Introduction
The fetal programming hypothesis proposes that in utero conditions may play an important role in the health of offspring not only shortly after birth but throughout life (Barker et al. Reference Barker, Eriksson, Forsen and Osmond2002; Gluckman et al. Reference Gluckman, Hanson, Cooper and Thornburg2008). Evidence from experimental studies supports the hypothesis that overexposure to stress hormones has a programming effect (Kapoor et al. Reference Kapoor, Dunn, Kostaki, Andrews and Matthews2006; Drake et al. Reference Drake, Tang and Nyirenda2007). Excessive glucocorticoids following stress in pregnancy can cross the placenta and cause dysfunction of hypothalamic-pituitary-adrenal (HPA) axis. This process may negatively affect fetal development and lead to adverse health outcomes in future life (Welberg & Seckl, Reference Welberg and Seckl2001; Fowden et al. Reference Fowden, Sibley, Reik and Constancia2006). There are also some preliminary evidence from humans suggesting specific exposure to stressful life event during pregnancy may induce negative effects on an individual's physiology and health (Mulder et al. Reference Mulder, Robles de Medina, Huizink, Van den Bergh, Buitelaar and Visser2002; Wadhwa, Reference Wadhwa2005; Kapoor et al. Reference Kapoor, Dunn, Kostaki, Andrews and Matthews2006; Drake et al. Reference Drake, Tang and Nyirenda2007; Cottrell & Seckl, Reference Cottrell and Seckl2009). The loss of a close relative has been classified as one of the most severe life events (Skodol & Shrout, Reference Skodol and Shrout1989) and it is likely to cause stress regardless of coping mechanisms (Stroebe et al. Reference Stroebe, Schut and Stroebe2007). We used maternal bereavement by the death of a close relative during prenatal life as a proxy of prenatal stress, but bereavement further leads to other effects such as social or familial disadvantages, which could contribute to an increased risk of mortality in offspring. Epidemiological studies have shown that preconceptional and prenatal stress due to maternal bereavement has been associated with a higher risk of spontaneous abortion, stillbirth and early childhood mortality (Wisborg et al. Reference Wisborg, Barklin, Hedegaard and Henriksen2008, Reference Wisborg, Barklin, Hedegaard and Henriksen2013; Class et al. Reference Class, Khashan, Lichtenstein, Langstrom and D'Onofrio2013, Reference Class, Mortensen, Henriksen, Dalman, D'Onofrio and Khashan2015; Laszlo et al. Reference Laszlo, Svensson, Li, Obel, Vestergaard, Olsen and Cnattingius2013; Marinescu et al. Reference Marinescu, Foarfa, Pirlog and Turculeanu2014), but little is known about the potential long-term effects of prenatal stress on mortality in later childhood, in adolescence, and in adulthood. Several register-based studies have shown that maternal bereavement shortly before and during pregnancy is associated with increased utilization of primary healthcare service in offspring (Li et al. Reference Li, Yang, Guldin, Vedsted and Vestergaard2015), hospital admissions due to mental and physical disease (Hansen et al. Reference Hansen, Lou and Olsen2000; Khashan et al. Reference Khashan, Abel, McNamee, Pedersen, Webb, Baker, Kenny and Mortensen2008), which can probably lead to an increased mortality risk in the long run.
We hypothesized that children born to mothers who lost a close relative during pregnancy or 1 year before pregnancy have a higher mortality rate. We used national registers from two Nordic countries, where mortality data from childhood to early adulthood are available. Bereavement due to the death of a child/spouse is considered as the most stressful type of bereavement, and bereavement by unnatural death might be more severe than bereavement from other causes (Li et al. Reference Li, Precht, Mortensen and Olsen2003, Reference Li, Laursen, Precht, Olsen and Mortensen2005, 2010; Zhu et al. Reference Zhu, Olsen, Sorensen, Li, Nohr, Obel, Vestergaard and Olsen2013). We thus expected that children of mothers who lost a child had a higher risk of death than other children. We also anticipated that the magnitude of the associations varies according to causes of death and age of the children (Li et al. Reference Li, Vestergaard, Cnattingius, Gissler, Bech, Obel and Olsen2014).
Method
Study population
We conducted a population-based cohort study by combining national registers from Denmark and Sweden. All live-born and new residents in Denmark and Sweden are assigned a unique civil personal identity code, which allows accurate linkage of individual information from various national registries (Li et al. Reference Li, Vestergaard, Obel, Cnattingus, Gissler and Olsen2011). We included all children born in Denmark from 1973 to 2004 (n = 19 44 899) and in Sweden from 1973 to 2006 (n = 33 08 609). As the causes of neonatal deaths (the first 4 weeks of life) are different from later occurring deaths (Heron, Reference Heron2013), we focused on mortality in children surviving beyond 28 days. Therefore, follow-up started 28 days after birth and ended at death, emigration, or end of follow-up (31 December 2009 in Denmark and 31 December 2008 in Sweden), whichever came first. Children who emigrated during the follow-up were treated as the censored at the time of emigration. We included only singletons because multiple births have different growth patterns (Alexander et al. Reference Alexander, Kogan, Martin and Papiernik1998; Alexander & Salihu, Reference Alexander and Salihu2005) and a higher infant mortality rate than singletons (Office for National Statistics, 2012).
Exposure
We classified offspring with prenatal exposure if their mother lost an older child, a spouse/partner, a sibling, or a parent during the period from 1 year before pregnancy to the day of delivery. We further dichotomized exposure into two categories: (1) losing a child or a spouse or (2) losing a sibling or a parent, assuming that the intensity of prenatal stress due to maternal bereavement after losing a child or a spouse is stronger than losing a sibling or a parent (Skodol & Shrout, Reference Skodol and Shrout1989). We also divided the causes of death in the deceased relative into natural deaths (from diseases or medical conditions: ICD-8 codes 000-799, ICD-9 codes 000-799, and ICD-10 codes A00-R99) and unnatural deaths (from external causes of injuries and poisoning: ICD-8 codes E800-E999, ICD-9 codes E800-E999, and ICD-10 codes V01-Y98) according to the European Shortlist for Causes of Death (Eurostat, 1998).
Outcomes
The main outcomes of interest were all-cause mortality, cause-specific mortality, and mortality by type of death (natural or unnatural death) in offspring. We obtained information on the cause of death from the Danish Causes of Death Register and the Swedish Causes of Death Register. In Denmark, the International Classification of Disease, 8th version (ICD-8) was used to classify the cause of death before 1994, whereas the ICD-10 has been used since 1994. In Sweden, the ICD-8 was used from 1973 to 1986, whereas the ICD-9 was used from 1987 to 1996, and the ICD-10 since 1997. We used the European Shortlist for Causes of Death to dichotomized offspring's type of death into natural death or unnatural death as defined above. We further evaluated cause-specific mortality rates from: infections and parasitic diseases (ICD-8 and ICD-9 codes 000–1399, ICD-10 codes A00–B999); neoplasms (ICD-8 and ICD-9 codes 140–2389, ICD-10 codes C00–C999); endocrine, nutritional, and metabolic diseases (ICD-8 and ICD-9 codes 240–2799, ICD-10 codes E00–E909); mental and behavioral disorders (ICD-8 and ICD-9 codes 290–3159, ICD-10 codes F00–F999); diseases of the nervous system and the sense organs (ICD-8 and ICD-9 codes 320–389, ICD-10 codes G00–H95); diseases of the circulatory system (ICD-8 and ICD-9 codes 410–414, 420–423, 425–429, ICD-10 codes I20–I25, I30–I33, I39–I52); diseases of the respiratory system (ICD-8 and ICD-9 codes 460–519, ICD-10 codes J00–J99); diseases of the digestive system (ICD-8 and ICD-9 codes 520–579, 4442, ICD-10 codes K00–K93); congenital malformations (ICD-8 and ICD-9 codes 740–759, ICD-10 codes Q00–Q99); transport accidents (ICD-8 and ICD9 codes 800–848, ICD-10 codes V01–V99); suicide and intentional self-harm (ICD-8 and ICD-9 codes 950–959, ICD-10 codes X60–X84); and other ICD codes not mentioned above were classified into the group of ‘other diseases’.
Covariates
Potential confounders were selected a priori based on previous literature (Li et al. Reference Li, Vestergaard, Cnattingius, Gissler, Bech, Obel and Olsen2014, Reference Li, Yang, Guldin, Vedsted and Vestergaard2015): preterm birth, birth weight, maternal age, parity, Apgar score at 5 min, maternal education, maternal social status, maternal smoking during pregnancy, country of birth, sex of child, birth year of the child (<1980, 1980–1989, 1990–1999, 2000–2008), and age of child (<1, 1–4, 5–9, 10–14, 15–19, ⩾20 years).
Statistical analysis
Kaplan–Meier method was performed to estimate cumulative mortality in the exposed and the unexposed. We used a Poisson regression model to estimate mortality rate ratios (MRRs) with 95% confidence interval (CI) to assess the effect of prenatal maternal bereavement on offspring mortality according to all-cause mortality and cause-specific mortality. When estimating the cause-specific mortality, death from other causes were considered as competing events and treated as censored cases. In addition, analyses were stratified by time of follow-up (28–364 days, 1–4, 5–9, 10–14, 15–19, ⩾20 years). We also divided the exposure window into five periods (7–12 months before pregnancy, 0–6 months before pregnancy, first trimester, second trimester, third trimester) to examine whether a potential effect of prenatal maternal bereavement on offspring mortality differed across these periods.
As the information on some variables was only available for specific time periods, we created missing indicators for variables with missing values. In a sensitivity analysis, we repeated the analysis while including only children born after 1978, 1979, 1980, 1982, 1987, 1991, 1994, 1997, and the year with full availability of all covariates (completed data analysis). We also restricted the analyses to offspring with a gestational age of ⩾37 weeks, a birth weight of ⩾2500 g, and an Apgar score of 10 at 5 min. We undertook further analysis for country of birth and sex of the offspring separately. In addition, we conducted sibling-matched analysis, comparing exposed child to his/her unexposed half-siblings born to the same mothers, to control for shared genetic, social and environmental factors. We performed stratified Cox regression for the sibling sub-cohort. The stratified Cox regression includes a separate stratum for each family identified by the mother's unique identification number in which only sibling pairs discordant for both maternal bereavement and death are informative and contribute to the effect estimate; thus, each family has its own baseline rate function reflecting the family's shared genetic and environmental factors. The association between prenatal maternal bereavement and mortality is analyzed among siblings in the same family and therefore adjusted for genetic and environmental factors that are shared among the siblings. The graphs were created by Stata/SE 12 (Stata Corporation, USA) and the analyses were conducted using SAS 9.2 (SAS Institute Inc., USA).
Results
Among 52 53 508 singletons followed up to 37 years of age, 1 29 733 (2.47%) individuals were born to mothers who lost a close relative during pregnancy or up to 1 year before pregnancy, including 20 955 (0.40%) individuals of mothers who lost a child or spouse and 1 08 778 (2.07%) individuals of mothers who lost a parent or sibling. A total of 2 75 419 (5.24%) individuals had emigrated at the end of follow-up [28–364 days: 15 591 (0.30%), 1–4 years: 55 582 (1.06%), 5–9 years: 41 410 (0.79%), 10–14 years: 19 510 (0.37%), 15–19 years: 30 184 (0.57%), ⩾20 years: 1 13 142 (2.15%)]. The exposed and unexposed cohorts were comparable in terms of most baseline characteristics at birth, although mothers who experienced bereavement were more likely to be older, to have higher parity, and to report smoking (Table 1).
Baseline characteristics of exposed and unexposed cohorts
Available periods of variables. Preterm birth, maternal age, parity: 1973–2004 in Denmark and 1973–2006 in Sweden; birth weight: 1979–2004 in Denmark and 1973–2006 in Sweden; Apgar score at 5 min: 1978–2004 in Denmark and 1973–2006 in Sweden; maternal education: 1979–2004 in Denmark and 1990, 1995, 2000, 2005 in Sweden; maternal social status: 1980–2004 in Denmark and 1980, 1985, 1990 in Sweden; maternal smoking during pregnancy: 1991–2004 in Denmark and 1982–2006 in Sweden.
During the total of 10 01 64 346 person-years at risk (median 18.9, interquartile range 11.1–27.2), 36 800 of 52 53 508 persons died (0.70%); 961 (0.74%) in the exposed cohort and 35 839 (0.70%) in the unexposed cohort. The exposed had a higher cumulative mortality than the unexposed in both countries (Supplementary Fig. S1). Prenatal exposure to maternal bereavement was associated with a 3–18% increased mortality rate in offspring (MRR 1.10, 95% CI 1.03–1.18). The association was statistically significant only in children born to a mother who lost a child or spouse (MRR 1.28, 95% CI 1.14–1.44), but not in children born to a mother who lost a sibling or a parent (MRR 1.04, 95% CI 0.96–1.12). The MRRs tended to be higher in offspring of mothers who lost a relative due to unnatural death (MRR 1.20, 95% CI 0.99–1.47) compared to those who lost relative to natural death (MRR 1.09, 95% CI 1.02–1.17), but the difference in risk estimates was small and did not reach statistical significance (Table 2). We found a stronger association between maternal bereavement and offspring mortality by natural causes (Table 2) than for offspring mortality by unnatural causes, where prenatal stress exposure was not significantly associated with increased risk. The second trimester appeared to be the most sensible period of exposure, but statistically significant association was only observed for maternal bereavement and offspring mortality due to natural cause (MRR 1.36, 95% CI 1.11–1.66) (Table 3).
Prenatal maternal bereavement and MRRs in offspring by relationship to the deceased relative and cause of death of the deceased relative
MRR, Mortality rate ratio; CI, confidence interval.
a Adjusted for preterm birth, birth weight, maternal age, parity, Apgar score at 5 min, maternal education, maternal social status, maternal smoking during pregnancy, and other factors related to offspring (country of birth, sex, birth year, age group).
MRR in offspring by the timing of exposure to prenatal maternal bereavement
MRR, Mortality rate ratio; CI, confidence interval.
a Adjusted for preterm birth, birth weight, maternal age, parity, Apgar score at 5 min, maternal education, maternal social status, maternal smoking during pregnancy, country of birth, sex of child, birth year of child.
The MRRs of bereavement by losing a child/spouse on all-cause mortality were high up until 10 years of age, after which the mortality attenuated to a level close to that seen in the background population (Fig. 1). The trend was only observed for natural causes of death.
Mortality rate ratio (MRR) in offspring by the length of follow-up time. (a) All-cause death mortality rate ratio over follow-up time; (b) natural death MRR over follow-up period; (c) unnatural death MRR over follow-up period).
Compared to unexposed, exposed children were more likely to have higher mortality rates in most groups of causes of death (Table 4). However, statistically significant associations were observed only in deaths from infectious and parasitic disease, endocrine, nutritional and metabolic diseases, disease of the nervous system, and congenital malformations. We found slightly higher effect estimates of bereavement on mortality rate in females (MRR 1.14, 95% CI 1.03–1.27) than in males (MRR 1.09, 95% CI 1.01–1.18), but the difference was not statistically significant. The complete data analysis and the sub-analysis restricted to individuals born after 1978, 1979, 1980, 1982, 1987, and 1991 showed that offspring exposed to prenatal stress had higher mortality, suggesting MRRs between 1.08 and 1.18. The association between any bereavement and offspring mortality was slightly stronger in the Danish population (MRR 1.13, 95% CI 1.01–1.26) than in the Swedish (MRR 1.05, 95% CI 0.97–1.13), but no differences were found for maternal loss of an older child or a spouse during pregnancy (MRR in Denmark: 1.20; 95% CI 1.00–1.43; MRR in Sweden: 1.19; 95% CI 1.02–1.38). When the analysis was restricted to the offspring with a gestational age of ⩾37 weeks, a birth weight of ⩾2500 g, and an Apgar score of 10 at 5 min, the overall MRR was 1.12 (95% CI 1.02–1.23), which was also similar to the result from analyzing the whole study population. Results obtained from sibling-matched analysis were similar to those from as the main analysis (Supplementary Table S1).
Cause-specific MRRs in offspring by relationship to the deceased relative
MRR, Mortality rate ratio; CI, confidence interval.
a Adjusted for preterm birth, birth weight, maternal age, parity, Apgar score at 5 min, maternal education, maternal social status, maternal smoking during pregnancy, and other factors related to offspring (country of birth, sex, birth year, age group).
Discussion
In this population-based cohort study, we found that maternal bereavement during pregnancy and 12 months before pregnancy was associated with an increased long-term mortality in offspring up to 37 years. The increased mortality rate tended to be greater for maternal loss of a child or spouse than for loss of a sibling or parent, and a stronger association between maternal bereavement and offspring mortality was observed for maternal loss by unnatural death than for loss by natural death. We only observed statistically significant associations in offspring in which the causes of their deaths were due to natural causes by diseases or medical conditions but not unnatural death by external causes of injuries and poisoning. The association between maternal bereavement and offspring mortality was more pronounced before the age of 10 years. We found the second trimester might be the most sensitive period for death by natural cause, but did not find a significant association in death by unnatural cause.
Possible pathways
Our findings that show a significant association between maternal bereavement and offspring mortality only by natural cause lend support to fetal programming theory (Barker et al. Reference Barker, Eriksson, Forsen and Osmond2002). Possible mechanisms have been proposed to explain intergenerational transmission of stress in humans with a special focus on HPA axis (Mulder et al. Reference Mulder, Robles de Medina, Huizink, Van den Bergh, Buitelaar and Visser2002; Glover, Reference Glover2015). Increased maternal glucocorticoids level after prenatal stress may activate the HPA axis, the sympathetic nervous system and adrenal medulla system, leading to the increased cortisol, corticotrophin-releasing hormone, (nor) adrenaline, and adrenocorticotropic-releasing hormone level. Excessive maternal cortisol can reduce the activity of the placental 11β-HSD2, which serves as a barrier to protect the fetus from relatively high maternal glucocorticoid levels, to reduce fetal growth by leading to increased transplacental passage of active cortisol. Another possible mechanism is that the activation of the sympathetic nervous system may result in reduced blood flow to the uterus and fetus, leading to fetal growth restriction. In addition, prenatal stress has been associated with increased cytokines, suggesting prenatal stress may affect the function of immune system and inflammation. In response to these changes, the developing fetus is able to adapt to the prevailing condition to promote survival, alter tissue size, structures, and function. These effects may have short-term adaptive benefits, but might also increase the risk of later diseases (Gitau et al. Reference Gitau, Cameron, Fisk and Glover1998; Drake et al. Reference Drake, Tang and Nyirenda2007; Cottrell & Seckl, Reference Cottrell and Seckl2009). Besides the direct physiological effect, unhealthy behaviors and changes in social network following bereavement may also contribute to the observed associations. Evidence from human studies and animal models have shown that maternal stress during pregnancy is associated with higher offspring risks, including neonatal mortality, neuroendocrine dysfunction, risks of cardio-metabolic disorders, preterm birth, and congenital heart defects (Hansen et al. Reference Hansen, Lou and Olsen2000; Patin et al. Reference Patin, Lordi, Vincent, Thoumas, Vaudry and Caston2002; Khashan et al. Reference Khashan, Abel, McNamee, Pedersen, Webb, Baker, Kenny and Mortensen2008, Reference Khashan, McNamee, Abel, Mortensen, Kenny, Pedersen, Webb and Baker2009; Li et al. Reference Li, Olsen, Vestergaard and Obel2010; Zhu et al. Reference Zhu, Olsen, Sorensen, Li, Nohr, Obel, Vestergaard and Olsen2013; Plana-Ripoll et al. Reference Plana-Ripoll, Liu, Momen, Parner, Olsen and Li2016).
Comparison with previous studies
To the best of our knowledge, this study is probably one of the largest relating maternal bereavement to childhood mortality, and the first to relate maternal bereavement to mortality in adolescence and in adulthood with adjustment for multiple confounders. As expected (Hansen et al. Reference Hansen, Lou and Olsen2000; Li et al. Reference Li, Precht, Mortensen and Olsen2003, Reference Li, Laursen, Precht, Olsen and Mortensen2005, 2010; Zhu et al. Reference Zhu, Olsen, Sorensen, Li, Nohr, Obel, Vestergaard and Olsen2013), maternal bereavement by the death of a child or her partner was associated with higher risk of adverse outcome in the offspring than the death of one of her siblings or a parent. This suggests that only the most stressful types of maternal bereavement may be related to increased mortality in the offspring.
We found a stronger association of maternal bereavement on offspring mortality due to natural causes of death, specifically deaths from infectious and parasitic disease, endocrine, nutritional and metabolic diseases, disease of the nervous system, and congenital malformations. The evidence in humans and rodents suggests that prenatal stress is associated with metabolic syndrome, increased healthcare use due to infections, and neuroendocrine disorders (Drake et al. Reference Drake, Tang and Nyirenda2007; Nielsen et al. Reference Nielsen, Hansen, Simonsen and Hviid2011; Li et al. Reference Li, Olsen, Vestergaard, Obel, Kristensen and Virk2012, Reference Li, Yang, Guldin, Vedsted and Vestergaard2015). The number of deaths attributable to theses causes were relatively small in our follow-up, and some of our previous studies did indicate that prenatal stress may increase the risk of psychosis (Li et al. Reference Li, Olsen, Vestergaard and Obel2010) in the offspring. Therefore, even we did not find an overall association between prenatal stress and death due to mental or behavioral disorders, this does not necessarily mean that the offspring may not have an excessive risk of these diseases.
We did not observe an association between prenatal stress and offspring mortality by unnatural causes, such as accidents and suicide. This finding is in line with a previous study on injury risk after the exposure (Virk et al. Reference Virk, Li, Lauritsen and Olsen2013). However, we have previously shown (Li et al. Reference Li, Vestergaard, Cnattingius, Gissler, Bech, Obel and Olsen2014; Guldin et al. Reference Guldin, Li, Pedersen, Obel, Agerbo, Gissler, Cnattingius, Olsen and Vestergaard2015) that the loss of a parent during childhood is associated with a marked increased risk of all-cause mortality and suicide. The mechanism underlying this difference between prenatal exposure and postnatal exposure is not clear, and future research is needed.
The mechanisms by which prenatal stress influence offspring health and mortality may depend on the timing of exposure (Cottrell & Seckl, Reference Cottrell and Seckl2009; Kapoor et al. Reference Kapoor, Kostaki, Janus and Matthews2009). There is little consistency in the literature about the timing of exposure in which prenatal stress is most harmful to the fetus and it is likely that the time window of susceptibility depends on the outcome studies (Glover, Reference Glover2015). Our results indicated that the second trimester of pregnancy may be an important period for long-term mortality, which is in agreement with our previous results of the effect of maternal bereavement on offspring risk of type-2 diabetes (Li et al. Reference Li, Olsen, Vestergaard, Obel, Kristensen and Virk2012) and preterm delivery (Laszlo et al. Reference Laszlo, Li, Olsen, Vestergaard, Obel and Cnattingius2016). A possible explanation is that the protective effect of human placental enzyme 11β-HSD2 that converts active glucocorticoids to inactive metabolites (Krozowski, Reference Krozowski1999) does not appear until 20–24 weeks of gestations.
The association only observed in natural cause of death could be due to genetic confounding in which genes-related mortality leads to a positive association. However, the results yielded from sibling-matched analysis controlling for genetic factors were consistent with those from the main analysis that prenatal maternal bereavement was associated with an increased risk in children who died of natural causes. Our study showed that the association between maternal bereavement and offspring mortality was more pronounced before the age of 10 years. As congenital malformations, infectious and parasitic disease, and disease of the nervous system are the most common causes of death in children under 10 years of age and account for 37% of deaths in Denmark and Sweden (Yu et al. Reference Yu, Qin, Cnattingius, Gissler, Olsen, Zhao and Li2016), the effect in children under 10 years could be partly explained by the association between maternal bereavement and increased mortality by these three aforementioned causes of death. No apparent association was found between maternal bereavement and mortality in adolescents (10–19 years). One possible explanation might be that injuries, poisoning, and cancer are the most common causes of death in this age group (Wolfe et al. Reference Wolfe, Macfarlane, Donkin, Marmot and Viner2014), which might not be associated with prenatal stress. We found slightly increased mortality caused by natural death in young adults (⩾20 years) if exposed to early life bereavement. However, this needs further evaluation to see whether the association is causal or due to chance.
We were able to perform sibling-match analysis to control for confounding related to genetic predisposition or family environment, which are often difficult to be control in a conventional cohort study. This approach is particularly suitable for exploring the association of prenatal maternal exposure on offspring outcome (Donovan & Susser, Reference Donovan and Susser2011). Compared with conventional analysis for cohort studies, sibling analysis has low statistical power because of the rarity of discordant pairs. However, the register data from two countries with a long follow-up time up to 37 years enable us to mitigate the influence of reduced sample size to some extent. Another potential limitation was that we cannot determine if the confounding was due to genetic predisposition, family environment or both. But our sibling-matched analysis basically yielded similar results as the cohort analysis, indicating it is less likely that our result was confounded by genes or shared family environment.
Strengths and limitations
Our study has several methodological strengths. First, bereavement is considered as a good indicator of severe stress, regardless of social support and ability to cope (Hansen et al. Reference Hansen, Lou and Olsen2000; Li et al. Reference Li, Precht, Mortensen and Olsen2003, Reference Li, Vestergaard, Cnattingius, Gissler, Bech, Obel and Olsen2014). High-quality register-based data from Denmark and Sweden provide prospective and precise measures of the loss of a close relative with respect to time, cause of death, and type of relative. Second, the large cohort also provided detailed information on covariates, which allowed us to adjust for a broad range of potential confounders. Third, the population-based cohort including all individuals with virtually complete follow-up over a long time period also eliminates the potential influence of selection bias, which is a common problem in case-control studies. Compared with the previous study on offspring mortality (Class et al. Reference Class, Khashan, Lichtenstein, Langstrom and D'Onofrio2013, Reference Class, Mortensen, Henriksen, Dalman, D'Onofrio and Khashan2015), we have a larger sample size with higher statistical power. This provides a unique opportunity to examine cause-specific mortality, and the long follow-up period enables us to investigate both short- and long-term effects of maternal bereavement on offspring mortality.
A number of limitations must also be acknowledged. We had limited data on maternal lifestyle factors and no data on alcohol and other substances abuse, caffeine consumption, not attending regular antenatal care, maternal social support networks, and risk factors in offspring after birth, which could potentially confound the association between prenatal stress and mortality in offspring. However, these negative lifestyle factors may be related to maternal smoking and socioeconomic status. Thus adjustment for maternal smoking and socioeconomic status may have partly controlled for their effects. In our study, we examined only the most severe types of stress. However, more studies are needed to decide whether the increased risk might also be found for more common but less severe types of stress. The data on subjective experience of the bereavement would help to classify the degrees of stress. However; we do not have such data, which is also a limitation of the study.
Conclusion
Severe prenatal maternal bereavement was associated with an increased long-term mortality risk in the offspring. Specifically, an association was only observed in children who died due to natural causes. Our results support the fetal programming hypothesis that prenatal stress may contribute to ill health from physical diseases later in life.
Supplementary material
The supplementary material for this article can be found at http://dx.doi.org/10.1017/S003329171600266X.
Acknowledgements
This study was supported by a grant from the European Commission's Seventh Framework Programme/European Research Council (ERC-2010-StG-260242-PROGEURO to J.L.). Data recruitment was also supported by grants from the Danish Medical Research Council (project no. 09-072986 to J.L.), the Swedish Council for Working Life and Social Research (grant no. 2010-0092 to S.C.), the Nordic Cancer Union (2013-78760 and 2015_176673 to J.L.), DFF-6110-00019 (To JL), and Karen Elise Jensens Fond (2016) (to J.L.). M.V. is supported by an unrestricted grant from the Lundbeck Foundation (grant no. R155-2012-11280). The funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Declaration of Interest
None.
