Suicide After Breast Cancer: an International Population-Based Study of 723 810 Women

Abstract

Women with breast cancer account for one in five of all cancer survivors in the United States, numbering approximately 2 million in 2001 ( 1 ) . Given the current 10-year relative survival of 78% ( 2 ) , most patients with breast cancer can expect to live for many years. Nonetheless, a diagnosis of breast cancer has been associated with excess risks of suicide of 30%–50% in Scandinavia ( 3 – 5 , 6 ) and Canada ( 7 ) , albeit based on small studies. Suicide risk among patients with breast cancer in the United States has not been evaluated. Further, only a few studies have examined the effect of patient age, race, disease stage, and calendar year on suicide risk or quantified risk 5 years or more after a breast cancer diagnosis ( 3 , 4 , 6 ) . Absolute risk of suicide among patients with breast cancer has not been estimated to date.

We evaluated 723 810 patients diagnosed with a first primary breast cancer between January 1, 1953, and December 31, 2001, who survived at least 1 year and were reported to one of 16 population-based cancer registries in Denmark, Finland, Norway, Sweden, or the United States through the Surveillance, Epidemiology, and End Results (SEER) Program ( 8 , 9 ) . The SEER program areas include Connecticut (from 1973), Hawaii (from 1973), Iowa (from 1973), New Mexico (from 1973), Utah (from 1973), San Francisco-Oakland (from 1973), Detroit (from 1973), Seattle-Puget Sound (from 1974), Atlanta (from 1975), Los Angeles (from 1992), San Jose-Monterey (from 1992), and Alaska (from 1992). Study dates varied according to registry ( Table 1 ). We included a subset of women with extended follow-up, described in previous reports ( 3 – 6 ) . Mean follow-up duration starting 1-year after breast cancer diagnosis was 7.7 years (range = <1 month to 49 years).

We identified suicide among patients reported to the Scandinavian registries through linkage with the World Health Organization Mortality database for the years 1954–2002 ( 10 ) and suicide among breast cancer patients reported to the SEER registries through mortality data collected and maintained by the National Center for Health Statistics and made available through the SEER program ( 11 ) . Mortality codes for suicide were as follows: E963 and E970–979 (International Classification of Diseases, 7th Revision [ICD-7]), E950–E959 (ICD-8 and ICD-9), and X60–X84 (ICD-10) ( 12 – 15 ) . We calculated standardized mortality ratios (SMRs) and 95% confidence intervals (CIs) ( 16 , 17 ) . We multiplied accumulated person-years at risk by suicide rates for each registration area, 5-year age group, and calendar-year interval to calculate the number of expected cases of suicide in each category. To test for heterogeneity in SMRs according to categories of covariates, such as registry, we used standard likelihood ratio tests that compared Poisson regression models, which included the covariate, with a simple model without the covariate ( 18 ) . We also used likelihood ratio tests to evaluate the statistical significance of the following interactions: 1) calendar year and time since breast cancer diagnosis and type of treatment and 2) age and time since breast cancer diagnosis, stage, and type of treatment. We compared models with main effects for the two variables plus an interaction term to simpler models that included only main effects. All tests were two-sided, with a .05 cutoff for statistical significance ( 18 ) . We also quantified the excess absolute risk (EAR) of suicide among patients with breast cancer compared with the background population ([number of observed cases – number of expected cases) × 100 000 person-years) and the crude probability of suicide among patients with breast cancer by use of a competing risk model ( 19 ) . Because suicide deaths could be miscoded as accidents ( 3 , 20 ) , we also examined mortality attributed to accidents and to other external causes.

In this cohort, 836 patients with breast cancer committed suicide (SMR = 1.37, 95% CI = 1.28 to 1.47; EAR = 4.1 per 100 000 person-years) ( Table 1 ). SMRs were lowest in Denmark (SMR = 1.25, 95% CI = 1.07 to 1.46) and Sweden (SMR = 1.27, 95% CI = 1.12 to 1.45) and highest in the United States, with 245 deaths in the US cohort (SMR = 1.49, 95% CI = 1.32 to 1.70), and in Finland (SMR = 1.53, 95% CI = 1.28 to 1.83), but differences across registries were not statistically significant ( P for heterogeneity = .19) ( Table 1 ). EARs were highest for Finland (EAR = 7.6 per 100 000 person-years) and Denmark (EAR = 6.3 per 100 000 person-years) and lowest for the United States (EAR = 3.0 per 100 000 person-years). The SMRs and corresponding EARs are not parallel because of variation in background suicide rates among countries ( 10 , 11 , 21 ) . Elevated SMRs were observed for all calendar-year periods, including for women diagnosed between 1990 and 2001 (SMR = 1.36, 95% CI = 1.18 to 1.57), but were slightly higher among those diagnosed before 1970 ( P for heterogeneity = .07). SMRs varied by race, with risk higher in black women (SMR = 2.88, 95% CI = 1.44 to 5.17) than white women (SMR = 1.36, 95% CI = 1.27 to 1.46) and other races (SMR = 1.02, 95% CI = 0.44 to 2.01) ( P for heterogeneity = .06). EARs, however, were similar in black women and white women. Slight variations in risk according to age at breast cancer diagnosis were not statistically significant. Elevated SMRs were evident throughout the follow-up period, including for 25 or more years after diagnosis (SMR = 1.35, 95% CI = 0.82 to 2.12), but were slightly, but not statistically significantly, higher in the initial years after diagnosis. Risk increased with increasing stage of breast cancer ( P for heterogeneity = .08); all suicides among women with distant disease occurred among white women. Small differences in risk according to initial treatment were not statistically significant. Analyses based on type of breast cancer surgery, which was available for US women (1983–2001), showed little difference in SMRs among women who received breast-conserving surgery (SMR = 1.22, 95% CI = 0.89 to 1.64) or radical mastectomy (SMR = 1.30, 95% CI = 1.04 to 1.63) (data not shown in table). None of the interactions between covariates that we evaluated was statistically significant. Cumulative probabilities of suicide 5, 10, 20, and 30 years after breast cancer diagnosis were 0.05% (95% CI = 0.04% to 0.06%), 0.10% (95% CI = 0.09% to 0.11%), 0.16% (95% CI = 0.15% to 0.17%), and 0.20% (95% CI = 0.18% to 0.22%), respectively. The SMR for accidents and other external causes of death was 0.95 (95% CI = 0.93 to 0.99).

This study is the largest population-based investigation to date of suicide after a breast cancer diagnosis and the first analytic study of US women. New findings include the persistence of elevated suicide risk than 25 or more years after breast cancer diagnosis, with a slight diminution in risk among patients diagnosed after 1969 compared with earlier periods, perhaps because of improved treatments ( 4 ) ; a statistically significant 50% increase in suicide among US women; increasing suicide risk with increasing breast cancer stage; and higher risk in black women than white women and women of other races.

Although most patients with breast cancer appear to adapt well, a minority suffer from significant psychosocial distress ( 22 ) . Estimates of the prevalence of clinical depression or depressive symptoms in patients with breast cancer vary greatly, ranging from 1.5% to 46% ( 23 ) . Importantly, psychosocial factors (including depression) and use of antidepressant medications do not appear to be risk factors for the development of breast cancer ( 24 , 25 ) . Depression, however, was an important predictor of late-stage breast cancer diagnosis in one study ( 26 ) but not in another ( 27 ) . A Danish investigation ( 3 ) that included a subset of our population found no association between suicide and stage at breast cancer diagnosis. However, in our analysis an association with stage was evident in the extended Danish cohort and overall. We did not have information on depression, but our results are consistent with reports of increased risks of hospitalization for affective disorders and anxiety in patients with breast cancer compared with the general population ( 20 ) .

Our findings should be considered in light of the strengths and limitations of cancer registry and mortality data. Population-based investigations circumvent the selection bias in clinical or hospital studies and permit evaluation of mortality among a large number of patients. Possible limitations include the problem of accurate ascertainment of suicide, which tends to be underreported as a cause of death ( 28 ) , possibly more so in cancer patients than in the general population ( 3 , 4 , 6 , 29 ) . Notably, mortality from accidents and other external causes was not elevated in our analysis. Ascertainment of suicide also varies according to certification procedures ( 28 ) , but all analyses were controlled for country and calendar year to take into account reporting differences ( 21 ) . We also lacked information on possible confounding factors, notably marital status and socioeconomic status. Because unmarried women have higher rates of suicide ( 28 ) and breast cancer ( 30 ) , adjustment for this variable might attenuate our observed risk estimates. In contrast, adjustment for socioeconomic status would likely make little difference because socioeconomic status is directly associated with breast cancer risk ( 30 ) but is not clearly associated with female suicide risk ( 31 , 32 ) .

Nonetheless, our international, population-based study highlights a small, but statistically significant, increased long-term risk of suicide after a breast cancer diagnosis. In contrast, suicide risk for most cancers decreases to expected levels several years after diagnosis ( 6 ). This difference may reflect the unique chronic and indolent nature of breast cancer with recurrences possible many years after diagnosis. Thus, although the cumulative probability of suicide is small, our results suggest that long-term follow-up programs for breast cancer survivors should include resources devoted to psychosocial concerns.

References