Death by Unnatural Causes During Childhood and Early Adulthood in Offspring of Psychiatric Inpatients

Context  Offspring of psychiatric inpatients are at higher risk of death from all causes, but their cause-specific risks have not been quantified.

Objective  To investigate cause-specific deaths at 1 to 25 years in offspring of parents previously admitted as psychiatric inpatients.

Design  Population-based cohort study.

Setting  The entire Danish population.

Participants  All singleton births (N = 1.38 million) from January 1, 1973, to December 31, 1997, with follow-up to January 1, 1999. Linkage to the national psychiatric register identified all previous parental admissions.

Main Outcome Measures  Deaths from all natural causes and all unnatural causes, specifically, accidents, homicides, suicides, and undetermined causes.

Results  The highest observed relative risk (RR) was for homicide in young and older children with affected mothers or fathers. Homicides were between 5 and 10 times more likely to occur in this group, according to child's age and whether the mother or father had been admitted. There was previous parental admission in approximately one third of all child homicides. We found no evidence of increased risk of homicide in exposed young adults, but this group had a 2-fold to 3-fold higher risk of suicide. In almost one fourth of the suicides, there was a history of parental admission. Young adults with 2 previously admitted parents were 6 times more likely to kill themselves than were their peers in the general population. Relative risk of suicide or open-verdict deaths by poisoning were higher than for such deaths occurring by other means.

Conclusions  Almost 99% of children studied survived to their mid-20s. However, they were more vulnerable to death from unnatural causes, notably, homicide during childhood and suicide in early adulthood. Further research is needed to establish how parental psychopathology contributes to increased risk of premature death in these offspring.

Links between parental psychopathology and risk of death in offspring have been investigated since the 1930s.¹ Some early studies reported no evidence of higher risk of death,^2-4 but these used outmoded methods and their results may have limited relevance today. More recently published studies have indicated a higher risk of perinatal and infant death associated with maternal schizophrenia^5,6 and psychotic disorders in general.⁷ However, there is no recent population-based evidence for cause-specific risk of death after infancy in offspring of mentally ill parents.¹ Our previous study indicated increased risk of deaths from all causes up to early adulthood in this group.⁸ We report herein new findings, specifically for risk of death from unnatural causes at ages 1 to 25 years, from the same Danish birth cohort. Higher risk of these outcomes may arise as a result of several factors, including adverse social environment and neglect,^9,10 intentional physical harm,^11-13 and inheritance of suicidal behavior.^14,15

We investigated the risk of death from unnatural causes in offspring of psychiatric inpatients. We hypothesized that the risk would be increased for deaths from any unnatural cause, and specifically for accidental death in children and young adults and suicide during early adulthood. We expected sparse event data to preclude precise estimation of relative risk of homicide in this age range.

The characteristics of the study cohort are described in detail in our previous report.⁸ All live births in Denmark between January 1, 1973, and December 31, 1997, were enumerated using the Central Population Register and were followed up to January 1, 1999. This birth cohort was restricted to singletons (97.5% of all births) and to offspring of Danish-born mothers. The cohort was then linked to the Psychiatric Central Register of all inpatient admissions since 1969¹⁶ and to the Causes of Death Register.¹⁷ A national unique identification number enabled virtually complete linkage between registers and between offspring and parent. The cohort consisted of more than 1.38 million subjects eligible for follow-up at age 1 year; almost 1.15 million children (born between January 1, 1973, and December 31, 1993) were eligible at age 5 years and 570 312 (born between January 1, 1973, and December 31, 1982) were eligible at age 16 years.

In the entire birth cohort, unnatural causes represented less than 2% of all deaths in the first year of life compared with almost half of those that occurred at ages 1 to 25 years. Cause-specific deaths during infancy are distinctly different and require special consideration, but for the purposes of this article, offspring age was stratified at 1 to 4 years (young children), 5 to 15 years (older children), and 16 to 25 years (young adults). Age 1 to 4 years was chosen as the youngest stratum because risk of death is high in this group. Age 5 years was used as a cutoff point because Danish children tend to begin preschool education (known as the 0th grade) at this age. In Denmark, legal adulthood begins at age 18 years, as in other countries such as the United States; however, we chose not to use this as a cutoff point because compulsory education ends at age 16 years in Denmark and because risk of death from unnatural causes begins to rise sharply again during midadolescence. Thus, by using 16 years as the cutoff point, most (84%) of the suicides that occurred during follow-up were included in our young adult group. The maximum age of follow-up in this study was restricted to age 25 years by availability of the registry data. Because the older childhood group is broad, we performed tests for interaction to assess whether effects varied within this stratum. The age-specific relative risks given in Table 1 and Table 2 were adjusted using more finely stratified age data, as described in the “Statistical Analyses” section.

Inpatient episodes were coded in the Psychiatric Central Register using the 8th and 10th revisions of the International Classification of Diseases (ICD-8 and ICD-10), with ICD-10 used since 1994. The following parental diagnostic categories were investigated:

• All psychiatric diagnoses: ICD-8: 290 to 315; ICD-10: F00 to F99

• Schizophrenia and related disorders (ie, schizophrenia, schizophrenialike, or schizoaffective disorder): ICD-8: 295, 296.8, 297, 298.39, and 301.83; ICD-10: F20 to F29

• Affective disorders (including bipolar disorder): ICD-8: 296.09, 296.19, 296.29, 296.39, 296.99, 298.09, 298.19, 300.49, and 301.19; ICD-10: F30 to F39

• Alcohol- and other drug-related disorders: ICD-8: 291, 294.30, 294.38, and 303; ICD-10: F10, F11, F16, F18, and F19

We mainly report effects associated with all parental psychiatric diagnoses (Tables 1, 2, and 3) because these analyses had the most statistical power and precision. Evidence of excess risks associated with specific parental diagnostic categories is also mentioned separately where applicable. Categorization of cause of death by ICD coding was as follows:

• All natural causes: ICD-8: all except E800 to E999; ICD-10: all except V01 to Y98

• All unnatural causes: ICD-8: E800 to E999; ICD-10: V01 to Y98

• Accident: ICD-8: E800 to E949; ICD-10: V01 to X59 and Y85 to Y86 (Transport-related [the great majority of these accidents involved motor vehicles rather than other modes of transport]: ICD-8: E800 to E845 and E940 to E941; ICD-10: V01 to V99 and Y85)

• Homicide: ICD-8: E960 to E969; ICD-10: X85 to Y09 and Y87.1

• Suicide: ICD-8: E950 to E959; ICD-10: X60 to X84 and Y87.0 (Poisoning: ICD-8: E950 to E952; ICD-10: X60 to X69)

• Open verdicts: ICD-8: E980 to E989; ICD-10: Y10 to Y34 and Y87.2 (Poisoning: ICD-8: E980 to E982; ICD-10: Y10 to Y19)

Suicide rates were estimated by exclusion of open verdicts, that is, deaths by unnatural causes of undetermined intent. Some commentators have argued for the inclusion of such cases in suicide rate estimates,¹⁸ which has led to divergent practices between countries. To maximize internal validity, we used the more stringent definition according to Danish tradition,¹⁹ with open-verdict deaths analyzed separately.

Analyses were performed using STATA software (Release 8.0; StataCorp LP, College Station, Tex). For each subject, the total person-years at risk was calculated and stratified according to parental first admission dates to create time-dependent exposure variables.^20,21 To eliminate potential reverse causality bias, subjects were only classified as being exposed if first parental psychiatric admission occurred before death (if applicable), because a recent Danish study reported increased risk of parental hospitalization after the death of a child.²²

Relative risk of cause-specific death was modeled using log-linear Poisson regression of aggregated person-years data, without significant evidence of overdispersion.²³ Two separate reference groups were used for estimation of relative risks given in Tables 1 and 2, with the unexposed groups being the complement of the exposed subjects. Thus, to estimate effects associated with maternal admission (Table 1), unexposed subjects were offspring of mothers without previous psychiatric admission, irrespective of paternal admission status. The unexposed group for effects linked to paternal admission was defined likewise (Table 1). Where number of events was sufficient, the interaction between finely stratified offspring age and 5-year period was fitted to comprehensively adjust for cohort effects.²¹

The confidence intervals of the relative risk estimates were not corrected for lack of statistical independence among sibships because of computational problems associated with analyzing multiple time-dependent variables in this large cohort without aggregating the data. Some validation analyses were conducted in more simple models, with a Huber-White variance estimator implemented to relax the assumption of independence.²⁴ The confidence intervals [CIs] were essentially unchanged in these models compared with those without correction for sibship clustering. In further validation models, exact Poisson confidence intervals were calculated for the relative risk of homicide, suicide, and open-verdict deaths, the rarest mortality outcomes. There was no material difference between these intervals and the Wald asymptotic estimates. Thus, for consistency, 95% asymptotic CIs are given throughout this article.

In the entire birth cohort, the lifetime prevalence of parental psychiatric admission increased with age, from 4.7% in young children to 7.5% in older children and to 10.8% in young adults. Among the more than 1.38 million children who were alive and eligible for follow-up at age 1 year, approximately 99.1% survived to their mid 20s according to the age-specific all-cause death rates.⁸ The rate of survival to 26 years was lower in children of psychiatric inpatients (98.7%) compared with those whose parents had not been admitted (99.2%). Overall, there were 5373 all-cause deaths at 1 to 25 years. Cause of death was unknown for less than 2% of these cases, with comparable levels of missing data in exposed and unexposed groups. Paternal identity was missing for 1.2% of all births, 2.3% of all deaths in childhood and early adulthood, 2.6% of deaths from unnatural causes, 3.7% of suicides, and 4.4% of homicides. Maternal identity was known for all births, and there were no missing data for any other items in the registry data.

In general, parental admission was not associated with increased risk of offspring death from natural causes (Table 1). The only exception was a borderline significant (P = .049) association between maternal admission and death from these causes in young adults. However, significant associations between maternal and paternal admission and deaths from unnatural causes were found in each offspring age group, with the highest relative risk (RR) in young children. Interaction tests were conducted to assess whether death from unnatural causes varied by age during the broad older childhood period. In the relation to maternal admission (χ² test, 1 df, 0.1; P = .75) and to paternal admission (χ² test, 1 df, 0.0; P =.91), the effect sizes were similar at ages 5 to 10 and 11 to 15 years.

Most deaths due to unnatural causes in children and young adults were accidental, in both exposed and unexposed offspring (Table 1). Increased risk for this outcome was found in association with either maternal or paternal admission, again with the highest risk observed in young children. In older children, there was evidence of an excess risk of accidental death associated with paternal alcohol- or drug-related disorders (n = 23 deaths; relative risk, 1.81; 95% CI, 1.19-2.74) vs other paternal disorders. This significant excess was confirmed by a Wald z test comparing regression coefficients within the Poisson model (z = 1.99, P = .047). Relative risk of death from other types of accidents was generally higher than death from transport accidents, especially in young children with affected mothers (n = 22; RR, 3.16; 95% CI, 2.04-4.87).

The relative risk of child homicide was markedly elevated in relation to either maternal or paternal psychiatric admission (Table 1). Relative risk of between 5.15 and 9.80 was found, the highest for paternal admission and homicide during early childhood. However, there was no evidence of higher risk of homicide in young adults with affected parents. The Figure shows the convergence in homicide rates between exposed and unexposed groups during early adulthood. Small numbers precluded estimation of relative risk associated with specific parental diagnostic categories, although 2 possible high-risk subgroups were indicated. Among homicides in older children, the mother had been diagnosed with an affective disorder in 6 of 11 cases with previous maternal admission. Also, in all but 1 of 9 homicides in older children with previous paternal admission, the father had been diagnosed with an alcohol- or drug-related disorder.

These outcomes were assessed in young adults only because there were just 3 suicides and 1 open-verdict death among children in the exposed groups. Relative risk of suicide was greater than 2 with maternal and with paternal admission (Table 1). Suicide risk was significantly higher (z = 2.72; P =.006) in offspring with 2 affected parents (n = 8; RR, 5.90; 95% CI, 2.89-12.02) vs 1 affected parent (n = 32; RR, 2.01; 95% CI, 1.37-2.95). The number of suicides was small, but relative risk of 2.17 to 2.73 was indicated in relation to most parental diagnostic groups (ie, maternal and paternal affective disorders, maternal and paternal alcohol- or drug-related disorders, and maternal schizophrenia and related disorders). However, an excess suicide risk above that associated with all other paternal disorders (z = 3.29, P =.001) was indicated in offspring of fathers with schizophrenia and related disorders (n = 7; RR, 8.01; 95% CI, 3.78-17.17). A relative risk of 5.09 was found for the association between previous maternal admission and open-verdict death (Table 1). There was also a suggestion of increased risk associated with paternal admission (Table 1), but this was nonsignificant (P = .07). Because of the sparse event data, effects specific to parental diagnostic categories were not estimated for risk of open-verdict death.

Suicide by poisoning, in particular, may be prone to misclassification as open-verdict death.²⁵ For this reason, and also because of sparse event data, the 2 outcomes were combined and relative risk was estimated separately for poisoning vs other methods. Almost half (45%) of 55 suicides or open-verdict deaths in offspring exposed to admission of either parent were from poisoning, compared with less than a fourth (24%) of 160 deaths in offspring with no previous parental admissions (χ² test, 9.3; 1 df, P = .002). For deaths by poisoning, RR 4.90 and 4.09, respectively, were found in offspring of affected mothers and fathers (Table 2). Smaller effect sizes were observed if these types of death occurred by means other than poisoning.

These results are presented in Table 3. Among deaths from all unnatural causes, the lifetime prevalence of previous parental admission rose from 11.1% in young children to 17.4% in young adults. Consistent with the relative risk estimates, the prevalence was higher in unnatural compared with natural causes of death. The prevalence of parental admission was especially high in homicides in children aged 1 to 4 years and 5 to 15 years and in suicides and open-verdict deaths during early adulthood. However, the prevalence of parental admission in young adult homicides was low compared with that in other causes of death at this age.

To our knowledge, this is the first population-based study of cause-specific deaths in children and young adults whose parents have experienced severe mental disorders, defined here as a previous psychiatric admission. We report greater risk of death from unnatural causes in this group but little evidence of higher numbers of deaths from natural causes. We anticipated that risk of accidental death would be increased in exposed offspring across the study age range. This was confirmed, but the relative risk of death from other unnatural causes was considerably greater.

Markedly high relative risk of homicide in young and older children was found. There was no evidence of increased risk of this outcome in young adults, although high risks of suicide and open-verdict death were observed in this age group. Links between parental illness and suicide risk may be mediated through increased incidence of psychiatric disorders in the offspring.²⁶ The relative risk of suicide or open-verdict deaths from poisoning was higher than the relative risk of such deaths by other means. This could reflect greater use of or access to prescribed medications and other toxic substances in young adults with mentally ill parents. Risk of suicide was especially high in offspring of 2 affected parents and those whose fathers had been admitted with schizophrenia or a related disorder. The latter result aside, there was no evidence of excess risks of death from unnatural causes associated with schizophrenia and related disorders compared with other categories of parental psychopathology. The lifetime prevalence of previous parental admission for deaths from unnatural causes at these ages was high, especially for child homicide and for suicide or open-verdict death in early adulthood.

Our previous investigation of all-cause mortality found no evidence of higher risk in older children with mentally ill mothers and only a small increased risk in children at this age with affected fathers.⁸ These new results, however, indicate that older children were 50% to 60% more likely to die of an unnatural cause than their peers in the general population, with admission for paternal alcohol- or drug-related disorders associated with an excess risk compared with other disorders.

The Danish national registers are complete in their coverage and follow-up. We found a small amount of missing data for cause of death and registered paternal identity, although maternal identity was complete. For practical and ethical reasons, biological paternity cannot be established from national population registers such as those maintained by the Nordic countries. A recent meta-analysis of clinic sample data reported from a number of countries other than Denmark estimated the prevalence of paternal discrepancy to be about 4%.²⁷ Therefore, it is likely that our estimates of effects associated with maternal admission are more robust than those linked to paternal admission. Also, while autopsy rates are known to have fallen in Denmark over recent decades, validity of cause of death coding is likely to be high in this study because multiple causes are rare and medicolegal inquests are routinely performed when deaths occur at young ages.¹⁷

The study has several other general limitations. First, data were sparse for some analyses despite the large size of the national birth cohort, precluding estimation of precise relative risk specific to parental diagnostic categories. Second, although the Central Psychiatric Register records all inpatient activity, selection bias was unavoidable because most mentally ill persons in Denmark, as elsewhere, are not admitted.²⁸ The findings are, therefore, relevant to the families of those whose disorders are sufficiently severe to require admission. Hospitalization incidence rates for schizophrenia reported in previous Danish registry studies (eg, the study by Munk-Jørgensen and Mortensen²⁹) are similar to the median level of incidences reported in a recent meta-analysis of studies conducted mainly in North America and Western Europe.³⁰ The degree of comparability between Denmark and other countries in admission rates for affective disorders and alcohol- or drug-related disorders is unknown. Danish psychiatric hospitalization rates declined during the study period. Interaction terms, however, indicated that the observed effects on offspring mortality risk were not restricted to either the earlier (1973-1984) or later (1985-1998) periods of observation and were not significantly different between these periods. The Central Psychiatric Register began recording all outpatient episodes in 1995; thus, in future years it may be possible to investigate offspring mortality risk according to severity of parental disorder, as indicated by inpatient vs outpatient care. Third, cause of death coding may not be entirely comparable over time because of inconsistencies between the ICD-8 and ICD-10.¹⁷ A smaller degree of disparity seems likely for unnatural causes, which may be less prone to temporal trends in diagnostic practice than natural causes of death. Fourth, in this study, we could not adjust the relative risk for socioeconomic status, which is an important predictor of suicide risk in the general population.³¹ It is, however, doubtful that such large relative risk, particularly for child homicide and self-poisoning in young adults, would be largely explained by confounding factors.³²

Specific sources of information bias may have exaggerated the strong association observed between previous parental admission and risk of child homicide. In cases in which there is a history of parental mental illness, coroners and juries may be more frequently given contextual evidence about adverse family circumstances. This could influence them to give unequivocal rather than open verdicts. Mentally ill parents may also be more likely to confess, either truly or falsely,³³ or be caught. Furthermore, homicide and suicide rates vary markedly between countries,³⁴ and, thus, the generalizability of the findings may be limited.

The high relative risk for child homicide is of particular concern for clinicians and policy makers across a range of health and social care agencies. Maternal filicide is the focus for much of the existing literature in this field.¹³ Thus, our findings are novel in indicating strong effects associated with severe forms of psychopathology in either parent. We report a link between parental mental illness and child homicide but have not identified any causal mechanisms. The currently available data do not enable identification of the perpetrators of these crimes, although negotiation to obtain such information is in progress. Possible explanations include the following: growing up in a more hazardous subculture or neighborhood in which violent death is generally more common³⁵; exposure to other potential perpetrators, such as partners, relatives, friends, or others; and direct harm by parents because of mental disorder or associated behavioral disturbance, or substance abuse. Whatever the causal explanation, a need for greater levels of coordinated support for these vulnerable families is indicated, from mental health, primary care, and social services. Further research is also needed to establish the likely causal pathways.

We report high relative risk for some causes of death, but the absolute risk of death from unnatural causes in these children and young adults was very low. This contrast between high relative risk and low absolute risk was especially marked in relation to child homicide. The rarity of death from unnatural causes at a young age and the low predictive value of parental psychiatric disorder present a major challenge to both practitioners and policy makers. This is widely recognized in developing programs to reduce rates of suicide and homicide in psychiatric populations. Even if effective treatment strategies for affected families were well established, the number needed to treat to save a single life would be huge, as would be the associated costs.³⁶ Nonetheless, mental health professionals should continue to emphasize that the overwhelming majority of parents with a history of psychiatric admission pose no threat to the survival of their children.

Correspondence: Roger T. Webb, MSc, Centre for Women's Mental Health Research, University of Manchester, Seventh Floor, Williamson Building, Oxford Rd, Manchester M13 9PL, England (roger.webb@manchester.ac.uk).

Submitted for Publication: February 23, 2006; final revision received May 19, 2006; accepted June 6, 2006.

Financial Disclosure: None reported.

Funding/Support: This study was supported by grant 073935 from the Wellcome Trust and by the Stanley Medical Research Institute. The National Centre for Register-Based Research, University of Aarhus, is supported financially by the Danish National Research Foundation. Psychiatric epidemiologic research at the Center is also funded in part through a collaborative agreement with the Centre for Basic Psychiatric Research, Aarhus Psychiatric Hospital.

Role of the Sponsors: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; or preparation, review, or approval of the manuscript.

Previous Presentation: A summary of these findings was presented at the 159th Annual Meeting of the American Psychiatric Association; May 24, 2006; Toronto, Ont.

Acknowledgment: We thank Carsten B. Pedersen, MSc, Thomas M. Laursen, MSc, and Heine Gøtzsche, National Centre for Register-Based Research, for linking the registers and providing statistical advice.