Article Text

Download PDFPDF

Characteristics influencing location of death for children with life-limiting illness
  1. Emily Chang1,
  2. Rod MacLeod2,
  3. Ross Drake1
  1. 1Starship Palliative Care Service, Starship Children's Health, Auckland City Hospital, Auckland, New Zealand
  2. 2Northern Clinical School, Greenwich Hospital, The University of Sydney, NSW 2006, Australia
  1. Correspondence to Dr Emily Chang, Starship Palliative Care Service, Starship Children's Health, Auckland City Hospital, Private Bag, 92024, Auckland 1142, New Zealand; emschang{at}


Objective To determine whether demographic and diagnostic characteristics were associated with location of death in a series of children with life-limiting illnesses.

Design A population-level case series was carried out by reviewing mortality records. Sociodemographic characteristics, diagnosis and referral to paediatric palliative care (PPC) were analysed for association with location of death.

Setting New Zealand

Participants Children and young people aged 28 days–18 years who died from a life-limiting illness between 2006 and 2009 inclusive.

Main outcome measures Location of death—home, hospital, other.

Results Of 494 deaths, 53.6% (256/494) died in hospital and 41.9% (203/494) died at home. Asian (OR=2.66, 95% CI 1.17 to 6.04) and Pacific children (OR=2.22, 95% CI 1.15 to 4.29) had an increased risk of death in hospital compared with European children, while children with cancer (adjusted OR=0.48, 95% CI 0.3 to 0.75) and children referred to the PPC service (adjusted OR=0.60, 95% CI 0.38 to 0.96) had a decreased risk. Population-attributable risk for referral to the PPC service was 28.2% (95% CI 11.25 to 47.75).

Conclusions Most children in New Zealand with a life-limiting illness die in hospital with a significant influence resulting from ethnic background, diagnosis and referral to the PPC service. These findings have implications for resourcing PPC services and end-of-life care.

  • Palliative Care
  • Paediatric Practice

Statistics from

What is already known on this topic

  • Most children with life-limiting illnesses die in hospital.

  • Infants are more likely than older children to die in hospital.

  • Children with non-malignant disease are more likely to die in hospital than children with cancer.

What this study adds

  • Children referred to the paediatric palliative care service are more likely to die in the community than those who were not referred.

  • Asian and Pacific children are more likely to die in hospital than European children.


Despite advances in medical technology, a substantial number of children and young people still die from life-limiting illnesses. In New Zealand, an estimated 2.3 per 10 000 children and young people died from a life-limiting illness in 2006.1 In many, death could be anticipated, giving children, their families and healthcare professionals the opportunity of end-of-life care in advance, including location of death (LOD). Studies from the UK and Australia have indicated that children and their families prefer to receive this care at home.2–4

In reality, most children with a life-limiting illness die in hospital.1 ,5–7 The development of paediatric palliative care (PPC) services can allow for end-of-life care to be provided at home and, in some countries, a children's hospice. New Zealand is a country in the South Pacific with a population of 4 027 947. Approximately 67% belong to the European ethnic group. The other major ethnic groups include Maori (indigenous peoples) (14%), Asian peoples (9.2%) and Pacific peoples (6.2%).8 The health system is mostly publically funded. There is no children's hospice, but in 1999, a PPC service was established at the Starship Children's Hospital in Auckland. This small multidisciplinary team provides care across the hospital to home continuum in greater Auckland, and advice to other healthcare services nationally.

LOD for children with life-limiting illnesses has been well described in several countries.6 ,7 ,9 However, the factors influencing outcome are less well understood. Research to date indicates that age, gender, ethnic group, socioeconomic status (SES) and diagnosis may influence LOD.10 ,11 Anecdotally, disease progression, family and healthcare resources, and cultural beliefs seem to influence where a child dies and insights into these characteristics may assist end-of-life care planning, including preferred LOD.

The aim of this study was to determine any demographic and diagnostic factors associated with LOD in a series of children who had died from a life-limiting illness. The possible influence of referral to palliative care was of primary interest, as it is most amenable to change. It was also hypothesised that age, ethnicity, SES and diagnosis may influence LOD.


A retrospective review of mortality records between 1 January 2006 and 31 December 2009 was conducted on children in New Zealand aged between 28 days and 18 years who died from a life-limiting illness and for whom death could have been anticipated. The records were from a database kept by the Child and Youth Mortality Review Committee for all individuals aged 28 days–24 years who died in New Zealand.

Children who died from a life-limiting illness were identified by matching cause of death to an International Classification of Diseases 10th revision (ICD-10) code chosen to reflect life-limiting illnesses for which death could have been anticipated and a palliative approach may have been beneficial (table 1). Box 1 outlines the exclusion criteria. However, cases from these groups could be included if there was a reasonable expectation that palliative care may have been beneficial, such as those who sustained significant brain injury. Children who died from surgical complications were excluded if the surgery had occurred within 30 days of death and the complication was listed as contributory. Similarly, children who died from complications of curative treatment were excluded if the treatment was given within 14 days of death and the listed cause was a known complication of treatment.

Box 1

Exclusion criteria

  • Sudden death in an apparently well person

  • Diagnoses not known to benefit from palliative care

  • Cause of death not related to underlying disease

  • Death due to surgical complications

  • Death due to complications of curative treatment

  • Not enough information

Table 1

ICD-10 codes used to identify children who died from life-limiting illnesses

Relevant variables were age, LOD, socioeconomic deprivation, ethnic group, diagnostic category and referral to PPC. Age was defined as the time between date of birth and date of death. LOD was categorised as home, hospital, hospice or other, based on addresses provided in each mortality record. Children who died as hospice inpatients were coded to have died in a ‘hospice’ as, occasionally, a child's care can be shared with an adult hospice. Children who died at a residential address differing from their main address were deemed ‘other’.

Socioeconomic deprivation was scored using the NZDep2006 index.12 This system accounts for nine dimensions of deprivation, including income, home ownership, support, access to a car and telephone, level of education and employment. Residential areas of New Zealand are divided into groups of approximately 87 people and a score allocated as a decile, with decile 1 being the least deprived and decile 10 being the most deprived. For each case, an NZDep2006 score was derived from their main place of residence. Due to the small number of cases in our analysis, deciles were amalgamated into quintiles with deciles 1 and 2 forming quintile 1, and so on.

Ethnic group was derived from data provided on the child's death certificate and followed the categories present in the 2006 New Zealand census.8

A diagnostic category was assigned to each case based on the underlying cause of death. In some instances, identified medical problems rather than the cause of death determined the category that was allocated. For example, the authors coded a case as ‘cardiac’ rather than ‘chromosomal/syndrome’ when significant cardiac lesions were identified, but the cause of death was recorded as Trisomy 21.

Cases who were referred to the PPC service were identified by matching a unique identification number (National Health Index number) to that held on the service's database.

Statistical methods

Our analysis considered the dependent variable as death in hospital, with independent variables comprising sociodemographic and diagnostic information. Analysis of data was undertaken using the JMP V.5.1, StatsDirect V.2.5.7  and R V. For categorical variables, such as age group, gender, ethnic group, differences between groups were investigated using Fisher's exact test, and effect estimates summarised using univariable ORs. We assessed the independent effects of sociodemographic and diagnosis variables on death in hospital using logistic regression. All independent variables were thought, a priori, to be important influences on the outcome variable, so all were included in the logistic model. We used both Wald and likelihood-ratio tests to assess whether independent variables were significantly related to the risk of death occurring in hospital. p Values < 0.05 were considered statistically significant. We tested for interactions between ethnic group and SES, and between SES and palliative care referral using the likelihood ratio test.

We further explored the effect of involvement with PPC on the outcome variable as children engaged with the PPC service were likely to differ systematically from those who were not. Propensity scoring (closest unit matching) was used to check the validity of the multivariable OR for the effect of this exposure and matched-pair effects were adjusted for in the logistic model of propensity-matched subjects.14


During the study period, a total of 1797 deaths were recorded in the age range, with 577 children fulfilling selection criteria. The children who did not fulfil the ICD-10 selection criteria died predominantly from unintentional injury (43%), sudden unexpected death between age 28 days and 2 years (20%) and acute medical illnesses for which no underlying chronic disease was recorded (19%).

Of the 577 children selected by ICD-10 codes, 494 children remained after exclusion criteria were applied (figure 1). The two cases lacking the necessary information to make a decision were pending a coroner's review and excluded as death was unlikely to have been anticipated. Five patients were excluded for deprivation analysis as their usual residence was outside of New Zealand.

Figure 1

Simple selection diagram. ICD-10, International Classification of Diseases 10th revision.

The majority of the 494 children either died in hospital (53.6%) or at home (41.9%). The remaining children died at a residential address other than their home or an adult hospice (table 2).

Table 2

Location of death

Paediatric palliative care referral

Twenty-one percent of the study population were referred to the PPC service. Factors associated with referral to PPC included a diagnosis of malignancy, age category, death outside of hospital and ethnic group (table 3). Children with malignancy had over twice the rate of referral compared with children with non-malignant disease (34.5% vs 16.0%, p<0.0001). Children aged 1–9 years were more likely to be referred compared with infants and young people aged 10–18 years. Of the total, Maori were less likely to be referred to palliative care, whereas Pacific children were more likely to be referred.

Table 3

Paediatric palliative care referral by demographic and diagnostic characteristics

Overall, children under PPC had a 40% decreased odds of dying in hospital (adjusted OR=0.60, 95% CI 0.38 to 0.96, p=0.03). The propensity matched analysis (n=210), for referral to PPC, yielded an adjusted OR similar to the unmatched analysis (propensity score matched adjusted OR=0.68; 95% CI 0.43 to 1.07), further strengthening the causal claims for this exposure. Population-attributable risk for children referred to the PPC service estimated an overall reduction in hospital deaths of 28.2% (95% CI 11.25 to 47.75). This implies that if all children in the study population had been referred for PPC a 28% reduction in risk of hospital death would be expected if the association measured in this study was indeed causal.

Demographic characteristics

Univariable analysis showed that children under 1 year of age (OR=1.98, 95% CI 1.34 to 2.95) were associated with a higher risk of dying in hospital (table 4) compared with all other age groups. However, the measure of association was attenuated in the multivariable analysis and was not statistically significant (adjusted OR=1.46, 95% CI 0.95 to 2.26) (table 5).

Table 4

Location of death by demographic characteristics

Table 5

Crude and adjusted ORs for demographic and diagnostic variables and death in hospital

European children made up the largest ethnic group and had the lowest risk of death in hospital (table 4). In comparison to the other groupings, children of Asian (OR=2.66, 95% CI 1.17 to 6.04, p=0.019) or Pacific ethnic group had a higher risk of dying in hospital (over twice the odds) even after adjustment for relevant confounding variables. Maori children had a similar prevalence of dying in hospital as the reference group.

Neither interaction (ethnic group and SES, or SES and palliative care referral) was statistically significant.

Diagnostic category

Children with malignant, cardiac and neurologic disease formed the three largest groups, with distribution differing according to age. Children aged less than 1 year died mostly from non-malignant disease: respiratory (24.3%), cardiac (23.7%) and neurological (18.4%). In the 15–18 years age group, the most common diagnoses were malignancy (43%) and neurological disease (29%).

Children with cancer were significantly less likely to die in hospital (adjusted OR=0.48, 95% CI 0.3 to 0.75, p<0.0001) compared with children with non-malignant disease.


This is the first study, to our knowledge, to demonstrate that those who were referred to PPC had a lower risk of hospital death, and that Pacific and Asian children were more likely to die in hospital than European children.


Using mortality records to determine whether a child could have benefited from a palliative approach was limited by lack of clinical data, so it was possible that children could have been assigned to the wrong category. Although information bias can be minimised by using ICD-10 criteria for sample selection, there is potential for bias when applying exclusion criteria. A child may have died unexpectedly while receiving palliative care and therefore been excluded. While unlikely, up to 25 children who were receiving palliative care may have been inadvertently excluded. It is difficult to quantify the effect of this on the LOD.

Paediatric palliative care referral

The effect of PPC referral on LOD has been relatively poorly studied. In the adult population, involvement of palliative care services was associated with improved odds of dying at home.15 ,16 In this study, the prevalence of hospital death in children receiving palliative care was comparable with a multicentre study of a similar population in Australia, Canada and the UK where the LOD was evenly distributed between hospital, home and hospice.9 In another Canadian study, the rate of hospital death was substantially higher at 58%.17

An interesting aspect of referral to the New Zealand PPC service was the reduced risk of hospital death, even after propensity score matching. Calculation of the population-attributable risk indicated that a further lowering of hospital deaths by about 28% was likely if all the study population were referred to the PPC service. This finding has implications for resource allocation and service delivery for PPC in New Zealand and other developed Western countries.


Variation in LOD between age groups has been reported in English5 and US6 studies, similar to this one. In these reports, about 80% of infants died in hospital, and this age group spent more time in hospital during the end-of-life period compared with older children.18 Infants (aged 28 days–1 year) in this study were at higher risk of hospital death after univariable analysis, but not after the multivariable results. One explanation for this finding is that age group is independent of hospital death. However, this conclusion is not supported by studies from other countries,5 ,18 ,19 and thus may represent a type II error (false negative due to small sample size).

Ethnic group

In other published studies, the chance of children with life-limiting conditions dying in hospital ranged from 52% to 74%,7 ,10 with ethnic group a likely influence.10 A retrospective analysis of children who died from complex, chronic conditions in the USA demonstrated black and Hispanic decedents were less likely to die at home,10 while a smaller study of adult patients with cancer in Texas indicated black ethnic group was associated with a significantly higher risk of death in an acute hospital.11

This study illustrated the same vulnerability in New Zealand for children of Asian and Pacific ethnicity. The reasons are likely to be complex, with immigrants’ attitudes towards healthcare speculated to be a contributing factor, reflecting their society of origin. Over 80% of Asian and 40% of Pacific peoples in New Zealand are immigrants.8 Studies of paediatric oncology in Asian countries suggest differing practices compared with Western settings. In Taiwan, one study reported that 78.8% of children with cancer died in hospital, with over half admitted to intensive care in the last month of life, and 24% receiving cardiopulmonary resuscitation (CPR).20 These rates are high when compared with the USA. For example, in a study from Ohio,21 10% of children who died from cancer were admitted to intensive care, and in a similar report from Tennessee, 7.6% received CPR.22

Our findings of ethnic differences in LOD may be explained, partly, by socioeconomic differences between the ethnic groups. Possible correlations between ethnic group and SES were explored using Spearman's rank correlation. Compared with all other ethnic groups, European ethnic group was moderately positively correlated with higher SES (0.40), whereas Pacific and Maori ethnic groups were negatively correlated with SES (−0.24 and −0.29 respectively). Middle Eastern, Latin American, African (MELAA) and Asian groups were independent of SES, compared with all others.

These patterns of correlation are unlikely to explain the pattern in adjusted measures of association between ethnic group and hospital death; for example, Asian peoples were at highest risk of hospital death compared with European, but had little correlation with SES compared with other ethnic groups.

Diagnostic factors

The decreased risk of hospital death for children with cancer (compared with all other diagnoses) was consistent with other studies.1 ,9 ,18 This is likely to result from the illness trajectory being more predictable, giving healthcare professionals and families an opportunity to plan for palliative and end-of-life care. Furthermore, in New Zealand, support for paediatric oncology patients in the community is more extensive than for children with non-malignant disease. For example, oncology outreach services are available throughout the country and charities such as the Child Cancer Foundation and CanTeen are able to provide psychosocial, practical and financial support.23 ,24 In contrast, charities supporting people with non-malignant diagnoses do exist but generally do not offer a similar level of support.


This study has identified protective and risk factors for death in hospital among children and young people with life-limiting illnesses. Children with cancer were less likely to die in hospital as were children referred to the PPC team. Conversely, children with an Asian or Pacific background had higher rates of hospital death.

These factors will help inform the resourcing of paediatric palliative and end-of-life services. The findings provide guidance for delivering tailored and equitable palliative care for children with life-limiting illnesses and their families.


We are grateful to Drs Nick Baker, Gabrielle McDonald and the Child and Youth Mortality Review Committee for access to their database, guidance in study design and implementation. The authors thank the Starship Children's Research Centre and Dr Simon Thornley for statistical assistance.



  • Contributors All authors contributed to study conception and design. EC and RD contributed to the data analysis. The draft manuscript was drawn up by EC with critical review from RD and RM. All three authors approved the final version to be submitted for publication. EC contributed to all stages, and can take responsibility for this study. There are no other people who meet the criteria for authorship.

  • Funding Emily Chang was supported by a Starship Foundation Fellowship.

  • Competing interests None.

  • Ethics approval This study was approved by the New Zealand Health and Disability Ethics Committee.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.