Widowhood and mortality: a Danish nationwide register-based cohort study

Aims Widowed people have increased mortality compared to married people of the same age. Although most widowed people are of older age, few studies include the oldest old. As life expectancy is increasing, knowledge of widowhood into older age is needed. This study aimed to examine mortality and widowhood in older age by comparing mortality in widowed and married people by sex, age, time since spousal loss and cause of death. Methods A Danish register-based matched cohort study of 10% of widowed persons ⩾65 years in the years 2000–2009. For each randomly drawn widowed person, five married persons were matched on sex and age. Mortality rate ratios (MRR) were calculated using Poisson regression, and stratified according to sex and 5-year age intervals. MRRs were furthermore calculated by time since spousal loss and by specific cause of death. Results The study included 82 130 persons contributing with 642 914.8 person-years. The overall MRR between widowed and married persons with up to 16 years of follow-up was 1.25 (95% CI 1.23–1.28). At age ⩾95 years for men, and ⩾90 years for women, no differences in mortality rates were seen between widowed and married persons. Mortality in widowed persons was increased for most specific causes of death, with the highest MRR from external causes (MRR 1.53 [1.35–1.74]) and endocrine diseases (MRR 1.51 [1.34–1.70]). Conclusions Widowhood was associated with increased mortality in older age for both men and women until age ⩾95 and ⩾90 years, respectively. Increased mortality was observed for almost all causes of death.


Introduction
Widowed people have increased mortality compared to married people (Moon et al., 2011;Shor et al., 2012), a phenomenon known as 'the widowhood effect' (Sullivan and Fenelon, 2014;Vable et al., 2015;Ennis and Majid, 2019). For people at age 65 years or above, the relative increase in mortality in widowed people compared to married people has been estimated to be 11% (Manzoli, 2007).
The risk of mortality seems to be highest immediately after spousal loss (Moon et al., 2011) but studies have found 'a widowhood effect' even after 25 years of follow-up (Shor et al., 2012). Results from different studies are however conflicting, and further knowledge of the longitudinal association is needed (Ennis and Majid, 2019).
Various conditions have been shown to be increased in widowed people compared to married people, including cardiovascular risk factors, chronic pain, cancers, lower functional status and mental disorders (Ennis and Majid, 2019). Also mortality in widowhood has been shown to be increased from various conditions including both acute events such as infections and acute myocardial infarction, and chronic diseases (Elwert and Christakis, 2008). This may be due to reduced social and emotional support, leading to reduced self-care of diseases (Elwert and Christakis, 2008). Furthermore, increased death from external causes in widowed people has been shown (Brenn and Ytterstad, 2016). This is partly due to suicides (Ajdacic-Gross et al., 2008) but an increased risk of accidents such as falls has also been found (Elwert and Christakis, 2008;Einio and Martikainen, 2019). Similar to the acute and chronic diseases described above, death from accidents could be related to less self-care and neglect of own needs, as well as be due to increased risk behaviours (Einio and Martikainen, 2019). Reduced self-care, increased risk behaviour and an increased risk of suicide have been associated with depression (Hawton and van Heeringen, 2009;Einio and Martikainen, 2019). As the prevalence of depression is high in widowed people (Kristiansen et al., 2019a(Kristiansen et al., , 2019b, this could be a possible explanation for some of the excess mortality in widowhood, and depression could thus be a relevant target for preventive interventions in the future. However, to be able to design and aim relevant interventions, more knowledge of causes of death, also in the oldest old, is needed, in order to understand the pathways into the increased mortality in widowhood. Few studies of the widowhood effect examine how this is associated with age in the oldest old (Shor et al., 2012). In the meta-analysis by Moon et al., age was dichotomised as younger than 65 years and 65 years or above (Moon et al., 2011). Shor et al. studied age also in the older population, more specifically by using 10-year age strata, however, with their oldest age stratum pooling those aged 80 years and above (Shor et al., 2012). With the increasing proportion of older people in most societies (United Nations, 2017), and with most widowed people being of older age (The Loomba Foundation, 2015), more knowledge about this group is needed. This is particularly important, as the increased mortality in widowed people has been shown to be increasing through the past decades (Moon et al., 2011;Shor et al., 2012).
The aim of this study was to examine widowhood and mortality in older age, including the oldest old. The study aimed to (1) examine widowhood and overall mortality compared to married persons by age, sex and time since spousal loss and (2) compare cause-specific mortality between widowed and married persons.

Methods
This is a Danish register-based matched cohort study. In Denmark, each citizen is assigned a unique lifelong personal identification number (the CPR number) which enables linkage between different registers at an individual level (Pedersen, 2006). Data in this study were drawn from the Danish Civil Registration System (Pedersen, 2006), the Income Statistics Register (Baadsgaard and Quitzau, 2011), the Danish Register of Causes of Death (Helweg-Larsen, 2011) and the Danish National Patient Register (Schmidt et al., 2015). For project protocol see Appendix 1.

Study population
The cohort consisted of a random sample of 10% of people in Denmark who became widowed at age 65 years or above in the period 2000-2009, both years included. Using incidence density sampling, five married persons were matched to each widowed person on age and sex.
Each person was followed in the registers from the date of inclusion until the date of death or censoring, which could be any day after inclusion. Thus, people could be followed for one day up to 16 years. Censoring of widowed persons was due to remarriage, emigration or end of study (31 December 2015). Censoring of married persons was due to becoming widowed, divorced, emigration or end of study. Married persons who became widowed were censored as married and were subsequently re-included as widowed on the date they lost their spouse and were followed as widowed until death or censoring. Each person contributed with person-years (PY) according to their current marital status at all times.

Exposure
The exposure studied was widowhood. This information was obtained from the Danish Civil Registration System (Pedersen, 2006). A person could be included either as widowed (exposed) or married (unexposed). Exposure status could change during the study from unexposed (married) to exposed (widowed) but not reverse. This strategy was chosen as widowed persons are potentially at an increased risk of death due to the experience of being widowed. Therefore, persons who have been widowed have been exposed to the potential long-term consequences of widowhood and could therefore not be re-included as unexposed.

Outcome
The outcome was death. Date and cause of death were extracted from the Danish Civil Registration System (Pedersen, 2006) and the Danish Register of Causes of Death (Helweg-Larsen, 2011). The cause of death used was the underlying cause of death, defined as the disease or condition, which started the process that led to death (Helweg-Larsen, 2011). The causes of death were grouped according to ICD-10 categories (World Health Organization, 1992)  The ICD-10 chapters of mental and behavioural disorders (F00-F99) and diseases of the nervous system and sense organs (G00-H59) were combined because the vast majority of deaths from these chapters were related to dementia.

Covariates
Time since spousal loss was examined as different periods of follow-up times from inclusion: 1 week, 1 month, 3 months, 6 months, 1 year, 2 years, 5 years, 10 years and 15 years, respectively. Widowed persons were followed from the date of spousal death (which equals the date of inclusion). Married persons were followed from the date of inclusion (which equals the day they were matched with the widowed person, and thus the date the widowed person lost his/her spouse).
Age was examined in 5-year intervals from 65-69 to ⩾100 years in the analyses stratified according to age. Socioeconomic status (SES) was measured as the total disposable personal income in the year prior to inclusion divided into deciles. Deciles were calculated across the study population in the calendar year of inclusion to enable comparison over the study period. The income from the year prior to inclusion was used to avoid false high estimates in the first year of widowhood due to death grants from pension funds, inheritances, etc. Comorbidity was defined according to the Charlson Comorbidity Index (Charlson, 1987;Charlson et al., 1994), applying comorbidity groups in terms of ICD-10 codes in accordance with previous studies using the Charlson Comorbidity Index (Brink et al., 2017). For each person, a Charlson Comorbidity Score (CCS) was calculated including diagnoses given up to 5 years prior to inclusion based on data from the Danish National Patient Register (Schmidt et al., 2015). For persons who changed exposure status from married 2 C. Blanner et al.
to widowed, income and CCS were updated at the time of re-inclusion as widowed.

Statistical analysis
Mortality rates (MRs) were calculated as the number of deaths/10 000 PY for widowed and married persons, stratified according to sex and 5-year age intervals. Time since spousal loss and specific causes of death were considered. Overall and cause-specific mortality rate ratios (MRR) with 95% confidence intervals (CIs) were calculated using Poisson regression of the number of deaths with logarithmic transformation of PY as offset and adjusted for the potential confounders: age in years, sex, CCS (in categories 0, 1, 2 or ⩾3), income decile and calendar year. In the analyses of cause-specific mortality, a person was censored in case of death by other causes than the cause of death examined, and persons with missing causes of death were excluded. Analyses were conducted using the statistical software package, Stata 15 (2017).

Characteristics of the study population
A total number of 82 130 persons were included in the study. Of the total study population, 66.4% were women (n = 54 534). The mean age of the study population was 75.8 years (S.D. 6.9) at the time of matching. Further characteristics at the time of matching including income deciles and comorbidity scores are seen in Table 1.

Person-years at risk
The participants contributed with a total of 642 914.8 PY. Of these, widowed persons contributed with 223 715.3 PY (of which women contributed with 78.8%) and married persons contributed with 419 199.5 PY (of which women contributed with 65.7%). Each person contributed only with PY according to their current marital status at all times. A total of 26 143 (37.8%) of those who were matched as married (see Table 1) changed exposure status during the study. Of these, 4652 were men (20.1% of all married men), and 21 491 were women (46.7% of all married women). The mean age was 85.1 (S.D. 6.2) and 83.1 (S.D. 6.0) years for men and women, respectively, at the time of re-inclusion into the study as widowed.

Overall mortality
The total number of deaths was 42 962. Of these, 55.0% (n = 23 620) were married and 45.0% (n = 19 342) were widowed at the time of death.  Table 2 shows the MRs and MRRs stratified according to age and sex. For both men and women, MRRs appeared to decrease with Cause-specific mortality Table 4 and Fig. 2 show the sex-specific MRs and MRRs for specific causes of death. For women, widowhood was associated with increased cause-specific mortality for each of the considered causes of death, except for death from diseases of the blood or immunological system, and from death from other causes. For men, there was no difference in cause-specific mortality for death from infections, diseases of the blood or immunological system, from neurological diseases or mental disorders or from the category other diseases. For all remaining causes of death, mortality was significantly increased in the widowed men compared to the married men.

Discussion
We examined the association between widowhood and mortality in a Danish study population aged 65 years and above for up to 16 years of follow-up. The increase in the overall MR in widowed persons was 25% (23% in women and 29% in men). This is similar to what has been found in a meta-analysis by Shor et al. (2012), but higher than the 11% increased risk found in the meta-analyses by Moon et al. (2011) and Manzoli (2007). We found a significant widowhood effect on mortality until age ⩾90 years for women and ⩾95 years for men. The widowhood effect appeared to decrease with increasing age. Shor et al. (2012) found results similar to ours with decreasing MRR with increasing age. They divided age into 10-year age intervals up to ⩾80 years of age and found significant differences in mortality between widowed and married in all age groups (Shor et al., 2012). They predicted that by ⩾90 years, the difference would be insignificant (Shor et al., 2012). Currently, almost 20% of the Danish population is aged 65 years or above, and about 4.6% is aged 80 years or above (Statistics Denmark, 2020). Statistics Denmark has predicted that in 2050, these numbers will be about 25 and 9.5%, respectively (Statistics Denmark, 2020). Thus, it is especially the proportion of the oldest old which is increasing. This overlap between most widowed people being of older age (The Loomba Foundation, 2015) and the fact that the proportion of those of older age is increasing in most societies (United Nations, 2017) underlines the importance of this finding.
We found that for women ⩾100 years of age, widowhood was associated with a lower risk of mortality. Although the number of PY is relatively low in this age stratum (337 PY in total), the phenomenon is interesting and has not previously been shown.
Marriage is known to be protective of health in general (Kiecolt-Glaser and Newton, 2011); however, the protective effect of marriage on health diminishes with decreasing health (Zheng and Thomas, 2013). Further, a study has found that widowed women have less risk of becoming frail than married women in a follow-up period of 3 years (Wilcox et al., 2003). As health often decreases with increasing age, it is possible that in very old age, the protective effect of marriage on women compared to widowed women is no longer present. Indeed, marriage might even be a health disadvantage, because married women might become caregivers for an ill spouse, which is known to be associated with poor health outcomes (Penning and Wu, 2016;de Zwart et al., 2017). Another reason could be that widowed women who live to be 100 years, despite becoming widowed, belong to a particularly resilient group. Shor et al. (2012) concluded in their meta-analysis that the association between widowhood and mortality was strongest during the first 2 years of widowhood, after which it decreased but remained significant for up to 20 years. We did not find a similar temporal association in the present study, where we were able to examine different lengths of widowhood in the same study population. For women, the MRR was highest during the first month of widowhood. When including up to 3 months, the MRR did no longer appear to be different from the one including up to 15 years of widowhood. This indicates that the overall increased mortality which is seen in widowed women compared to married women may be explained by an increased mortality within the first months of widowhood. In men, the MRR was insignificant during the first month. The non-significance did not seem to be explained by a lack of statistical power. A possible explanation is the high age of the people included in the present study compared to previous studies. Adjacic-Gross et al. (2008) examined the risk of suicide in widowed people compared to married people and found that the risk was highest in the first week of widowhood, after which it decreased but remained high in the first year. The increase in mortality was higher in the younger participants than in those of older ageespecially in the first week of widowhood (Adjacic-Gross et al., 2008). A similar association was found by Brenn and Ytterstad (2016), who also found a strong association of death from external causes immediately after spousal death, which was highest in the youngest age group examined and seemed to decrease with increasing age. The total number of deaths from external causes is relatively low compared to deaths from other causes both in the present study and in the study by Brenn and Ytterstad (2016). Nonetheless, the relative increase in the risk of mortality in widowed people compared to married people from this cause is high. In the study by Adjacic-Gross et al. (2008), the SMR was about 15 for widowed men ⩾60 years of age the first week and almost 90 for those under 60 years of age (Adjacic-Gross et al., 2008). Similarly, the HR for widowed men v. married men aged 55-64 years was above 15 for death from external causes in the first week after widowhood in the study by Brenn and Ytterstad (2016). Thus, although the total number of deaths from this cause is low, its relative contribution to the increased risk of mortality in early widowhood is high. In comparison, we found a lower MRR from death from external causes for widowed men compared to married men which was about 1.8. Thus, it might be that the increased risk of mortality immediately after spousal loss previously seen in widowed men is explained primarily by the very high risk of death from external causes immediately after spousal loss in younger samples. Nonetheless, despite this difference, external causes are still an important cause of death in widowed people also at older age. Since suicide most often occurs in people who are depressed (Hawton and van Heeringen, 2009), and the prevalence of depression in widowed people is known to be high Fig. 1. Mortality rate ratios for widowed persons compared to married persons by sex and time since spousal loss. MRR, mortality rate ratio adjusted for age, calendar year, Charlson Comorbidity Score and income. 95% CI, 95% confidence interval.

6
C. Blanner et al. PY, person-years at risk; n, number of deaths; MR, mortality rate pr. 10 000 PY. MRR, mortality rate ratio for widowed compared to married; CI, confidence interval. a Adjusted for age, income and Charlson Comorbidity Score. MRRs for both sexes are also adjusted for sex. b A total of 1029 causes of death were missing. These observations are excluded from the cause of death analysis and the PY presented in this table.
Epidemiology and Psychiatric Sciences (Kristiansen et al., 2019a(Kristiansen et al., , 2019b, a possible target for preventive interventions in terms of mortality in widowhood could be depression and consequently suicide in widowed people. In line with previous research, we found that mortality was increased in widowed persons for most specific causes of death (Elwert and Christakis, 2008;Brenn and Ytterstad, 2016). Elwert and Christakis (2008) examined a large number of specific causes of death and found that widowhood increased mortality by both acute events such as sepsis, pneumonia and influenza as well as chronic diseases such as congestive heart failure, diabetes and chronic obstructive pulmonary disease. The increased MR in widowed persons across most causes of death indicates that widowhood may be considered a general risk to the health. It has been shown that some physical health disparities may occur prior to widowhood (Wilcox et al., 2003;Einio et al., 2017). In the present study, we controlled for comorbidity 5 years before inclusion using register-based data with very high coverage of all hospital contacts (Schmidt et al., 2015). The distribution of CCS at time of matching appeared to be similar in widowed and married persons (see Table 1). Thus, if the increased mortality in widowed persons was explained by health disparities prior to widowhood, these are not evident from the hospital records (in this study represented by data from the National Patient Registers used in the CCS), indicating that the diseases were either diagnosed and treated in primary care (as information from primary care is not included in the National Patient Register) or not diagnosed and treated at all in widowed persons. The latter might be important as qualitative research has suggested that widowed people who have been caregivers of an ill spouse may deliberately have postponed taking care of their own health problems (DiGiacomo et al., 2013). It could also be that the differences in health disparities prior to widowhood which have previously been shown (Wilcox et al., 2003;Einio et al., 2017) are less pronounced in a Danish sample, as all people in Denmark have the same universal and free access to the healthcare system, reducing differences in health due to for an example SES. The study by Einio et al. is however conducted in Finland, and as the two countries are comparable regarding both healthcare (Eide et al., 2017), as well as the association of income and mortality in older age (Mortensen et al., 2016), this does not seem to be the full explanation. In order to further explore the influence of the time before spousal death, future studies should include information on the health, as well as the causes of death from both spouses. This could enlighten possible effects from caregiving, as well as effects of specific causes of deaths from the predeceased spouse such as expected v. unexpected deaths, all factors which have been proposed to explain some of the widowhood effect (Elwert and Christakis, 2008;Sullivan and Fenelon, 2014;Siflinger, 2017).
Cause-specific mortality from the ICD-10 chapter 'symptoms, signs and ill-defined causes' was increased in widowed persons for both sexes. There are no previous studies examining death from unknown and ill-defined causes in widowed persons. This might be because the World Health Organization discourage the use of the ICD-10 codes R00-R99 (symptoms, signs and illdefined causes) as the cause of death unless all measures have been taken to establish the cause of death (Ylijoki-Sorensen et al., 2014). The occurrence of the R00-R99 codes in death certificates in Denmark might be due to a low rate of autopsies (Helweg-Larsen, 2011). A study found that people over 70 years were the most likely age group to be registered with an unknown or ill-defined cause of death and least likely to be autopsied (Ylijoki-Sorensen et al., 2014). The authors speculated that doctors might not have been as meticulous about finding the correct cause of death in older people (Ylijoki-Sorensen et al., 2014).

Limitations
The study has some limitations. First, since the persons who were reincluded as widowed must be of older age at the time of Cause-specific mortality rate ratios for widowed persons compared to married persons stratified according to sex. MRR, mortality rate ratio adjusted for age, calendar year, Charlson Comorbidity Score and income. 95% CI, 95% confidence interval.
widowhood than they were at the time of matching, the mean age in widowed persons in the analyses is higher than in married persons. To account for this bias, all analyses were either stratified according to or adjusted for age. However, since age becomes an increasingly dominating risk factor of mortality as age increases, some age-related residual confounding in the age-adjusted analyses cannot be ruled out. Another limitation is the use of personal yearly income as the only measure of SES. Other measures of SES, including wealth and education, have been found to account for up to a third of the increased mortality in widowed people in an American study (Sullivan and Fenelon, 2014). A study from the Netherlands did not find that income was associated with mortality in widowed people (van den Berg et al., 2011). Considering the extensive welfare system in Denmark, the influence of income in widowhood might be similar to that of the Netherlands. Thus, using income alone as a proxy for SES might not be sufficient and there might be residual confounding due to other measures of SES. However, due to the age range of the study population, it was not possible to meaningfully adjust for educational level or latest employment status. The educational attainment register was not fully operating until the 1970s; therefore, many persons in the study population had finished their education before the register had sufficient coverage (Statistics Denmark, 2019). Similarly, due to the age of the study population, the recent employment status was not meaningful, as the majority of the study population was retired.
The study aimed to examine mortality in widowed people at older age specifically compared to married people, as losing a spouse to death is a frequent event in older age. The results of the study are therefore restricted to enlightening the relationship between mortality in widowed and married people only. Other groups of people could have been considered for comparison including singletons, people cohabiting with another person than a spouse, as well as a comparison of those who stay widowed and those who remarry. For future studies, including other exposure groups besides married and widowed people could be relevant in order to further understand the causality of the increased mortality in widowed people compared to married people.

Conclusion
We found an increased mortality in widowed persons compared to married persons and have shown that the increased mortality persists until very old age, however, is no longer significant at ages over 90 and 95 years for women and men, respectively. Mortality is increased immediately after widowhood for women, whereas for men the effect of widowhood on mortality seems less dependent on the time since spousal loss. Widowhood was associated with increased mortality from most specific causes of death, with the largest effect on death from endocrine diseases and death from external causes, including suicides and accidents.
Data. Data have been made available for the first author specifically via Statistics Denmark and can therefore not be shared.

Background
Increased mortality associated with widowhood, also known as the 'widowhood effect', is a well-known phenomenon, documented most recently in two meta-analyses (Moon et al., 2011, Shor et al., 2012. Many studies upon the widowhood effect have been conducted (Moon et al., 2011, Shor et al., 2012; however, the comparability of these studies is often limited due to differences in the methods, samples and adjustment for possible mediators and confounders. Many factors might influence the outcome of widowhood both as risk factors and protective factors. In order to understand this phenomenon properly, these factors need to be investigated.

Aims of the study
The aim of this study is to examine the risk of mortality associated with widowhood both regarding all-cause and specific causes of death. Furthermore, to examine the effect of age and gender on the mortality risk. Lastly, to examine if there is an association between time since spousal death and mortality risk.

Design
A register-based population-based cohort study. The exposure studied is becoming widowed.

Population
All people in Denmark who are married and became widowed at the age of 65 years and above in the period 2000-2009.

Sample
A cohort consisting of a 10% random sample of those who were widowed at age 65 years and above in 2000-2009 is created. For each widowed person, five persons who match the widowed on age (±2 years) and gender, and who are married at the time the widowed person lost his/her spouse are included by random selection.

Study period
The exposed are included from the date they become widowed and are followed up until death (outcome), or end of study on December 31 2015. If a participant remarries, he/she is censored.
The unexposed are included from the day the exposed became widowed, and are followed up either until death (outcome) or end of study on 31 December 2015. If a non-exposed becomes widowed, he/she is censored from the study, as he/she is no longer unexposed.