Cancer incidence in parents who lost a child

A nationwide study in Denmark

Authors

  • Jiong Li M.D., M.Sc.,

    Corresponding author
    1. Department of Epidemiology and Social Medicine, the Danish Epidemiology Science Center, University of Aarhus, Aarhus, Denmark
    • Department of Epidemiology and Social Medicine, the Danish Epidemiology Science Center, University of Aarhus, Vennelyst Boulevard 6, DK-8000, Aarhus C, Denmark
    Search for more papers by this author
    • Fax: 011-45-8613-1580

  • Christoffer Johansen M.D., Ph.D.,

    1. Department of Psychosocial Cancer Research, Institute of Cancer Epidemiology, Danish Cancer Society, Copenhagen, Denmark
    Search for more papers by this author
  • Dorthe Hansen M.D., Ph.D.,

    1. Department of Epidemiology and Social Medicine, the Danish Epidemiology Science Center, University of Aarhus, Aarhus, Denmark
    2. National Institute of Public Health, Copenhagen, Denmark
    Search for more papers by this author
  • Jørn Olsen M.D., Ph.D.

    1. Department of Epidemiology and Social Medicine, the Danish Epidemiology Science Center, University of Aarhus, Aarhus, Denmark
    Search for more papers by this author

Abstract

BACKGROUND

It has been debated whether psychological stress causes cancer, but the scientific evidence remains contradictory. The objective of this study was to investigate whether the death of a child is related to cancer risk in bereaved parents.

METHODS

The authors undertook a follow-up study based on national registers. All 21,062 parents who lost a child from 1980 to 1996 were recruited for the exposed cohort together with 293,745 randomly selected, unexposed parents. Cox proportional hazards regression models were used to evaluate the relative risk of cancer incidence up to 18 years after the bereavement. The main outcomes of interest were all incident cancers, breast carcinoma, smoking-related malignancies (International Classification of Diseases [ICD] 7 codes 140, 141, 143–149, 150, 157, 160–162, 180, and 181), alcohol-related malignancies (ICD7 codes 141, 143–146, 148–150, 155, and 161), virus/immune-related malignancies (ICD7 codes 155, 171, 191, 200–202, and 204), lymphatic/hematopoietic malignancies (ICD7 codes 200–205), and hormone related malignancies (ICD7 codes 170, 172, 175, and 177).

RESULTS

The authors observed a slightly increased overall cancer risk in bereaved mothers (relative risk [RR], 1.18; 95% confidence interval [95%CI], 1.01–1.37; P = 0.028) at 7–18 years of follow-up. There was an increased risk for smoking-related malignancies (RR, 1.65; 95%CI, 1.05–2.59; P = 0.010) among bereaved mothers during the 7–18 years of follow-up. The authors observed no significantly increased relative risk of breast carcinoma, alcohol-related malignancies, virus/immune-related malignancies, or hormone-related malignancies.

CONCLUSIONS

The current data suggest that the death of a child was associated with a slightly increased overall cancer risk in mothers and that the increase may be related to stress-induced adverse life styles. Cancer 2002;95:2237–42. © 2002 American Cancer Society.

DOI 10.1002/cncr.10943

The belief that psychological stress causes malignant disease has a long history,1 although scientific evidence for such an association remains contradictory to date.2 Some case-control studies have shown a strong effect of stress,3, 4 whereas others have reported negative findings.5 Cohort studies have shown little evidence of the association.6–10 It has been pointed out that most of the case-control studies have been subjected to weaknesses in methodology, like recall bias, inadequate control of confounders,3–5 and prospective studies with limited observation periods and loss of patients to follow-up.10 Many of the studies were based on self-reported exposure or outcome, which are vulnerable to bias.11, 12

Psychological stress has effects on the endocrine and immune systems, which may enhance neoplastic growth by increasing levels of corticosterone, prolactin, and estrogen or by decreasing the number and activity of lymphocytes and natural killer (NK) cells.13, 14 Stress also may increase adverse life-style behaviors, like smoking and heavy alcohol drinking,15–17 which may increase cancer incidence after some induction and latency time.

We investigated the effect of parental bereavement on the overall and specific incidence rates of malignant disease in parents who lost a child. The study was conducted as a large, longitudinal study based on nationwide registers, which have complete and unbiased information on both exposure and outcomes of interest. We used the death of a child as the indicator of stress to avoid misclassification of the exposure or insufficient exposure.18, 19 We avoided selection bias related to incomplete follow-up, and we used detailed data on socioeconomic factors for confounder control.20

MATERIALS AND METHODS

Study Design

First, we identified all children in Denmark who died before age 18 years from 1980 to 1996 from the Register of Birth and Death Statistics.21 We then identified the parents of the deceased children by using the unique personal identification number, which is given to all citizens at birth by the Register of Population Statistics.21 These parents were included in the exposed cohort. The family structure of the exposed families was identified on January 1st in the year the child died. We then randomly sampled 15 times as many families from the remaining population, matched with the exposed family structure according to the number of parents and the number of children in 6 age subgroups (ages < 1 year, 1–2 years, 3–6 years, 7–9 years, 10–14 years, and 15–17 years) on January 1st in the year the index child died. Parents in these families were included in the unexposed cohort.

Information on the deceased children (age, gender, date of birth, date of death, and cause of death) was obtained from the Fertility Database.21 Information on cohort members (date of birth, date of death, gender, education level, residence, cause of death, and cancer incidence) was obtained from the Prevention Register20 and from the Danish Cancer Registry.22 All register linkages were made by means of the unique personal identification number mentioned above. The National Board of Health and The Danish Data Protection Agency approved the project.

Outcome and Follow-Up

Follow-up started when participants were recruited to the cohorts and ended when they were diagnosed with malignant disease, died, or emigrated or at the end of 1997, whichever came first. The main outcomes of interest were all incident malignancies, breast carcinoma, and four subgroups of malignancies that were hypothesized, a priori, to be associated with psychological stress: smoking-related malignancies (International Classification of Diseases [ICD] 7 codes 140, 141, 143-150, 157, 160–162, 180 and 181), alcohol-related malignancies (ICD7 codes 141, 143-146, 148–150, 155, and 161), virus and immune-related malignancies (ICD7 codes 155, 171, 191, 200–202, and 204), lymphatic/hematopoietic tissue malignancies (ICD7 codes 200–205), and hormone-related malignancies (ICD7 codes 170, 172, 175, and 177).3–10

Statistical Analysis

Relative risk (RR) estimates and 95% confidence intervals (95%CIs) were calculated as the measurement of association between the exposure and the outcomes using Cox proportional hazards regression analysis23 according to the SAS PHREG procedure (SAS Institute, Cary, NC).24 Sociodemographic characteristics of the cohort members, such as age at entry (ages < 30 years, 30–39 years, and ≥ 40 years), gender (male or female), education level (basic, secondary or higher, or no information) and residence (cities with a population > 100,000 inhabitants or other places) were included in the analyses as potential confounders. Two parents at the same address were considered to be cohabitating, which is equivalent to marriage in Denmark. The place of residence was included because it is related to social support. The number of children also was included because it is related to both emotional and social support as well as coping capacity in bereaved parents.

We examined the overall relative differences of cancer incidence between exposed and unexposed cohort members according to the length of follow-up, the type of death of the deceased child (either unexpected death [sudden death by unknown cause, ICD8 codes 795.0–795.9 and ICD10 codes R95–R97; motor vehicle accidents, ICD8 codes 810.0–823.0 and ICD10 codes V01–V89; suicide, ICD8 codes 950.0–959.9 and ICD10 codes X60–X84; other accidents and violence, ICD8 codes 800.0–807.9, 825.0–949.9, and 960.0–999.9 and ICD10 codes V90–V99, W00–X59, and X85–Y89] or death by other causes). Similar analytic strategies also were applied to specific malignant diseases and to subgroups of malignant diseases.

RESULTS

A total of 12,512 children died in Denmark from 1980 to 1996. There were 440 children in the population registers without a link to their parents at the beginning of the year the child died. The remaining 12,072 children were linked to their families. Characteristics of the deceased children (age group and type of death) are shown in Table 1.

Table 1. Characteristics of the Deceased Children that Defined the Exposed Cohort
CharacteristicNo. of decreased children (%)
BoysGirlsTotal
  • a

    Unexpected death: sudden death by unknown cause, International Classification of Disease (ICD) 8 codes 795.0–795.9 and ICD 10 codes R95–R97; motor vehicle accidents, ICD8 codes 810.0–823.0 and ICD10 codes V01–V89; suicide, ICD8 codes 950.0–959.9 and ICD10 codes X60–X84; other accidents and violence, ICD8 codes 800.0–807.9, 825.0–949.9, and 960.0–999.9 and ICD10 codes V90–V99, W00–X59, and X85–Y89.

Total7128494412,072
Age group   
 <1 month2595 (36.4)1900 (38.4)4495 (37.2)
 1–11 months949 (13.3)633 (12.8)1582 (13.1)
 1–2 yrs1070 (15.0)863 (17.5)1933 (16.0)
 3–9 yrs938 (13.2)663 (13.4)1601 (13.3)
 10–17 yrs1576 (22.1)885 (17.9)2461 (20.4)
Type of death   
 Unexpected deatha2306 (32.3)1306 (26.4)3612 (29.9)
 Other death4822 (67.7)3638 (73.6)8460 (70.1)

There were 21,062 parents in the exposed cohort and 293,745 parents in the unexposed cohort. The 314,807 cohort members had an average follow-up of 10.5 years, equal to 3,316,649 person-years. Table 2 summarizes the baseline characteristics of the cohorts. We found no large differences in mean follow-up, age, residence, education level, or family size between the exposed group and the unexposed group.

Table 2. Baseline Characteristics of the Two Cohorts
CharacteristicNo. of parents (%)
Exposed cohortUnexposed cohort
Total patients21,062293,745
Age at study entry (yrs)  
 <308566 (40.7)110,032 (37.5)
 30–398610 (40.9)126,823 (43.2)
 40+3886 (18.5)56,890 (19.4)
 Mean32.732.8
 Range22–4722–47
Gender  
 Male9841 (46.7)135,828 (46.2)
 Female11,221 (53.3)157,917 (53.8)
Follow-up (yrs)  
 Mean10.410.5
 Range (yrs)2–182–18
Education  
 Basic14,624 (69.4)198,460 (67.6)
 Secondary or higher3632 (17.2)60,009 (20.4)
 No information2806 (13.3)35,276 (12.0)
Residence  
 Cities6011 (28.6)91,041 (31.0)
 Other places15,020 (71.4)202,498 (69.0)
 No information31 (0.0)206 (0.0)
No. of children in the family  
 112,451 (59.1)172,393 (58.7)
 25590 (26.5)80,365 (27.4)
 ≥ 33021 (14.4)40,987 (14.5)
No. of parents in the family  
 12053 (9.8)27,521 (9.4)
 219,009 (90.2)266,224 (90.6)

Table 3 presents the overall cancer risk among parents. We observed a small but statistically significant increased risk in mothers at 7–18 years of follow-up (RR, 1.18; 95% CI, 1.01–1.37; P = 0.028). However, the risk was close to expected values among fathers. Parents who lost a child unexpectedly experienced a slightly higher RR, but the difference was not statistically significant. The age of the child on the date of death did not modify the effect of bereavement (data not shown). With the exception of one, all risk estimates for mothers were higher compared with the risk estimates for fathers.

Table 3. Relative Risk of Cancer Incidence in Bereaved Parents According to the Length of Follow-Up and the Type of Death of the Deceased Child: Cox Regression Analyses
Length of follow-up and type of death of the deceased childBoth parentsFathersMothers
E/URR (95%CI)E/URR (95%CI)E/URR (95%CI)
  • RR: relative risk (adjusted for age group, gender, education, residence, number of children in the family, and number of parents in the family); E/U: number of patients in the exposed group/number of patients in the unexposed group; 95%CI: 95% confidence interval.

  • a

    Unexpected death: sudden death by unknown cause, International Classification of Diseases (ICD) 8 codes 795.0–795.9 and ICD10 codes R95–R97; motor vehicle accidents, ICD8 codes 810.0–823.0 and ICD10 codes V01–V89; suicide, ICD8 codes 950.0–959.9 and ICD10 codes X60–X84; other accidents and violence, ICD8 codes 800.0–807.9, 825.0–949.9, and 960.0–999.9 and ICD10 codes V90–V99, W00–X59, and X85–Y89.

All follow-up      
 All child deaths461/62371.09 (0.99–1.20)176/25201.00 (0.86–1.17)285/37171.15 (1.02–1.30)
  Unexpected deathsa180/62371.15 (0.99–1.33)72/25201.07 (0.84–1.35)108/37171.20 (0.99–1.45)
  Other deaths281/62371.03 (0.97–1.09)104/25200.98 (0.89–1.08)177/37171.06 (0.98–1.14)
1–6 yrs of follow-up      
 All child deaths155/21971.03 (0.86–1.20)54/8500.90 (0.68–1.18)101/13471.10 (0.91–1.34)
  Unexpected deathsa63/21971.12 (0.88–1.45)23/8501.02 (0.67–1.54)40/13471.20 (0.87–1.64)
  Other deaths92/21970.91 (0.88–1.09)31/8500.91 (0.76–1.08)61/13471.02 (0.90–1.16)
7–18 yrs of follow-up      
 All child deaths306/40401.13 (1.01–1.27)122/16701.06 (0.88–1.27)184/23701.18 (1.01–1.37)
  Unexpected deathsa117/40401.15 (0.96–1.39)49/16701.09 (0.82–1.45)68/23701.19 (0.94–1.52)
  Other deaths189/40401.06 (0.98–1.14)73/16701.02 (0.91–1.15)116/23701.08 (0.98–1.19)

With regard to specific disease sites, we found a RR for breast carcinoma in exposed mothers of 1.10 (95% CI, 0.89–1.35; P = 0.386). The RR for all smoking-related malignancies was 1.65 (95% CI, 1.05–2.59; P = 0.010) in mothers at years 7–18 of follow-up, and we did not observe excess risks for alcohol-related malignancies, virus/immune-related malignancies, or hormone-related malignancies (Table 4).

Table 4. Relative Risk of Cancer Incidence for Subgroups of Bereaved Parents According to the Length of Follow-Up: Cox Regression Analyses
Disease site and length of follow-upBoth gendersFathersMothers
E/URR (95%CI)E/URR (95%CI)E/URR (95%CI)
  1. RR: relative risk (adjusted for age group, gender, education, residence, number of children in the family, and number of parents in the family); E/U: number of patients in the exposed group/number of patients in the unexposed group; 95%CI: 95% confidence interval.

Smoking-related malignancies      
 All follow-up (yrs)70/7631.33 (1.04–1.70)46/5091.28 (0.95–1.73)24/2541.42 (0.94–2.16)
  1–614/2050.97 (0.56–1.66)11/1431.06 (0.58–1.95)3/620.72 (0.23–2.29)
  7–1856/5581.47 (1.12–1.93)35/3661.37 (0.97–1.94)21/1921.65 (1.05–2.59)
Alcohol-related malignancies      
 All follow-up (yrs)15/2011.08 (0.64–1.83)11/1561.01 (0.55–1.86)4/451.35 (0.48–3.75)
  1–63/600.72 (0.23–2.28)3/450.95 (0.30–3.02)0/15
  7–1812/1411.24 (0.69–2.24)8/1111.04 (0.51–2.13)4/302.05 (0.72–5.80)
Virus/immune-related malignancies      
 All follow-up (yrs)154/21031.08 (0.92–1.27)56/8790.92 (0.70–1.21)98/12241.19 (0.97–1.46)
  1–656/7661.06 (0.81–1.39)19/3060.89 (0.56–1.41)37/4601.17 (0.84–1.64)
  7–1898/13371.09 (0.89–1.34)37/5730.94 (0.68–1.32)61/7641.20 (0.94–1.56)
Lymphatic/hematopoietic malignancies      
 All follow-up (yrs)24/3431.01 (0.67–1.53)10/2010.70 (0.37–1.32)14/1421.47 (0.85–2.54)
  1–6 years8/1280.89 (0.44–1.82)5/800.88 (0.36–2.16)3/480.90 (0.28–2.91)
  7–18 years16/2151.08 (0.65–1.80)5/1210.59 (0.24–1.44)11/941.76 (0.94–3.30)
Hormone-related malgnancies (mothers only)      
 All follow-up113/15851.08 (0.89–1.31)
  1–6 years14/1541.17 (0.71–2.13)
  7–18 years99/14311.06 (0.86–1.30)

DISCUSSION

We found that the death of a child was associated with a slightly increased overall cancer incidence in mothers. There was an excess risk for smoking-related malignancies, especially lung carcinoma. We did not observe an excess risk for breast carcinoma, alcohol-related malignancies, hormone-related malignancies, or immune/virus-related malignancies at follow-up. Cancer incidence was similar in the two cohorts during the first 6 years of follow-up.

Our study is in line with most of the previous studies, suggesting that an increased overall risk of developing malignant disease attributed to stressful life events is small if it exists at all.2, 6–10 Only bereaved mothers, compared with bereaved fathers, experienced a slightly increased risk, which does not support a general association. If stress has a dose-response effect, then severe types of loss would be expected to have more detrimental effects on health.25 We noted that the RR for parents who lost a child unexpectedly was similar to the RR for parents whose loss of a child was not unexpected. These observations may refute any strong role of psychological stress in the causation of malignant disease. Previous studies have shown that stress is related to a higher prevalence of smoking, heavy drinking, and lower levels of physical activity.15–17 In addition, mothers often experience more psychological stress compared with fathers after the loss of a child.26, 27 Consequently, exposed mothers may have more adverse risky behaviors, which may explain the slight increase in overall cancer incidence among mothers.

It is reassuring that this study showed no association with breast carcinoma. The importance of psychological factors related to breast carcinoma has been debated heavily in the literature.3–10 We found no effect, even after this extreme stressor, which is a strong argument against the hypothesis of a correlation between stress and breast carcinoma. Negative findings for other hormone-related malignancies are not in agreement with one previous report.9 Our data also did not corroborate the elevated risks for lymphatic/hematopoietic malignancies and for immune/virus-related malignancies that were found in another study.8 Although stress has been linked to reduced functioning in immune and endocrine systems that may influence the development of malignant disease,13, 14 our findings did not support this hypothesis. The increased risk in parents for developing malignancies at sites associated with smoking also may be attributed mainly to stress-induced, adverse behavior factors, consistent with the findings from one previous study.28

We present for the first time a nationwide prospective study based on the most severe stressor we know of, the loss of a child. We expected this exposure to create stress in all parents, regardless of personality, coping style, social support, or social network.18, 19 If there is a threshold effect for stress, and if this threshold is within the levels of public health concern under normal conditions, then we believe that most of the exposed cohort members were above this threshold. The large sample size enabled us to study bereaved fathers and mothers separately. This study is without loss to follow-up, recall, or selection bias, which have been weaknesses in the methods used for a majority of previously published studies in this area of research. Regarding incident cancer, the Cancer Registry in Denmark is considered almost complete; therefore, the possibility of differential misclassification is minimal, because we used the same diagnostic system during the period of follow-up.29 Reliable information on socioeconomic factors provided data for confounder control.20 All data were collected for administrative purposes, independent of the hypothesis under study, and were of comparable and good quality for the two cohorts.20–22, 29

It may be argued that the increased risk for some malignancies may be confounded by baseline risky life-style factors, like smoking and drinking alcohol, for which we have no data. The lack of an overall effect in fathers and no difference during the first 6 years of follow-up may refute this. The enrolled parents were young and had a low risk of developing malignant disease, which reduced the statistical power, especially for rare types of malignancies. These estimates should be interpreted with caution. The current data suggest a slightly increased risk of developing malignant disease in bereaved mothers, probably due to adverse life-style behaviors after parental bereavement.

Acknowledgements

The authors thank Mr. Jørn Hansen Schmidt and Mr. Søren Leth-Sørensen for their valuable help in establishing the cohorts.

Ancillary