![]() |
Abstract |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Methods Subjects consisted of women aged 35 in a longitudinal study of a nationally representative sample. Marital status and covariates were enumerated at a baseline interview in 1984/85 and a follow-up interview in 1991/92. Death data up to May 1997 were obtained from the National Health Service Central Register. Cox regression was used to estimate hazard ratios (HR) for the single, divorced and widowed states in relation to the married state.
Results Having adjusted for age and martial selection factors, being single (HR = 1.45) was significantly associated with higher all-cause mortality. Being divorced and being widowed showed no excess mortality risk (each HR = 1.09).
Conclusions Being single was associated with higher mortality. A causal interpretation is plausible. Being divorced and being widowed were not associated with higher mortality.
Keywords Marital status, mortality, selection, British women
Accepted 5 August 1999
![]() |
Introduction |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
The 1990s has seen several longitudinal studies on this issue. Using census-linked and population registry data, researchers in Sweden and Finland have shown an association between marital status and mortality after adjusting for socioeconomic factors that may have a selection effect.1012 An American longitudinal study of the elderly also showed that marital status appeared to have a modest effect on health and mortality.13 However, these datasets did not allow adjustment for marital selection related to risk-taking behaviour and/or health status. In Britain there is considerable research interest about excess mortality during bereavement. Longitudinal studies of those widowed took place as early as in the 1950s.14 Some researchers have reviewed the bereavement literature.1517 They pointed out considerable methodological problems in the literature, such as not having a proper control group and inadequate control for confounders. Two longitudinal studies of British men had the advantage of including subjects of all marital status and adjustment for behavioural, health and socioeconomic factors measured at baseline.18,19 They showed that unmarried men suffered higher mortality than their married counterparts. This association persisted after adjustment for various potential selection factors. Relatively little is known about British women, single and divorced women in particular.
In the last few decades, there was a worldwide trend of increasing mortality differentials by marital status.20 In Britain the relative mortality risk between unmarried and married women has also been on the increase.21 Marriage rate is decreasing and divorce rate is on the rise.22 Women tend to marry men older than themselves, and have a life expectancy at age 65 of about 4 years longer than that of men.23 This generates a demographic pressure of producing more widows than widowers. Given these demographic conditions, the impact of marital status on mortality in British women is a matter of public health significance. To address this public health issue and fill the gap in the literature, the present study focuses on the mortality differences of women of various marital status.
![]() |
Methods |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
The present study included female respondents aged 35 in HALS1. Death data from HALS1 to May 1997 were available for the analysis, giving a follow-up period of about 12.3 years. Among 3497 eligible respondents, 73 were not flagged on the NHS Central Register. Of the 73, 15 participated in HALS2. These 15 subjects were included in the analysis and were censored at the time of HALS2; the other 58 were excluded in the main analysis. Also excluded were 61 subjects who did not give all the relevant information for the covariate adjustment. The main analysis involved 3378 respondents.
Legal marital status and cohabitation were enumerated in both HALS1 and HALS2. In women born prior to the 1940s cohabitation was rare.26 A negligible proportion (1%) of the subjects reported themselves as cohabiting in HALS1 or HALS2. This study only considered legal marital status. Marital status was created as a time-varying independent variable. Responses in HALS2 were used to update it. Age at beginning of a marital spell was also made time-dependent accordingly. For brevity, in this paper single refers to never married, and divorced includes separated women.
Five variables measured in HALS1 were taken as indicators of the health, socioeconomic and behavioural aspects of marital selection. Recent longitudinal studies have shown that self-reported health is powerful in predicting both physiological health and mortality.25,27 This variable was coded as excellent, good, fair, or poor. Adult height is related to health and nutrition during paediatric years,28 and to perceived physical attractiveness.29 Some studies have demonstrated its usefulness in predicting mortality.30,31 These properties make adult height a useful indicator of marital selection. The HALS interviews included an item on self-reported height. In addition, a home visit was made by a nurse who carried out physiological measurements, including height. The height variable was based on the measured height if available (80% of the subjects), and self-reported height otherwise (20%). Among the 80% of subjects who had both data available, intra-class correlation between the measured and self-reported values were 0.82. Highest education attainment was categorized as below O Level, O Level or equivalent, and A Level or above. In the survey this variable was also strongly associated with housing tenure and social class. Smoking represents one aspect of an unhealthy lifestyle, and is closely related to other risk-taking behaviour.32 The subjects were classified as non-smoker, ex-smoker, 20 cigarettes/day, and >20 cigarettes/day. Drinking was based on several questions about drinking pattern and a diary recording alcohol consumption in the last 7 days. The subjects were classified as non-drinker, ex-drinker, light drinker or moderate/heavy drinker. The boundary of light and moderate/heavy drinkers was 14 alcohol units per week, which was the Health Education Authority's recommendation for sensible drinking.33 Since only seven subjects drank at the harmful level (>35 units), all subjects consuming over 14 units were grouped into the moderate/heavy category. The differentiation between ex-smokers/ex-drinkers and non-smokers/non-drinker was needed because some people may stop smoking and drinking as a result of health problems.34 They are not comparable to people who never smoke or drink.
The primary concern of the present study was all-cause mortality. Causes of death were coded according to the Ninth Revision of the International Classification of Diseases (ICD). For subsidiary analysis, the causes were classified into three broad groups: cardiovascular diseases (CVD) (ICD codes 390 458), neoplasms (ICD codes 140239) and other diseases (non-CVD/non-cancer).
Descriptive statistics about non-response, marital transition, number of deaths, etc., were presented to facilitate interpretation of subsequent regression analysis. The main analysis was based on Cox regression models. One set of analysis adjusted for a second order polynomial of age; another set adjusted for the selection factors in addition. A statistical test proposed by Grambsch and Therneau, which was based on scaled Schoenfeld residuals, was used to assess the proportional hazards assumption.35 If a global test showed non-proportionality, the covariate-wise residuals were examined. For the analysis of all-cause mortality, a probability value of 5% was taken as statistically significant. It was appreciated that the sample size was not ideal for the subsidiary, cause-specific analysis. The point estimates related to cause-specific mortality were interpreted in a tentative manner regardless of statistical significance. Sensitivity analyses were carried out to check how (not) using time-varying covariates according to HALS2 data could affect the main research findings.
![]() |
Results |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Table 1 shows the 3378 subjects' follow-up status and marital status in HALS2 by marital status in HALS1. Among the survivors, the probabilities of non-response in HALS2 were 29%, 31%, 39% and 33% for the married, single, divorce, and widowed, respectively. Only 263 subjects reported a change in marital status. Marital transition during the 7-year period was uncommon among women single or widowed in HALS1. Among women married in HALS1, transition was mainly to widowhood. Among the divorced, transition was mainly to marriage. In the survival analysis below, marital status was a time-varying variable updated if a subject participated in HALS2 and reported a change in marital status. Otherwise, HALS1 marital status was used throughout.
|
|
|
|
|
|
![]() |
Discussion |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Marital status was updated by using information from HALS2. This reduced the classification errors that some previous studies were unable to avoid.13,18 Sensitivity analysis showed that, in this particular data set, updating marital status or not did not make any material impact. About one-third of the surviving subjects did not participate in HALS2 and their marital status could not be updated. Official statistics showed that people who had not married by their early thirties were unlikely to marry at all.36 Remarriage at advanced age was also rare.36 Therefore the single and widowed subjects had entered a stable marital career by HALS1. From the mid-1980s to mid-1990s the median age at divorce was around 34,37 which was much lower than the average age of the married subjects (i.e. 52). As such, one would expect that the subjects married in HALS1 had relatively higher probability of transition to widowhood than to divorce. The findings shown in Table 1 agreed with these general demographic patterns, so the non-response in HALS2 would cause little problem in the classification of the single and widowed states. Had they been interviewed in HALS2, some married people would have reported as widowed and, to a lesser extent, divorced; some of those divorced would have reported as (re)married. This misclassification of exposure might result in some underestimation of the HR.
Since changes in health and risk-taking behaviour are expected to be dependent upon marital status,18,38,39 it is unreasonable to use HALS2 data to update the health and behavioural variables among those not reporting a change in marital status. It is arguable whether they should be updated among the 263 subjects who reported a change in marital status, because of the uncertainty of the (causal) relation between changes in marital status, health status and behavioural pattern. Sensitivity analysis with the four variables updated gave virtually the same results. This is not surprising because only 30 deaths were affected by the updating. Educational attainment is usually achieved by young adulthood and is stable among the middle-aged and the elderly so the issue did not concern this variable.
The impact of marital status on mortality, if any, may consist of two components. One is the impact of being in a marital state; another is the short-term impact of transition into a marital state. Previous studies has found evidence of a short-term impact of transition from marriage to divorce and widowhood.10,19,40 The present study was not designed to capture the effect of a marital transition. The results were interpreted as the effect of being in a marital status.
The single state was associated with a more favourable health, educational and behavioural background; the divorced state was associated with an unfavourable background. The primary concern of the present study is all-cause mortality. Adjusting for the effect of age but not other variables, there was little evidence to suggest any excess all-cause mortality among the unmarried groups. Adjustment for the selection factors demonstrated a statistically significant excess mortality for the single (HR = 1.45). The adjustment made the estimates for the divorced and widowed closer to the null value (each HR = 1.09). As previously mentioned, the HR might have been somewhat underestimated but not overestimated. Having considered the issues of sample representativeness, marital selection effect, and classification errors, it seems that being single did have a causal impact on mortality in British women aged over 35. There was no evidence to suggest that being divorced and widowed would cause an excess mortality. This is similar to the finding of a nationally representative study of British men, which showed that with partial and full statistical adjustment the HR for the single was 1.6 and 1.4, respectively (each P < 0.05).19 In either case of statistical adjustment divorced and widowed men did not appear to suffer higher all-cause mortality (HR between 1.11.2; P > 0.05).
Keeping in mind the reduced sample size and the play of chance resulting from multiple testing, the marital patterns in cause-specific mortality are interpreted in a more tentative manner. None of the unmarried states was associated with higher CVD mortality. It supports previous research that being a widow did not obviously increase the chance of having a broken heart.40,41 None of the unmarried groups had a general increase in all three types of mortality. Being single or widowed were more associated with non-CVD/non-cancer deaths than with CVD or neoplasm deaths. Being divorced was more associated with death from neoplasms than with other types of death. Some researchers have questioned the validity of the hypothesis of general susceptibility.14,42 The findings here lead to a speculation that different unmarried status has different psychosomatic and behavioural implications, which may have specific rather than general effects on health.
Grambsch and Therneau's test showed that the proportional hazards assumption was valid for the all-cause and CVD mortality models. The assumption for the other two cause-specific mortality models was rejected. Non-proportionality in the neoplasm model was related to widowhood; non-proportionality in the non-CVD/non-cancer model was not related to marital status at all. A plot of residuals against survival time suggested that the impact of widowhood on neoplasm mortality was hazardous in the short run but appeared to be protective in the long run. That there may be a short-term increase in mortality during bereavement is not surprising. It has been demonstrated and discussed in the literature.10,14,40,43 The long-term protective effect is likely to be spurious. As time went by more married women became widows. If becoming widowed was hazardous but being widowed was not, the increase over time in misclassification of marital status would produce a seemingly protective effect for widowhood. A recent American study has also discussed this problem.13 This problem means that the HR for the widowed might be underestimated, and highlights the importance of developing better systems to capture marital changes.
![]() |
Conclusion |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
![]() |
Acknowledgments |
---|
![]() |
Notes |
---|
![]() |
References |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
2 Joung IMA, van de Mheen HD, Sronks K, van Poppel FWA, Mackenback JP. A longitudinal study of health selection in marital transition. Soc Sci Med 1998;46:42535.[ISI][Medline]
3 Lillard LA, Panis CWA. Marital status and mortality: the role of health. Demography 1996;33:31327.[ISI][Medline]
4 Mastekaasa A. Marriage and psychological well-being: some evidence on selection into marriage. J Marriage Family 1992;54:90111.[ISI]
5 Cheung YB. Can marital selection explain the differences in health between married and divorced people? Public Health 1998;112: 11317.[ISI][Medline]
6 Cheung YB, Sloggett A. Health and adverse selection into marriage: evidence from a study of the 1958 British birth cohort. Public Health 1998;112:30911.[ISI][Medline]
7 Fu H, Goldman N. Incorporating health into models of marriage choice: demographic and sociological perspectives. J Marriage Family 1996;58:74058.[ISI]
8 Kiernan K. Demographic Experience in Early Adulthood: A Longitudinal Study. PhD thesis. London: University of London, 1987.
9 Kiernan K. Who remains childless? J Biosoc Sci 1989;21:38798.[ISI][Medline]
10 Martikainen P, Valkonen T. Mortality after death of spouse in relation to duration of bereavement in Finland. J Epidemiol Community Health 1996;50:26468.[Abstract]
11 Hemstrom O. Is marriage dissolution linked to differences in mortality risks for men and women? J Marriage Family 1996;58:36678.[ISI]
12 Kaprio J, Sarna S, Fogelholm M, Koskenvuo M. Total and occupationally active life expectancies in relation to social class and marital status in men classified as healthy at 20 in Finland. J Epidemiol Community Health 1996;50:65360.[Abstract]
13 Korenman S, Goldman N, Fu H. Misclassification bias in estimates of bereavement effects. Am J Epidemiol 1997;145:9951002.[Abstract]
14 Young M, Benjamin B, Wallis C. The mortality of widowers. Lancet 1963;ii:45456.
15 Bowling A. Mortality after bereavement: a review of the literature on survival periods and factors affecting survival. Soc Sci Med 1987; 24:11724.[ISI][Medline]
16 Bowling A, Windsor J. Death after widow(er)hood: an analysis of mortality rates up to 13 years after bereavement. Omega J Death Dying 1995;31:3549.[ISI]
17 Stroebe MS, Stroebe W, Gergen KJ, Gergen M. The broken heart: reality or myth? Omega J Death Dying 1981;12:87104.[ISI]
18 Ben-Shlomo Y, Smith GD, Shipley M, Marmot MG. Magnitude and causes of mortality differences between married and unmarried men. J Epidemiol Community Health 1993;47:20005.[Abstract]
19 Ebrahim S, Wannametee G, McCallum A, Walker M, Shaper AG. Marital status, change in marital status, and mortality in middle-aged British men. Am J Epidemiol 1995;142:83442.[Abstract]
20 Hu Y, Goldman N. Mortality differentials by marital status: an international comparison. Demography 1990;27:23350.[ISI][Medline]
21 Hajdu P, Mckee M, Bojan F. Changes in premature mortality differentials by marital status in Hungary and in England and Wales. Eur J Public Health 1995;5:25964.
22 Kiernan K. Partnership behaviour in Europe: recent trends and issues. In: Coleman D (ed.). Europe's Population in the 1990s. Oxford: Oxford University Press, 1996, pp.6291.
23 Grundy E. Population ageing in Europe. In: Coleman D (ed.). Europe's Population in the 1990s. Oxford: Oxford University Press, 1996, pp.26796.
24 Cox BD, Blaxter M, Buckle ALJ et al. The Health and Lifestyle Survey. London: Health Promotion Research Trust, 1987.
25 Cox BD, Huppert FA, Whichelow MJ. The Health and Lifestyle Survey: Seven Years On. Aldershot: Dartmouth, 1993.
26 Gershuny J, Brice J. Looking backwards: family and work 1900 to 1992. In: Buck N, Gershuny J, Rose D, Scott J (eds). Changing Households: The BHPS 1990 to 1992. Essex: ESRC Research Centre on Micro-social Change, 1994, pp.2760.
27 Idler EL, Benyamini Y. Self-rated health and mortality: a review of twenty-seven community studies. J Health Soc Behavior 1997;38:2137.[ISI][Medline]
28 Tanner JM. Growth as a measure of the nutritional and hygienic status of a population. Horm Res 1992;38(Suppl.1):10615.[ISI][Medline]
29 Macintyre S, West P. Social, developmental and health correlaties of attractiveness' in adolescence. Sociol Health Illness 1991;13:15267.
30 Leon DA, Smith GD, Shipley M, Strachan D. Adult height and mortality in London: early life, socioeconomic confounding, or shrinkage? J Epidemiol Community Health 1995;49:59.[Abstract]
31 Peck AMN, Vagero DH. Adult body height, self perceived health and mortality in the Swedish population. J Epidemiol Community Health 1989;43:38084.[Abstract]
32 Strand GC, Garr MS. Driving under the influence. In: Hirschi T, Gottfredson MR (ed.). The Generality of Deviance. New Brunswick, NJ: Transaction, 1994, pp.13148.
33 Health Education Authority. That's the Limit. A Guide to Sensible Drinking. London: Health Education Authority, 1989.
34 Shaper AG. The unhealthy abstainers' question is still important. Addiction 1995;90:48890.[ISI][Medline]
35 Grambsch PM, Therneau TM. Proportional hazards tests and diagnostics based on weighted residuals. Biometrika 1994;81:51526.[ISI]
36 Office for National Statistics. Marriage, Divorce and Adoption Statistics. Series FM 2, No. 23. England and Wales. London: The Stationery Office, 1998: Table 3.12.
37 Office for National Statistics. Marriage, Divorce and Adoption Statistics. Series FM 2, No. 23. England and Wales. London: The Stationery Office, 1998: Table 4.1.
38 Miller-Tutzauer C, Leonard KE, Windle M. Marriage and alcohol use: a longitudinal study of maturing out. J Stud Alcohol 1991;52:43440.[ISI][Medline]
39 Umberson D. Gender, marital status and the social control of health behaviour. Soc Sci Med 1992;34:90717.[ISI][Medline]
40 Jones DR. Heart disease mortality following widowhood: some results from the OPCS Longitudinal Study. J Psychosom Res 1987;31:32533.[ISI][Medline]
41 Fox J, Goldblatt P. OPCS Longitudinal Study, 19715, Socio-Demographic Mortality Differentials. Series LS, No. 1. London: HMSO, 1982.
42 Marmot MG, Shipley MJ, Rose G. Inequalities in deathspecific explanations of a general pattern? Lancet 1984;i:100306.
43 Jones DR, Goldblatt PO, Leon DA. Bereavement and cancer: some data on deaths of spouses from the Longitudinal Study of Office of Population Censuses and Surveys. Br Med J 1984;289:46164.[ISI][Medline]