Search tips
Search criteria 


Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
Am J Public Health. Author manuscript; available in PMC 2010 February 16.
Published in final edited form as:
PMCID: PMC2636447

Variation in the Effect of Widowhood on Mortality by the Causes of Death of Both Spouses

Felix Elwert, Ph.D.a and Nicholas A. Christakis, M.D., Ph.D., M.P.H.b,c



We investigate the effect of spousal bereavement on mortality, and we extend prior work by documenting cause-specific bereavement effects by the causes of death both of the pre-decedent spouse and the bereaved survivor.


We examined a nationally representative cohort of 373,189 elderly married couples in the United States, followed from 1993 to 2002, using competing-risk and Cox models. Covariates included the baseline health of both spouses.


In both men and women, there is significant variation in the effect of widowhood on mortality as a function of the causes of death of both spouses. The death of a spouse due to almost all causes, including various cancers, infections, and cardiovascular diseases, increases the mortality of the surviving spouse, albeit to varying degrees. Conversely, the death of a spouse increases survivor’s cause-specific mortality for almost all causes, including cancers, infections, and cardiovascular diseases, albeit to varying degrees.


The effect of widowhood on mortality varies substantially by the causes of death of both spouses, suggesting that the widowhood effect is not restricted to one aspect of human biology.


The increased probability of the recently bereaved to die—often called the “widowhood effect”—is one of the best documented examples of the effect of social relationships on health.(1) The widowhood effect has been found in bereaved men and women of all ages throughout the world.(2-5) Recent longitudinal studies put the excess mortality of widowhood (compared to marriage) among the elderly between 30% and 90% in the short run, and around 15% in the long run.(1, 6-8) These estimates are remarkably robust to rigorous statistical controls for confounding as well as complex estimation strategies,(1, 6, 8, 9) prompting increasing confidence in a causal basis of the widowhood effect.(6, 8, 10, 11)

Most previous studies on the widowhood effect, however, have focused on all-cause mortality. By comparison, much less is known about the relationship between widowhood and cause-specific risks of mortality. Cause-specificity in the widowhood effect can be traced along two dimensions: the cause of death of the pre-decedent spouse and the cause of death of the bereaved partner (if he or she dies). Research on either dimension of cause specificity is scarce, particularly with respect to accounting for the cause of death of the pre-decedent spouse. This is regrettable since cause-specificity of the widowhood effect may help illuminate the specific mechanisms by which the death of a spouse increases the mortality of the survivor and may thus help identify opportunities for health interventions.

Prior work in this area, often using a narrow list of disease categories, has yielded mixed results. For example, while several large studies find that spousal bereavement is associated with increased cancer mortality,(12-14) several other studies,(4, 7, 15-19) including the only two longitudinal studies considering multiple causes of death in the United States,(4, 7) found no statistically significant or inconsistent evidence for increased cancer mortality after widowhood, net of covariates.

The present study is the largest nationally representative and longitudinal study of cause-specificity in the widowhood effect in the United States, and also the first such study to investigate heterogeneity in the widowhood effect by the causes of death of both spouses, i.e., of the pre-decedent spouse and the bereaved partner.



We developed a very large longitudinal sample of elderly married couples in the United States from Medicare databases.(20, 21) Married couples in the 1993 Medicare Denominator file, which contains 96% of all elderly Americans,(22) were identified using a published detection algorithm.(23) In order to satisfy computer hardware constraints we drew an 8% simple random sample from the pool of all identified married couples, which we further restricted to those couples in which both spouses were either white or black, aged 67–98 at baseline, not enrolled in an HMO, living in the 50 states or D.C., and sharing the same zip code (to exclude married but separated couples). The final sample contains 373,189 married couples, or 746,378 men and women. Table 1 compares this Medicare-based sample of elderly married couples to elderly married couples in the 5% Public Use Micro Sample of the 1990 Census (using corresponding sample restrictions). This comparison demonstrates close agreement between our sample and the Census with respect to spouses’ age, poverty status, race, and region of residence (all variables common to both datasets). Follow up extends over 9 years (January 1, 1993 to January 1, 2002). There is no loss to follow up for reasons other than the terminal study outcome, death.

Table 1
Comparison between 1993 Medicare-based and 1990 5% PUMS Census samples of elderly married couples, aged 67–98 years


Death and Widowhood

The Medicare Vital Status (VS) file provides death dates for both members of every couple in the entire sample, if they died. From these death dates, we derive the outcome (time to death since January 1, 1993) and the key independent variable (widowhood). Survival times are censored on January 1, 2002. During the nine years of follow-up, 52.3% of the husbands and 32.7% of the wives in this sample died.

Cause of Death

Medicare records do not contain death certificates, but they do contain prospectively collected diagnostic histories for all individuals up to their date of death. We adapted an algorithm described by the Medicare Payment Advisory Commission to derive causes of death from the complete set of inpatient and outpatient diagnostic records during the last two years of decedents’ lives.(24) This algorithm was designed to identify the underlying health condition most likely to have killed the decedent. For example, an individual with no prior conditions who was hospitalized with a diagnosis of lung cancer three months prior to death and who later sought outpatient treatment for an eye infection would be assigned a cause of death of lung cancer. We note that, with this procedure, 48.2% of assigned causes death coincide with the primary diagnosis of decedents’ hospital record on the day of their death since most patients in the US are hospitalized at the time of their death. We grouped causes of death into 16 categories based on ICD-9. Decedents who did not have hospital or outpatient records in the two years prior to death were assigned the category “cause unknown” (8.1%) (as a 17th “cause”). Table 2 shows the frequency distribution of causes of death for decedent husbands and wives; the distribution of causes of deaths bears a satisfactory overall resemblance to the distribution of causes in elderly decedents as reported by the CDC.(25)

Table 2
Number and percentage distribution of causes of death in a Medicare-based sample of 373,189 married couples aged 67–98 years between 1993 and 2002


The analysis adjusts for numerous potentially confounding medical, social, and contextual covariates. Importantly, this is the first study on cause-specificity in the widowhood effect to adjust for the baseline characteristics of both members of the couple. On the individual level, we adjust for the baseline age and race (black or white) of both spouses.(8, 26, 27) We construct a couple-level poverty indicator at baseline from spouses’ dual Medicare-Medicaid eligibility.(28) We extract detailed health histories to adjust for differences in baseline morbidity from the 1991 and 1992 Medicare Provider Analysis and Review (MedPAR) files. We summarize the chronic disease burden at baseline by computing Charlson comorbidity scores from hospitalization records separately for each spouse in 1991 and 1992.(29) We trichotomize this measure of health burden into low, moderate, and severe (Charlson scores of 0, 1, and 2 or higher, respectively).(30) We further adjust for the number of days each partner had spent in the hospital in 1991 and in 1992. These detailed, physician-ascertained statistical controls for confounding by health status significantly exceed the health information available to previous cause-specific studies of the widowhood effect. We also adjust for couples’ Census division of residence as well as detailed measures for residential context. At the county level, we adjust for population density, violent crime rates, and the availability of medical care as reported in the Area Resource File.(31) At the zip-code level, we adjust for urbanization, demographic composition (age, race, nativity, linguistic isolation), male unemployment rates, median home value, log median income, and median education, all drawn from the Census Summary Tape File 3B.

Statistical Methods

This study presents two analyses, each estimated separately for men and women. We refer to the focal individual whose mortality risk during bereavement is being assessed—whether a husband or a wife—as the “partner” of the deceased spouse. For readability, we suppress individual-level subscripts i. First, we estimate a continuous-time Cox competing risk model to analyze the impact of the pre-decedent spouse’s death (from any cause) on partner’s hazard of dying from each of 17 causes of death.(32)

(Eq. 1)

Eq. 1 partitions partners’ hazard of dying from specific cause c at time t, hc(t), into the product of a cause-specific baseline hazard that varies freely with time, h0,c(t), and a function of the vector of explanatory variables, such that changes in the explanatory variables induce proportional shifts in the baseline hazard. The model contains a time-varying widowhood indicator, W(t), that switches from 0 to 1 on the day of spouse’s death, and a time-invariant vector of baseline covariates, X. The impact of each variable on partners’ hazard of death is allowed to vary freely across partners’ potential causes of death. Thus, the key parameter of interest, b2,c, is the effect of widowhood on a partner’s hazard of dying from cause c, net of covariates, X. Under the assumption that the cause-specific hazards of death are independent given observed covariates, the model likelihood factors naturally into c components that each depend only on parameters relating to one specific cause. We can then estimate Eq. 1 as a series of c independent cause-specific Cox models, where the outcome is censored if partners die of a different cause first.(32)

The second analysis uses standard continuous-time Cox models to estimate the impact of spouse’s death from a specific cause on partner’s overall hazard of death.

(Eq. 2)

Eq. 2 partitions partner’s overall hazard of death at time t, h(t), into a time-varying baseline hazard, h0(t), and a function of explanatory variables. The vector of covariates, X, is the same as in Eq. 1. Here, however, we enter not one overall indicator of widowhood but a vector of 17 separate, cause-specific, widowhood indicators, Wc(t), one each for every one of spouse’s possible causes of death (“spouse alive” is the reference category). The vector of key parameters of interest, b2,c, gives the impact of spouse’s death from cause c on partner’s overall hazard of death, net of covariates, X.

More complicated analyses (not shown) that simultaneously accounted for the full 17×17 matrix of potential causes of death in the partners and their spouses did not illuminate the present topic beyond the results obtained from Eqs. 1 and 2. All results reported in this study are fully adjusted for the covariates listed above. Analyses were performed using the Stata software package, release 9.2.(33)


Mortality after widowhood is significantly elevated for male and female partners. The death of a wife is associated with an 18% increase in all-cause mortality for men (hazard ratio (HR): 1.18, CI95[1.16; 1.19]), and the death of a husband is associated with a 16% increase in all-cause mortality for women (HR: 1.16, CI95[1.14; 1.17]), net of covariates.

Widowhood Effects by the Causes of Death of Surviving Partners

Figure 1 shows hazard ratios and 95% confidence intervals for the estimated effects of spouse’s death (from any cause) on partner’s cause-specific hazards of death, net of covariates. The effects of causes for which the confidence interval overlaps with the horizontal line of no effect (HR=1) are not statistically significant at the conventional 5% level. The effect of widowhood differs across partners’ 17 possible causes of death, and this difference is statistically significant (p-value<0.01) for both male and female partners. That is, widowhood does not raise the risk of all causes of death uniformly. We further note that the cause-specific widowhood effects for male partners correlate strongly with the corresponding cause-specific widowhood effects for female partners (ρ=0.77), although men on average tend to suffer somewhat stronger effects.

Figure 1
The effect of widowhood (from any cause) on the bereaved partner’s cause-specific mortality. Hazard ratios and 95% confidence intervals. Estimates from Cox models adjusted for all baseline covariates, estimated separately for male and female partners. ...

Wife’s death exerts statistically significant effects (p-values<0.05) on men’s cause-specific hazards of death for 15 out of 17 causes of death. Wife’s death increases men’s cause-specific hazards of death by more than 20% for six causes of death (in decreasing order: COPD, diabetes, accidents/serious fractures, infections/sepsis, all other known causes, and lung cancer) and also for unknown causes of death. The effect exceeds 10% for seven more causes of death (colon cancer, ischemic heart disease, CHF, nephritis/kidney disease, CVA/stroke, other heart and vascular diseases, and other cancers). The effects of wife’s death on husband’s hazards of death from influenza/pneumonia and Alzheimer’s/Parkinson’s disease are not statistically significant.

Husband’s death exerts statistically significant effects on women’s cause-specific hazards of death also for 15 out of 17 causes of death. The estimated effects exceed 20% for four causes of death (COPD, colon cancer, accidents/serious fractures, and lung cancer) and also for unknown causes. The effects exceed 10% for another seven causes (other known causes, infections/sepsis, influenza/pneumonia, nephritis/kidney disease, diabetes, other heart or vascular disease, and CHF). The impact remains statistically significant yet falls below 10% for three causes of death (CVA and stroke, ischemic heart disease, other cancers). We do not detect a statistically significant impact of husband’s death on wife’s hazard of death from rapidly fatal cancers or Alzheimer’s/Parkinson’s disease.

For male and female partners alike, the estimated effect of spouse’s death on dying from Alzheimer’s/Parkinson’s disease is negative; i.e., the death of a spouse reduces the risk of dying from these diseases, but neither effect reaches statistical significance at the conventional 5% level.

Widowhood Effects by the Causes of Death of Pre-Decedent Spouses

Figure 2 shows the impact of the pre-decedent spouse’s cause of death on the survivor’s all-cause hazard of death, net of covariates. We note that the effect of widowhood on partner’s all-cause morality varies considerably according to the cause of death of the pre-decedent spouse (p-value<0.01). As in the preceding analysis, the profile of effects found for male partners strongly correlates with the profile for female partners (ρ=0.72).

Figure 2
The effect of widowhood on partner’s all-cause mortality by the cause of death of his or her pre-decedent spouse. Hazard ratios and 95% confidence intervals. Estimates from competing risk models adjusted for all baseline covariates, estimated ...

The death of a husband or wife is associated with a statistically significant (p-value<0.05) increase in the all-cause mortality of the surviving spouse for almost all causes of death of the pre-decedent spouse. The sole exception for both men and women is bereavement due to Alzheimer’s or Parkinson’s disease, which is not associated with a statistically significant increase in the survivor’s all-cause mortality.

Men’s hazard of death increases by more than 20% if their wives died of lung cancer, infections/sepsis, COPD, other heart/vascular diseases, or diabetes. Men’s hazard of death increases by less than 20% if their wives died of any other causes. Only men whose wife died of Alzheimer’s/Parkinson’s disease do not experience a statistically significant increase in mortality.

Women’s hazard of death increases by more than 20% after widowhood only for two of their pre-decedent husbands’ causes of death, COPD and influenza/pneumonia. Women’s hazard of death increases by less than 20% in response to their husbands’ deaths from all other causes. Only women whose husband died of Alzheimer’s/Parkinson’s disease did not experience a statistically significant increase in their own mortality.


We find that the widowhood effect is not monolithic. The extent to which widowhood increases the mortality of a surviving spouse depends on the cause of death of their pre-decedent spouse. Moreover, surviving partners are at risk of some causes of death more than of others after the death of their spouse. This variation in the widowhood effect according to causes of death of both spouses provides some analytic leverage to better understand the nature and mechanism of the widowhood effect. Since over 13 million Americans are widowed,(34) and since the excess risk of mortality imposed by widowhood is non-trivial, this phenomenon is of substantial public health significance.

Our work advances prior work in this area in a number of ways. Existing research often considers only a small number of causes of death, such as the broad categories of cancer or violent deaths,(15, 35-38) uses samples with relatively small numbers of cases for each individual cause of death, uses samples that are not nationally representative,(4, 35, 36, 39) uses a cross sectional methodology,(12, 14) or lacks statistical controls for confounding beyond age, sex, and race.(12, 14, 17, 37, 38) By contrast, this study considers 17 causes of death separately for both spouses in a longitudinal and nationally representative sample of 373,189 married couples experiencing a total of 317,300 deaths while adjusting for a wide range of covariates, including baseline health, that are measured for both members of the couple.

Our results suggest several observations regarding the effect of widowhood on partners’ cause-specific mortality. First, in contrast to several,(4, 7, 15-19) but not all,(12-14) previous studies, we find a statistically significant effect of widowhood on cancer mortality for elderly men and women. The effect is particularly large for deaths from colon and lung cancer. It is much smaller (and statistically insignificant for female partners) for certain rarer cancers that typically and predictably lead to death quickly (cancer of the head and neck, upper gastrointestinal tract, liver, central nervous system, pancreas, or melanoma, lymphoma, or leukemia), and it is small but still significant for all other cancers. Second, in agreement with most,(12, 13, 16, 17, 39) though not all,(4, 7, 18) previous studies, we find a clear, positive, and statistically significant association between widowhood and partners’ mortality from vascular diseases for men and women. The effect is moderately strong, and it is approximately the same (no statistically significant difference) for all four categories of cardiovascular disease (ischemic heart disease, congestive heart failure, CVA/stroke, and other heart/vascular disease). Third, the broad effect of widowhood on the different types of cardiovascular disease, cancer, and other causes of death suggests that widowhood triggers a broad set of biopsychosocial mechanisms that affect mortality, and that the effect is not limited to one aspect of human biology. Widowhood appears to have particularly strong effects on death from causes that are either acute health events (infections/sepsis, accidents) or chronic diseases that require careful patient management to treat or prevent (diabetes, COPD, colon cancer). This points to the role of the loss of social support and social integration in widowhood as a possible origin for the widowhood effect.(40)

Many of the same observations also apply to the differences in surviving partners’ all-cause mortality as a function of the specific cause of death of the pre-decedent spouse. The lack of increased all-cause mortality following spouses’ death from Alzheimer’s/Parkinson’s disease confirms a similar finding from a British study which attributed the lack of a widowhood effect to anticipatory grief, i.e. the ability of caregivers to prepare adequately for the predictable death of their spouse.(36) The theory of anticipatory grief is also consistent with our finding of similarly small (albeit statistically significant) widowhood effects in the wake of spouse’s death from cancers that typically lead to death quickly. This suggests that it may be the predictability of the death per se, rather than the duration of spouse’s terminal illness, that may shield the survivor from some of the adverse consequences of bereavement. The lack of a widowhood effect following spouse’s death from Alzheimer’s or Parkinson’s disease, however, is also consistent with an alternative explanation that patients suffering from these diseases may simply cease to contribute to their partner’s health long before they die, or that the health consequences of caring for a person with Alzheimer’s has already been absorbed by the end of the sick spouse’s life, such that there is no measurable discontinuity in the survivor’s mortality at the actual time of spousal death.(41) Lastly, we note that the differences in male partners’ all-cause mortality across their pre-decedent wives’ illnesses or causes of death appear to be somewhat larger, in relative terms, than the differences observed for female partners. This may indicate that for women it mostly matters that their husband has died, whereas for men it additionally matters what their wife has died of.

Our analysis has several limitations. Despite adjusting for a more extensive set of potentially confounding variables than most prior research, including baseline health, there may be some residual confounding in the data that may explain certain features of the reported findings. Specifically, we note that the large effects of widowhood on mortality from COPD and lung cancer in male and female partners may be owed in part to shared behaviors, such as smoking,(1, 2) and the large effect on death by accidents/fractures may be owed to incidents involving both spouses.(6, 13, 17) On the other hand, the analysis treats same-day deaths as non-widowed deaths, and if confounding were driving all cause-specific results, we would expect that contagious diseases more generally would be associated with large effects, which is the case for infections/sepsis but not for influenza/pneumonia. Second, causes of death are derived from decedents’ diagnostic history rather than official death certificates. This leads to an unavoidable underascertainment of causes that lead to death suddenly or without prior detection.

The results of this longitudinal study of 373,189 elderly American couples show that the effect of widowhood on mortality varies substantially by the causes of death of both spouses. These results were found for male and female partners, even after adjusting for a wide range of potentially confounding factors, including the baseline health of both spouses. Widowhood increases survivors’ risk of dying from almost all causes, including cancer, but it increases the risk for some causes more than for others. The converse also holds: widowhood increases survivors’ all-cause mortality due to almost all causes of death of the pre-decedent spouse, but the actual cause of death of the pre-decent spouse makes a difference. The death of a spouse, for whatever reason, is a significant threat to health and poses a substantial risk of death by whatever cause.


This research was supported by a grant from NIH (R-01 AG17548-01) to N.A. Christakis and a Harvard University Eliot Fellowship to F. Elwert. The authors thank Laurie Meneades for expert data programming required to develop the analytic dataset, Andrew Clarkwest for computing Census statistics, T. Jack. Iwashyna for advice on developing the cause of death classifications, and Peter A. DeWan and Lawrence Wu for helpful discussions.

Human Subject Protection

This study was approved by the institutional review board of Harvard Medical School.


Felix Elwert is an Assistant Professor in the Department of Sociology at the University of Wisconsin-Madison. He is currently working on the causes and causal consequences of marital dissolution by death and divorce. Nicholas A. Christakis is Professor in the Department of Health Care Policy at Harvard Medical School and Professor in the Department of Sociology at Harvard University. He is currently studying the transmission of health phenomena in social networks.


1. Schaefer C, Quesenberry CP, Jr., Wi S. Mortality following conjugal bereavement and the effects of a shared environment. Am J Epidemiol. 1995;141(12):1142–52. [PubMed]
2. Kraus AS, Lilienfeld AM. Some epidemiologic aspects of the high mortality rate in the young widowed group. J Chronic Dis. 1959;10:207–17. [PubMed]
3. Parkes CM, Benjamin B, Fitzgerald RG. Broken heart: a statistical study of increased mortality among widowers. Br Med J. 1969;1(5646):740–3. [PMC free article] [PubMed]
4. Helsing KJ, Comstock GW, Szklo M. Causes of death in a widowed population. Am J Epidemiol. 1982;116(3):524–32. [PubMed]
5. Hu YR, Goldman N. Mortality differentials by marital status: an international comparison. Demography. 1990;27(2):233–50. [PubMed]
6. Martikainen P, Valkonen T. Mortality after the death of spouse in relation to duration of bereavement in Finland. J Epidemiol Community Health. 1996a;50(3):264–8. [PMC free article] [PubMed]
7. Johnson NJ, Backlund E, Sorlie PD, Loveless CA. Marital status and mortality: the national longitudinal mortality study. Ann Epidemiol. 2000;10(4):224–38. [PubMed]
8. Elwert F, Christakis NA. Widowhood and race. American Sociological Review. 2006;71(1):16–41.
9. Subramanian SV, Elwert F, Christakis NA. Widowhood and mortality among the elderly: The modifying role of neighborhood concentration of widowed individuals. Social Science and Medicine. 2008;66:873–884. [PMC free article] [PubMed]
10. Lillard LA, Panis CW. Marital status and mortality: the role of health. Demography. 1996;33(3):313–27. [PubMed]
11. Lillard LA, Waite lJ. Til death do us part: marital disruption and mortality. American Journal of Sociology. 1995;100:1131–56.
12. Burgoa M, Regidor E, Rodriguez C, Gutierrez-Fisac JL. Mortality by cause of death and marital status in spain. European Journal of Public Health. 1998;8(1):37–42.
13. Martikainen P, Valkonen T. Mortality after the death of a spouse: rates and causes of death in a large Finnish cohort. Am J Public Health. 1996b;86(8):1087–93. [PubMed]
14. Gove WR. Sex, marital status, and mortality. American Journal of Sociology. 1973;79(1):45–67. [PubMed]
15. 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 (Clin Res Ed) 1984;289(6443):461–4. [PMC free article] [PubMed]
16. Jones DR, Goldblatt PO. Cause of death in widow(er)s and spouses. J Biosoc Sci. 1987;19(1):107–21. [PubMed]
17. Kaprio J, Koskenvuo M, Rita H. Mortality after bereavement: a prospective study of 95,647 widowed persons. Am J Public Health. 1987;77(3):283–7. [PubMed]
18. Ebrahim S, Wannamethee 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(8):834–42. [PubMed]
19. Gore JL, Kwan L, Saigal CS, Litwin MS. Marriage and mortality in bladder carcinoma. Cancer. 2005;104(6):1188–94. [PubMed]
20. Lauderdale DS, Furner SE, Miles TP, Goldberg J. Epidemiologic uses of Medicare data. Epidemiol Rev. 1993;15(2):319–27. [PubMed]
21. Mitchell JB, Bubolz T, Paul JE, Pashos CL, Escarce JJ, Muhlbaier LH, et al. Using Medicare claims for outcomes research. Med Care. 1994;32(7 Suppl):JS38–51. [PubMed]
22. Hatten J. Medicare’s common denominator: the covered population. Health Care Financing Review. 1980;27(1):53–64. [PMC free article] [PubMed]
23. Iwashyna TJ, Zhang JX, Lauderdale DS, Christakis NA. A methodology for identifying married couples in Medicare data: mortality, morbidity, and health care use among the married elderly. Demography. 1998;35(4):413–9. [PubMed]
24. Hogan C, Lynn J, Gabel J, Lunney J, O’Mara A, Wilkinson A. Medicare beneficiaries’ costs and use of care in the last year of life. Washington, DC: 2000.
25. Arias E, Smith B. Deaths: Preliminary Data for 2001. National Center for Health Statistics; Hyattsville, MD: 2003.
26. Arday SL, Arday DR, Monroe S, Zhang J. HCFA’s racial and ethnic data: current accuracy and recent improvements. Health Care Financ Rev. 2000;21(4):107–16. [PMC free article] [PubMed]
27. Lauderdale DS, Goldberg J. The expanded racial and ethnic codes in the Medicare data files: their completeness of coverage and accuracy. Am J Public Health. 1996;86(5):712–6. [PubMed]
28. Clark WD, Hulbert MM. Research issues: dually eligible Medicare and Medicaid beneficiaries, challenges and opportunities. Health Care Financ Rev. 1998;20:1–10. [PMC free article] [PubMed]
29. Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis. 1987;40(5):373–83. [PubMed]
30. Zhang JX, Iwashyna TJ, Christakis NA. The performance of different lookback periods and sources of information for Charlson comorbidity adjustment in Medicare claims. Med Care. 1999;37(11):1128–39. [PubMed]
31. Health Resources and Services Administration BoHP . Area Resource File (ARF) System. Quality Resource Systems, Inc; Fairfax, VA: 1996.
32. Kalbfleisch JD, Prentice RL. The Statistical Analysis of Failure Time Data. Wiley; Hoboken, NJ: 2002.
33. StataCorp . Stata Statistical Softare: Release 9.2. StataCorp; College Station, TX: 2006.
34. Schoenborn CA. Marital Status and Health: United States, 1999-2002. National Center for Health Statistics; Hyattsville, MD: 2004.
35. Ryan DH. Death in dementia: a study of causes of death in dementia patients and their spouses. International Journal of Geriatric Psychiatry. 1992a;7:465–472.
36. Ryan DH. Is there a desolation effect after dementia? A comparative study of mortality rates in spouse of dementia patients following admission and bereavement. International Journal of Geriatric Psychiatry. 1992b;7:331–339.
37. Smith JC, Mercy JA, Conn JM. Marital status and the risk of suicide. Am J Public Health. 1988;78(1):78–80. [PubMed]
38. Luoma JB, Pearson JL. Suicide and marital status in the United States, 1991-1996: is widowhood a risk factor? Am J Public Health. 2002;92(9):1518–22. [PubMed]
39. 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(3):200–5. [PMC free article] [PubMed]
40. Iwashyna TJ, Christakis NA. Marriage, widowhood, and health-care use. Soc Sci Med. 2003;57(11):2137–47. [PubMed]
41. Christakis NA, Allison PD. Mortality after the hospitalization of a spouse. N Engl J Med. 2006;354(7):719–30. [PubMed]