Background: We consider whether widowhood increases mortality risk. Although commonly observed, this “widowhood effect” could be due to selection effects, as married couples share various characteristics related to the risk of death. We therefore consider the widowhood effect by various causes of spousal death; some causes of death are correlated with shared characteristics in couples, while others are not.
Methods: Using data from the Scottish Longitudinal Study, we compare outcomes for men and women by the causes of death of their spouse, controlling for a range of individual- and household-level characteristics.
Results: The widowhood effect in these data is greater than has been found in other recent studies, with adjusted hazard ratios of 1.40 (95% confidence interval = 1.33–1.47) for men and 1.36 (1.30–1.44) for women. The risk is highest shortly after widowhood, but remains raised for at least 10 years. There was little evidence that these hazard ratios differed by any classification of the cause of death of the spouse, but interactions were found for those with pre-existing illness or other risk factors. The hazard ratios for widowhood were lower for persons with preexisting risks.
Conclusions: Our analysis of the widowhood effect uses 3 methods of classifying the causes of spousal death in an attempt to control for potential selection effects. Our results are highly consistent and suggest that this is a causal effect, rather than a result of selection.
From the Longitudinal Studies Centre—Scotland, University of St. Andrews, St. Andrews, Scotland, United Kingdom.
Submitted 21 January 2010; accepted 13 August 2010; posted 3 November 2010.
The authors were funded by the Economic and Social Research Council while this work was conducted.
Supplemental digital content is available through direct URL citations in the HTML and PDF versions of this article (www.epidem.com).
Correspondence: Paul J. Boyle, School of Geography & Geosciences, University of St. Andrews, St. Andrews KY16 9AL, Scotland, United Kingdom. E-mail: P.Boyle@st-andrews.ac.uk.
The “widowhood” or “bereavement” effect has been demonstrated in numerous studies.1–7 The consensus is that the death of a spouse raises the risk of mortality for the surviving spouse by 10%–40%.8–10 This result appears reasonably consistent across countries, datasets, and methodological approaches.11–15
That widowhood shortens life is remarkably persuasive evidence that social circumstances influence mortality. Such an effect is consistent with studies demonstrating the benefits that marriage provides. Married people live longer, healthier lives than single people.4,12,14,16 Being widowed removes the protective effects of marriage,8 and if this effect is causal it is convincing evidence that social determinants have a powerful influence on mortality.
However, there are other possible noncausal explanations of elevated mortality postwidowhood, as married couples may share characteristics that predict mortality, including the following:
1. Shared socioeconomic background
2. Shared health-related life styles
3. Common access to and utilization of heath care resources
4. Shared attitudes to risk.
If the selection effect (or bias from unmeasured confounding) is due to (1), we would expect that controlling for socioeconomic variables would attenuate the widowhood effect and that it would be less pronounced for spousal causes of death unrelated to socioeconomic factors. If it is due to (2) or (3), we would expect that the widowhood effect would differ for spousal causes of death that could be avoided by healthier lifestyles or health care interventions and if (4) we would expect the widowhood effect to be more pronounced when the spouse died of a risk-related cause.
To identify causes of death related to socioeconomic factors, we used the classification by Espinosa and Evans14 that contrasts informative (ie, related to socioeconomic factors) and noninformative deaths. For avoidable mortality we used Page et al's17 3-fold classification into “amenable” (avoidable by treatment of the disease after onset), “preventable” (by changes in individual behavior), and “unavoidable”—a classification that was recently validated for UK data by Wheller et al.18 For risk-related causes, we used the classification by Martikainen and Volkonen12 that identifies “risky” causes mainly due to accidents, violence, smoking, or alcohol use.
Other studies have investigated covariates and causes of spousal death5,8,12,14,16 as explanations of the widowhood effect on mortality, but none as comprehensively as here. Our large sample also allows us to investigate whether it is modified by socioeconomic factors and by time after widowhood.
The data are from the Scottish Longitudinal Study,19 which collates information from the 1991 and 2001 national censuses, as well as from vital events, for a 5.3% sample of the Scottish population. It includes approximately 264,000 people in 1991; we extracted those aged 16 years and above who were living in married couples. Linked death records through December 2006, with causes of death, were available for study subjects (and their spouses, if they predeceased them). Deaths within 30 days after widowhood were excluded if they were due to common causes (eg, accidents). The data set included 58,685 men and 58,415 women who were living with a spouse, and with both members of the couple enumerated in the 1991 Census. During follow-up, 5002 men (8.5%) and 9628 women (16.5%) were widowed; 40% of widowers (n = 2015) and 26% of widows (n = 2548) died during this period. Individual characteristics, including age, qualifications, social class, ethnicity, and self-reported health status were available for study subjects and their spouses from the 1991 Census. The Census also provided data on the following household characteristics: household size, tenure, car availability, presence of central heating, and an area-based deprivation measure (the Carstairs score), which was calculated for the 1003 postcode sectors in Scotland.
We modeled time to death using a Cox proportional hazards model to obtain hazard ratios (HRs) and associated 95% confidence intervals (CIs). Widowhood was included as a time-varying covariate, whereas the remaining individual and household variables were time-invariant. We investigated the following interactions: whether the widowhood effect was modified by the cause of death of the spouse, by time from widowhood, and by other socioeconomic characteristics of the widowed person as measured at baseline.
Summary statistics and details of the causes of death are in eTables 1 and 2 (http://links.lww.com/EDE/A436). Table 1 gives the numbers of deaths by informative, avoidable, and risky causes. The smaller group of deaths due to risky causes was almost all in the preventable or amenable categories of the avoidable classification (over 99% for both men and women).
The hazard ratios for death after widowhood compared with not being widowed, adjusted only for age, are substantial: 1.49 (95% CI = 1.41–1.57) for men and 1.45 (1.38–1.53) for women. Adjusting for other socioeconomic variables reduces, but does not eliminate, the effect: 1.40 (1.33–1.47) for men and 1.36 (1.30–1.44) for women.
Adjusting only for age, there is evidence that the widowhood effect for men is lower for noninformative, unavoidable, and not-risky causes of death of the spouse, but there is no such evidence for women (Table 2). After adjustment for covariates, these interactions for men are much reduced, with evidence of an interaction present only for risky versus non-risky causes of death; there is still no evidence of interactions by cause of death for women. There is evidence for the effect of widowhood within every subgroup.
The adjusted models were extended to investigate how the widowhood effect varies with time since widowhood (Fig.). The number of deaths contributing to the hazard ratio in each period after widowhood is detailed in eTable 3 (http://links.lww.com/EDE/A436). The effect of widowhood is greatest for both men and women in the first 6 months after the death of the spouse, but it remains elevated for at least the following 10 years.
The baseline (1991) characteristic that modifies the effect of widowhood to the greatest extent is the person's health status (Table 3). As expected, long-term illness itself increases mortality risk. For widows who are ill the hazard ratio, although still elevated, is lower than for widows not ill at baseline. Table 3 also gives results for the widowhood-by-age interaction. Women widowed at a young age are at increased risk, but there are too few of them to distinguish this from a chance effect. Women widowed at the age of 80 or older are at increased risk compared with those widowed at ages 60–74. A similar but less pronounced effect is seen for men. There was also evidence that the hazard ratio for widowhood was lower for those already at increased risk of mortality (lack of qualifications, living in social rented housing, no car availability, no central heating, low social class, deprived area). Two examples are shown in Table 3.
Previous studies have found an increased risk of death of (hazard ratios of 1.1–1.4) following the loss of a spouse, although the results have been less consistent for women than men. Results from our data are at the high end of this range for both men and women, even controlling for a range of individual- and household-level variables. Some previous studies find high risks in the periods immediately after the death of a spouse.16,20 Our results suggest that the size of this effect, if present, is modest and that the widowhood effect persists for many years, confirming other previous studies.15,21
The widowhood effect is not removed once the cause of the spouse's death is controlled. Adjusting for other factors, only male widows whose wives died of a risky cause have an increased risk of death compared with men whose wives died of unrisky causes (although both groups have increased mortality). Perhaps this reflects shared risk-taking that is not captured by covariate adjustment. But, if so, a similar result would be expected for women whose husbands died of a risky cause-and this was not found. The majority of risky causes of death are related to smoking, so other explanations in terms of changed behavior after widowhood are possible. This effect was seen for both men and women in a Finnish study.12
Our finding that widows in poor health at baseline had a lower hazard ratio of death than healthy widows has been reported elsewhere.15 The effect of widowhood may be lower, on a hazard-ratio scale, for those with increased hazards due to other causes.
There are limitations with this study. First, some married couples may have been separated during follow-up, but they would still be recorded as widows when their previous partners died. Given that the relationship between these pairs will be weaker than for married couples, we would expect that this will make our results conservative. Second, our control for socioeconomic circumstances was measured only at baseline, and the breakdown by causes of death may be incomplete due to misclassification. Since confounding by these factors was small, it is unlikely that more accurate measures would change the results.
The combination of our evidence and that from previous articles that dealt with selection effects seems to provide evidence of a social impact on mortality. Although this effect was identified as long ago as the middle of the nineteenth century,2 it continues to be substantial.
1. Cox PR, Ford JR. The mortality of widows shortly after widowhood. Lancet
2. Farr W. The influence of marriage on the mortality of the French people. In: Hastings GW, ed. Transactions of the National Association for the Promotion of Social Sciences
. London: Parker and Son; 1858:504–513.
3. Gove WR. Sex, marital status and mortality. AJS
4. Lillard LA, Panis CWA. Marital status and mortality: the role of health. Demography
5. Manor O, Eisenbach Z. Mortality after spousal loss: are there socio-demographic differences? Soc Sci Med
6. Parkes CM, Benjamin B, Fitzgerald RG. Broken heart: a statistical study of increased mortality among widowers. Br Med J
7. Waite LJ. Does marriage matter? Demography
8. Elwert F, Christakis NA. Wives and ex-wives: a new test for homogamy bias in the widowhood effect. Demography
9. Elwert F, Christakis NA. Widowhood and race. Am Sociol Rev
10. Manzoli L, Villari P, Pirone GM, Boccia A. Marital status and mortality in the elderly: a systematic review and meta-analysis. Soc Sci Med
11. Hu Y, Goldman N. Mortality differentials by marital status: an international comparison. Demography
12. Martikainen P, Valkonen T. Mortality after the death of a spouse in relation to duration of bereavement in Finland. J Epidemiol Community Health
13. Gardner J, Oswald A. How is mortality affected by money, marriage and stress? J Health Econ
14. Espinosa J, Evans W. Heighted mortality after the death of a spouse: marriage protection or marriage selection? J Health Econ
15. Schaefer C, Quesenberry CP Jr, Wi S. Mortality following conjugal bereavement and the effects of a shared environment. Am J Epidemiol
16. Waite LJ, Gallagher M. The Case for Marriage: Why Married People are Happier, Healthier, and Better Off Financially
. New York: Broadway Books; 2000.
17. Page A, Tobias M, Glover J, Wright C, Hetzel D, Fisher E. Australian and New Zealand Atlas of Avoidable Mortality.
Adelaide, Australia: PHIDU, University of Adelaide; 2006.
19. Boyle P, Feijten P, Feng Z, et al. Cohort profile: the Scottish Longitudinal Study (SLS). Int J Epidemiol
20. Lichtenstein P, Gatz M, Berg S. A twin study of mortality after spousal bereavement. Psychol Med
21. Helsing K, Szklo M. Mortality after bereavement. Am J Epidemiol