particularly initiatives targeting individuals with multimorbidity.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291717002380
Acknowledgements
Data support was given by the Danish National Center for Integrated Register-Based Research, Aarhus University, Denmark. The study was supported by an unrestricted grant (grant number: R155-2012-11280) from the Lundbeck Foundation (MEPRICA) and the General Practice Fund of the Central Denmark Region. The funding sources had no role in study design; in the collection, analysis, and interpretation of data; in the writing of the report; and in the decision to submit the paper for publication.
Declaration of Interest
None.
