It is unknown whether severe emotional stress due to loss of a child influences the risk of cancers susceptible to immune modulation such as infection-related cancers. We conducted a historic cohort study in 1990 to 2004 on the basis of the Swedish Multi-Generation Register including 4,687,073 parents. Death of a child was identified through the Causes of Death Register. Poisson regression was used to derive the relative risks (RR) and 95% confidence intervals (CI) of infection-related cancers, comparing the incidence rates of parents who lost a child with those who never lost a child. A total of 101,306 parents (2%) had lost a child during follow-up, among whom 1,608 subsequently developed infection-related cancers. After adjustment for age, sex, calendar year, educational level, and civil status, the overall RR of 14 cancers studied was 1.07 (95% CI: 1.02–1.12). Parents who lost a child were particularly at a higher risk for cancers potentially associated with human papilloma virus (HPV) infection such as cervical cancer (RR: 1.46; 95% CI: 1.17–1.80). Higher RRs for most cancers were obtained within 5 years after child loss and excess risk for liver and stomach cancers was confined to that period. No association was observed for lymphoma and nonmelanoma skin cancer at any time point after child loss. Although potential confounding by unmeasured factors cannot be ruled out, our findings lend support to the hypothesis that severe life stressors, such as child loss, may raise the risk for several, chiefly HPV-related, cancers. Cancer Res; 71(1); 116–22. ©2011 AACR.
The potential influence of psychological stress on cancer development has received considerable research attention during the last decades. Still, reviews (1), editorials (2), and commentaries (3) have concluded that the evidence for such an association remains unconvincing. In addition to methodologic limitations, most previous studies have focused on breast cancer (1, 4–7) or lumped all cancers together (8, 9), leaving the potential influence of psychological stress on individual cancer sites undetected. Meanwhile, stress has been shown to be associated with increased levels of cortisol (10) and also immune suppression (11). Immune resistance has been suggested to be chiefly involved in infection-related cancers (12, 13). Thus, if any, the impact of stress might be greatest for cancers that have strong indication for infection. Immune-suppressed individuals through HIV infection or organ transplantation have been reported to have a drastically increased risk of lymphoma, Kaposi's sarcoma, and cervical cancer (14). The impact of immune modulation by stressful life events on the risk of infection-related cancers, on the other hand, awaits further study.
The loss of a child is one of the strongest emotional stressors that a person can encounter. To this end, we utilized the unique Swedish national registries to investigate the risk of developing infection-related cancers in parents who have lost a child through death.
Materials and Methods
We conducted a historic cohort study on the basis of the Swedish Multi-Generation Register (MGR) that includes people born in 1932 or later (“index persons”) together with their parents. Since 1991, tax offices responsible for the local population registration in Sweden have supplied almost complete data to the MGR (15); familial linkage is available for more than 90% of individuals who were alive in 1991. We included all parents who were born in Sweden and alive on January1, 1990, and had at least 1 child recorded as alive on that day (if born earlier) or born thereafter in the MGR. In total, 4,936,672 parents meeting these criteria were included in the analyses.
The loss of a child through death was used as an indicator ofsevere psychological stress. A total of 6,296,765 children of the study participants were identified in the MGR as alive onJanuary 1, 1990, or born afterward. Via the national registration numbers (NRN, unique identifiers assigned to all Swedish residents), we linked these children to the Causes of Death Register to identify any occurrence of death. Information on the date and cause of the death as well as age of the child at death was thereby obtained.
Using the NRNs, linkages to the Cancer Register, Causes of Death Register, and Emigration Register were conducted to identify cancer outcomes among the parents and calculate person-years. The study participants were followed from January 1, 1990, or the birth date of their first child, whichever came later, to the date of cancer diagnosis, death, emigration out of Sweden, or the end of follow-up (December31, 2004), whichever came first. A total of 249,599 parents were excluded from the cohort due to either a prevalent cancer diagnosis (n = 159,837) or inconsistencies discovered in record linkages, including death (n = 94) and emigration out of Sweden (n = 89,668) before entry to the cohort, leaving 4,687,073 (95%) in the final analyses. “Unexposed person-time” was accumulated from (a) all person-time contributed by parents who never lost a child during follow-up and (b) person-time contributed before child death from parents that did lose a child during follow-up. Among parents who lost a child, person-time accumulated after the date of the child death was used as “exposed person-time.” If a parent lost more than 1 child, the first loss was counted.
A diagnosis of infection-related cancer as recorded in the Cancer Register was defined as an outcome. As suggested by Grulich etal. (14), infection-related cancers were categorized in following groups: Epstein–Barr virus (EBV)-related cancers [International Classification of Disease (ICD)-7: 201 (Hodgkin's disease); 200 and 202 (non–Hodgkin's lymphoma)], hepatitis B/C virus (HBV/HCV)-related cancers [ICD-7: 155 and 156 (liver cancer); 1550 (primary hepatocellular carcinoma)], human papillomavirus (HPV)-related or possibly related cancers [ICD-7: 140 (lip cancer), 143–147 (cancer in oral cavity and pharynx), 150 (esophageal cancer), 1541 (anal cancer), 161 (cancer in larynx), 171 (cancer in cervix uteri), 176 (cancer in vulva and vagina), 179 (penile cancer), 191 (nonmelanoma skin cancer), and 192 (eye cancer)], and Helicobacter pylori–related cancers [ICD-7: 151 (stomach cancer)] (Table 1). Esophageal and cervical cancers were further classified into squamous cell carcinoma and adenocarcinoma, given the potentially different roles of H. pylori and HPV in their corresponding carcinogenesis (16–18). To compare the associations of child loss with invasive versus in situ cervical cancers, we conducted additional analysis for cervical cancer in situ. Similar analysis was performed for cancer in vulva (ICD-7: 1760). To avoid potential influence of shared genetic features between children and their parents, parent–child pairs with identical cancer diagnosis (parents) and underlying cause of death (children) were excluded from the analyses.
We used log-linear Poisson regression models to calculate the relative risks (RR) and 95% confidence intervals (CI), as the ratio of incidence rates of cancers among parents who lost a child to that of parents who never lost a child. RRs were adjusted for sex, age at follow-up, calendar year of follow-up, educational level (<9 or ≥9 years), and civil status [cohabitated (married or living as married) or noncohabitated (single, divorced, or widowed)] in all statistical models. Age at follow-up was adjusted for in either 5 year groups or 4 year groups (≤54, 55–64, 65–74, and ≥75 years) depending on statistical power. Information on educational level and civil status for all individuals was respectively obtained from the Swedish Education Register and the Swedish Household Census in 1990. Further analyses were performed for site-specific cancers and by time since loss (<5 and ≥5 years after loss) for overall cancer risk and the risks of the relatively common cancers (non–Hodgkin's lymphoma, liver cancer, stomach cancer, HPV-related or possibly related cancers, and nonmelanoma skin cancer).
We pooled all cancers that appeared to be associated with loss of a child (excluding Hodgkin's disease, non–Hodgkin's lymphoma, nonmelanoma skin cancer, and penile cancer), to conduct exploratory analyses among parents who lost a child by time since loss (<2, 2–4, 5–9, and ≥10 years), age at death of the deceased child (<25, 25–49, and ≥50 years), and causes of child death [unexpected (sudden death: suicide or nonsuicide) and expected (all other deaths that are not classified as sudden death: cancer or noncancer)].
To allay potential concern of confounding by parental behavior, such as excessive alcohol consumption, which could presumably have led to a higher risk of child death as well as a higher risk of certain cancers, we performed an additional analysis excluding all parents who had been hospitalized for alcoholism or liver cirrhosis. Previous hospitalization for alcoholism and liver cirrhosis was identified through linkage to the Swedish Inpatient Register which includes data on the NRNs, up to 8 discharge diagnoses, as well as time of hospital admission and discharge. To address the hypothesis that loss due to expected (e.g., cancer) but not unexpected (e.g., suicide) causes may have raised stress levels and correspondingly increased risk of infection-related cancers before the actual date of child death, we performed an additional analysis by artificially assigning a “time of stress onset” as 5 years before the date of child death and compared the incidence rates of all cancers studied during this period of time among parents who lost a child due to the “expected” causes with that of the unexposed group. The exposed parents contributed less unexposed person-time accordingly in this analysis. Finally, to test the specificity of our hypothesis, we ran similar models on bereaved parents to assess the risk of cancers with no consistent evidence for infectious etiology that is, breast cancer and prostate cancer. Pearson's χ2 test was used to assess the goodness of fit for all models. Statistical analyses were conducted using SAS version 9.1 (SAS Institute, Inc.).
The study was approved by the Ethics Review Board at the Karolinska Institutet.
A total of 56,536,390 unexposed person-years were accumulated during follow-up with 69,907 cases of infection-related cancers (incidence rate: 123.6/105 person-years). Among all study participants, 101,306 parents (2%) lost a child through death, leading to 620,376 exposed person-years with 1,624 cases of infection-related cancers (incidence rate: 261.8/105 person-years). Sixteen of the 1,624 cancers were excluded because the parents were diagnosed with the same cancer that their child died from, leaving 1,608 cancers in the final analyses (incidence rate: 259.4/105 person-years). The average age at entry to the cohort was 47.4 years (SD: 17.7) among the unexposed group, whereas 66.0 years (SD: 17.3) among the exposed group. After adjustment for age, sex, and calendar year, the overall RR of the 14 cancers studied was 1.08 (95% CI: 1.03–1.13). After additional adjustment for educational level and civil status, the overall RR became 1.07 (95% CI: 1.02–1.12). We, thus, only present results with additional adjustment for educational level and civil status.
As shown in Table 1, parents who lost a child did not have a significantly altered risk for EBV- and H. pylori–related cancers overall, whereas they had a higher risk of liver cancer (RR: 1.14; 95% CI: 1.00–1.30), especially primary hepatocellular carcinoma (RR: 1.22; 95% CI: 0.99–1.52). Among the HPV-related cancers, the risks of cancer in the cervix uteri, vulva and vagina, as well as anus appeared to be associated with child loss, although some of the excess risks were based on small numbers and not statistically significant (Table 1). Thirteen cases of cervical cancer among the exposed group were adenocarcinomas (RR: 1.32; 95% CI: 0.72–2.19); 60 cases were squamous cell carcinomas (RR: 1.33; 95% CI: 1.02–1.70); and 12 cases had unknown histology (RR: 3.18; 95% CI: 1.66–5.53). In addition, we observed 194 cases of cervical cancer in situ among mothers who lost a child, giving an RR of 1.12 (95% CI: 0.96–1.28). We also observed 10 cases of in situ cancer in vulva among the exposed mothers, giving an RR of 1.13 (95% CI: 0.56–2.00). Among possibly HPV-related cancers, statistically significant associations were observed for cancers in oral cavity and pharynx, esophagus, larynx, and eye (RRs: 1.43, 1.25, 1.49, and 1.58 respectively). Fifty-one cases of esophageal cancer were squamous cell carcinomas with an RR of 1.50 (95% CI: 1.12–1.97) and 19 adenocarcinomas with an RR of 0.77 (95% CI: 0.47–1.18). The other 7 cases had unknown histology. Nonmelanoma skin cancer was not associated with the loss of a child (Table 1). Among 489 individuals with nonmelanoma skin cancer, 432 were squamous cell carcinomas giving an RR of 0.97 (95% CI: 0.88–1.06) and 22 were basal cell carcinomas with an RR of 1.01 (95% CI: 0.66–1.54). The other 26 individuals had unknown histology.
Table 2 shows the RRs for infection-related cancers by time since child loss. An increased overall risk was seen only during the first 5 years after loss. For most site-specific cancers, the RRs were higher during the first 5 years after loss compared with thereafter, although the RRs of nonmelanoma skin cancer were similar across the time periods. The excess risk of liver cancer was confined to the first 5 years, whereas an increased risk for all HPV-related cancers was noted both before and beyond 5 years after loss. Parents who lost a child had an excess risk of stomach cancer during the first 5 years after loss but a decreased risk during longer follow-up (Table 2).
In a more detailed analysis among all cancers positively associated with the loss of a child, the increased risk was seen only during the first 5 years after loss overall (Table 3). The age of the child at death did not clearly alter the cancer risk, nor was the association different with regard to whether the death of the child was expected or unexpected. Only child death at the age of 25 to 49 years reached statistical significance. Among “unexpected deaths,” suicide entailed a slightly stronger risk of cancer among the parents compared with nonsuicide death. A similar difference was noted for cancer versus noncancer deaths. A child's death from infection-related cancers yielded a higher RR than a child's death from non–infection-related cancers; the difference was however not statistically significant.
In additional analysis, we excluded all parents with hospitalization for alcoholism or liver cirrhosis (n = 119, 209; 2.5%); the RRs of overall infection-related cancers (RR: 1.06; 95% CI: 1.01–1.12) as well as site-specific cancers were largely unchanged. For example, the RRs were 1.28 (95% CI: 1.16–1.42) for HPV-related or possibly related cancers but nonmelanoma skin cancer; 1.46 (95% CI: 1.17–1.82) for cervical cancer; 1.11 (95% CI: 0.96–1.28) for liver cancer; 1.54 (95% CI: 1.21–1.95) for cancer in oral cavity and pharynx; and 1.57 (95% CI: 1.14–2.09) for squamous cell esophageal cancer. In a second additional analysis using the artificial “time of stress onset” of 5 years before the actual date of child death, the overall RR for all cancers studied during this 5-year time window was 1.16 (95% CI: 1.06–1.27) among parents who lost a child to cancer and 0.87 (95% CI: 0.68–1.10) among parents who lost a child to suicide. During the first 5 years after the child's death, the corresponding RRs were: 1.15 (95% CI: 1.04–1.27) and 1.15 (95% CI: 0.94–1.38). Finally, parents who lost a child did not have, compared with nonbereaved parents, higher risk of breast cancer (RR: 0.96; 95% CI: 0.89–1.02) or prostate cancer (RR: 0.96; 95% CI: 0.91–1.02).
Our nationwide study of 4.6 million Swedes indicates that parents who lose a child are, compared with nonbereaved parents, at increased risk for cancers with an established or suspected infectious etiology. This association was particularly consistent for cancers linked to HPV infection. The elevated risk was more prominent within the first 5 years after child loss.
To our knowledge, our study is the largest systematic investigation of the hypothesized association. We are only aware of 2 previous studies addressing associations between child loss and the risk for some infection-related cancers (8, 9). Li etal. studied all cancers and found a weakly increased risk for “virus/immune-related malignancies” among 21,062 bereaved Danish parents, particularly for mothers, but the overall association was not statistically significant and no data for individual cancer sites were presented (9). Levav etal. found that compared with the cancer incidence in the general population, 6,284 Israeli parents who had lost an adult son to war or accident had an increased risk of lymphatic/hematopoietic cancers (8). The reason for the discrepancy between these findings and ours is not obvious and given the lack of data on individual cancers in these earlier studies, a direct comparison to our results is not possible.
The strengths of the present study include the large sample size, population-based study design, prospectively collected data, and the complete follow-up. A number of limitations should, however, be noted. First, the excess risks observed are generally modest (overall around 1.2 to 1.5). Although, we believe that the loss of a child is still the strongest global indicator of life events inducing severe emotional stress, other stressful life events are widespread in the general population and may, as such, also influence cancer risk. This might entail an underestimation of the relative risks among “exposed” compared with “unexposed” individuals.
With a lack of detailed covariates in register-based studies, the major limitation of our study is unmeasured confounders including potentially shared genetic or environmental vulnerabilities between the parents and children. For instance, the shared vulnerability for infections, infection-related malignancies in particular, may lead to a premature death of the child and a later diagnosis of infection-related cancer of the parent. We have made every effort to address such a concern. Already when designing the study, we excluded all child–parent pairs with the child dying from and parent diagnosed with the same cancer. We further calculated the RR among the bereaved parents by the cause of child death which showed that there was no statistically significant difference between an “expected” and “unexpected” child death. Although the highest RR was obtained for a child death due to infection-related cancers, the impact of psychological stress was evident for a child death due to other causes. Specifically, a child death due to non–infection-related cancers entailed a statistically significantly 19% increased risk of infection-related cancers among the bereaved parents.
The distinct time-dependent pattern of the observed associations in our data further alleviates such concern. First, the impact of confounding by shared vulnerabilities (e.g., genetics), if true, should be consistent before and after child loss, whereas our additional analyses show that the risks of infection-related cancers were only elevated after a time point when the parents are potentially emotionally alarmed, that is, before a child death to cancer but after a child death to suicide. Second, if such confounding only appears after child loss (e.g., lifestyle changes including smoking), we would expect the risk of infection-related cancers to be increasingly elevated along with time after loss since it presumably takes some period of time for changed lifestyle factors to influence cancer risk. This was clearly illustrated in a Danish study showing that the risk of smoking-related malignancies increased only beyond 7 years after loss (9).
Other indications in our data further argue against major concerns about specific potential confounders including smoking, alcohol consumption, socioeconomic status, and educational level. Firstly, the fact that the RR of cervical adenocarcinoma is similar to that of squamous cell carcinoma provides evidence against a pure confounding by smoking. In a recent study using serum samples from 5 Nordic countries, smoking was shown to be associated with squamous cell carcinoma, but not cervical adenocarcinoma, even when HPV infection was carefully adjusted for (19). Similarly, the lack of association between child loss and stomach cancer beyond 5 years after loss also argues against confounding by smoking, given that smoking has been shown as an established risk factor for stomach cancer (20). Secondly, we did not find any association between child loss and breast cancer which has repeatedly been shown to be associated with alcohol consumption (21). Unchanged results observed in the additional analysis excluding parents with alcoholism or liver cirrhosis further alleviate the concern about alcohol consumption.
Both child loss and infection-related cancers may be associated with socioeconomic or marital status. We used the highest attained educational level as a proxy for socioeconomic status and obtained marital status data from the Household Census in 1990. With time, bereaved parents might differ from nonbereaved parents with respect to change in educational level or marital status, potentially leading to a different risk of infection-related cancers. However, recent studies from Sweden have shown that parents who lost a child due to cancer were 4 to 9 years later at similar educational level (22) and even showed a slightly lower risk of divorce (23), compared with other parents. Finally, the virtually identical RRs observed in different educational level and marital status groups also help to alleviate such a concern (data not shown).
Although confounding likely remains as a partial explanation for our findings, a competing hypothesis is that the emotional stress of child loss may accelerate the cancer genesis of an already established infection. A biological mechanism for an association between severe psychological stress and infection-related cancers, especially during the recent years after the onset of stress, is plausible. Psychological stress has been associated with modified cellular defense processes against carcinogenesis, such as DNA damage repair and apoptosis as well as downregulated immune surveillance of tumor cells by reduced natural killercell activity (11, 24). Interestingly, the strongest impact of child loss was observed for cancers with well established indication for viral infection, for example, cervical cancer (HPV infection for both squamous cell carcinoma and adenocarcinoma; ref. 25). Previously, higher levels of perceived stress have been reported to increase the risk of an impaired HPV-specific immune response in women with cervical dysplasia (26).
In conclusion, in our large, population-based investigation, we found parents who lost a child to have an increased risk of some, chiefly HPV-associated, malignancies. Our findings provide evidence for the role of psychological stress in cancer development. If confirmed in other studies with detailed covariate assessment, further investigation of stress-related mechanisms that affect aspects of oncogenic viral infections, such as clearance and viral load, are warranted (27–29).
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest are disclosed.
The study was funded by the Swedish Council for Working Life and Social Research (2008-1310) and the Swedish Research Council (K2009-70X-21087-01-2).
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
- Received February 9, 2010.
- Revision received October 4, 2010.
- Accepted October 11, 2010.
- ©2010 American Association for Cancer Research.