Context. End-of-life care is an important yet underdeveloped component of pediatric hospital services.
Objectives. We sought 1) to describe the demographics of children who die in children’s hospitals, 2) to describe the prevalence of complex chronic conditions (CCCs) among these cases, and 3) to test the hypotheses that cases with a greater number of CCC diagnoses experience longer periods both of mechanical ventilation and of hospitalization before death.
Design and Methods. We identified all deaths of patients 0 to 24 years old that occurred in the 60 hospitals contributing discharge data to the National Association of Children’s Hospitals and Related Institutions data consortium for the years 1991, 1994, and 1997. We classified discharge diagnoses into 9 major categories of CCCs (cardiovascular, neuromuscular, malignancy, respiratory, renal, metabolic, gastrointestinal, hematologic/immunologic, and other congenital/genetic).
Results. Of the 13 761 deaths identified, 42% had been admitted between 0 and 28 days of life, 18% between 1 and 12 months, 25% between 1 and 9 years, and 15% between 10 and 24 years. Fifty-three percent were white, 20% were black, and 9% were Hispanic. The principal payer was listed as a governmental source for 42% and a private insurance company for 35%. Based on all the discharge diagnoses recorded for each case, 40% had no CCC diagnosis, 44% had diagnoses representing 1 major CCC category, 13% had diagnoses representing 2 CCC categories, and 4% had diagnoses representing 3 or more CCC categories. Among cases that had no CCC diagnoses, the principal diagnoses were related to prematurity and newborn disorders for 32% of these cases, injuries and poisoning for 26%, and an assortment of acute and infectious processes for the remaining 42%. Mechanical ventilation was provided to 66% of neonates, 40% of infants, 36% of children, and 36% of adolescents. Cases with CCCs were more likely than non-CCC cases to have been mechanically ventilated (52% vs 46%), and to have been ventilated longer (mean: 11.7 days for CCC cases vs 4.8 days for non-CCC cases). The median duration of hospitalization was 4 days, while the mean was 16.4 days. After adjustment for age, sex, year, and principal payer, compared with patients with no CCC diagnoses, those with 1 major CCC category had a significantly lower hazard of dying soon after admission (hazard ratio [HR]: 0.60; 95% confidence interval [CI]: 0.57–0.62), those with 2 CCC categories even lower (HR: 0.53; 95% CI: 0.50–0.57), and those with 3 or more CCC categories the lowest hazard of rapid death (HR: 0.51; 95% CI: 0.46–0.57). This trend of diminishing hazard of rapid death was significant across the 3 groups of children with 1 or more CCCs.
Conclusions. Children’s hospitals care for a substantial number of dying patients, who differ widely by age and medical conditions. Children who die in the hospital with CCCs are more likely to experience longer periods of mechanical ventilation and hospitalization before death.
The past decade has seen a surge of interest in how our society and its medical system care for dying patients.1–5 The pediatric health care community more specifically has been seeking to improve palliative care services for dying children.6,7 These services ideally would attend to the needs of these children and their families throughout the dying process, from symptom management and quality-of-life enhancement care during earlier phases when the timing of death is most uncertain, to intensive peri-death care immediately around the time of death, through to bereavement care for the family and health care staff subsequently.6–15
Where should such services be located? In Washington State from 1980 to 1998, the majority of children and young adults under 25 years old who died did so in the hospital (52%), with an even larger proportion of deaths attributable to complex chronic conditions (CCCs) occurring in the hospital (75%).16 If such proportions are found in other settings, then palliative care services need to be at least partially hospital-based, integrated with community-based services in a manner to cover the inpatient and outpatient realms of care with the greatest degree of continuity possible and to facilitate, if desired, the occurrence of death at home. Together these services would strive to meet the needs of dying children who either are never admitted to a hospital, are cared for in a hospital for some interval but die at home, or die in a hospital.
To develop hospital-based palliative care services across the United States, a more comprehensive understanding of what kinds of children need what kinds of services would be helpful. To date, studies have examined the experience of children dying with particular conditions (such as cancer17 and cystic fibrosis18); children dying in particular locations (including specific neonatal or pediatric intensive care units,19–22 a network of 9 intensive care units in France, 23 and throughout single hospitals24–26); and children dying under the care of a hospice or palliative care team.27,28 These studies, while providing important insights, have limited generalizability. With the exception of the French study, none have examined the experience of children across multiple systems of health care. Furthermore, many of these studies predetermined which types of patients might be suitable for palliative care, using a variety of eligibility criteria that make comparisons across studies difficult.
To further our understanding of the circumstances of death for children who died in a hospital, we aimed to describe the demographics of children who die in children’s hospitals and to describe the prevalence of CCCs among these cases. Additionally, given how the duration of mechanical ventilation and of hospitalization influence the dying experience for child and for family, and our clinical experience that children with multiple concurrent medical conditions are treated very intensively, we also sought to test the hypotheses that cases with more CCC diagnoses experience both longer periods of mechanical ventilation and longer periods of hospitalization before death. For this study, we used administrative discharge data collected from 60 children’s hospitals located throughout the United States for the years 1991, 1994, and 1997.
Case Definition and Data Source
We identified all deaths in patients 0 to 24 years old that occurred in the 60 hospitals contributing discharge data to the National Association of Children’s Hospitals and Related Institutions data consortium for the years 1991, 1994, and 1997. We selected these years because 1991 was the first year of available data, 1997 the most recent year, and 1994 provided a mid-point year to increase the power of the study to detect change over time. We used as our upper age limit not 18 but rather 24 years old so as not to exclude deaths attributable to conditions that often fatally culminate in early adulthood, such as cystic fibrosis or muscular dystrophy. Each hospital contributed data for 1, 2, or 3 of the study years, resulting in a total of 136 hospital-year combinations.
We classified International Classification of Diseases, Ninth Revision, Clinical Modification (ICD9-CM) discharge diagnoses into 9 major categories of CCCs (cardiovascular, neuromuscular, malignancy, respiratory, renal, metabolic, gastrointestinal, hematologic or immunologic, and other congenital or genetic) as previously described.29,30 A review of the diagnostic codes categorized as representing a CCC led to a slight broadening of the range of ICD9-CM codes designating CCCs for congenital vascular malformations, gastrointestinal anomalies, and hereditary anemias, and to the exclusion of patent ductus arteriosus as a CCC.
Annual Number of Deaths and Average Daily Census
For each hospital we summed the number of deaths that occurred in each study year. This total number of annual deaths for each of the 136 hospital-year combinations was then divided by quartiles into 4 categories, <96, 96–117, 118–150, and >150 deaths per year. To derive the average daily census of children in each hospital who subsequently died during that admission, we used the length of stay as a count of the number of days that each case spent in the hospital, summed the total number of days spent in the hospital by such children for each hospital-year combination, then divided by 365 days a year; that is, with n = total number of subsequently fatal cases in each hospital each year. This calculation assumes that the number of cases that are admitted in 1 year, remain in hospital, and die the next year is constant 1 year to the next, and that the number of days each of these cases contributes is also constant over time. The distribution of this parameter across all the hospital-year combinations is reported in Table 2.
We examined the distribution of CCC status among cases by age, year, and ventilatory status, using the Pearson χ2 test to test for the presence of associations. Then, according to our a priori hypotheses, we tested whether CCC status was associated with the duration of mechanical ventilation and the duration of hospitalization before death, using Cox proportional hazards regression modeling to adjust for confounders and using robust variance estimates to account for the clustering of data by hospital sites. We also calculated the mean duration of both mechanical ventilation and the hospitalization along with 95% confidence intervals (CI). We performed all statistical analysis with Stata 7.0 software (Stata Corp, College Station, TX).31
Of the 13 892 deaths identified, 56% were male; 42% had been admitted between 0 and 28 days of life, 18% between 1 and 12 months, 24% between 1and 9 years, 15% between 10 and 24 years; and 1% over the age of 24 years; 53% were white, 20% black, and 9% Hispanic; 42% had governmental insurance listed as principal payer, 35% had private insurance, and 23% had other sources (Table 1). In subsequent analyses, those cases over 24 years old at the time of admission were omitted.
Average Daily Census
The daily census of subsequently fatal cases at a hospital is a function both of the number of fatal cases (which, in turn, is related to the volume of patients cared for at the hospital and their age distribution) as well as the length of stay for these subsequently fatal cases. This daily census ranged broadly, from a median of 3.5 for hospitals reporting 50 to 99 deaths a year, 5.0 for hospitals reporting 100 to 149 annual deaths, and 7.0 for hospitals reporting more than 150 annual deaths (Table 2).
Based on all the discharge diagnoses recorded for each case, 40% had no CCC diagnosis, 44% had diagnoses representing 1 major CCC category, 13% had diagnoses representing 2 categories, and 4% had diagnoses representing 3 or more CCC categories (Table 3). The most prevalent CCC categories varied by age, with cardiovascular, respiratory, and other congenital or genetic conditions most common in the newborn period, while by adolescence, malignancies and neuromuscular conditions predominated (Table 4). Among cases that had no CCC diagnoses, the principal diagnoses were related to prematurity and newborn disorders for 32% of these cases, injuries and poisoning for 26%, and an assortment of acute and infectious processes for the remaining 42%.
Mechanical ventilation was provided at some point during the hospitalization to 66% of neonatal cases, 40% of infant cases, 36% of child cases, and 36% of adolescent cases. Cases with CCCs were more likely than cases without CCCs to have been mechanically ventilated (52% vs 46%; P < .001). The proportion of patients with specific CCCs who were mechanically ventilated was greatest for those with renal conditions (68%) and least for those with malignancies (36%), with the proportion varying by age for each CCC category (Table 5). Although this data set did not identify whether ventilation was discontinued because of death, withdrawal of care, or some other reason, the patient records did specify how many days patients were mechanically ventilated during some period of the terminal admission. If ventilated, cases with CCCs were ventilated longer (mean: 11.7 days [95% CI 10.8–12.6] vs 4.8 days [95% CI 4.3–5.3]), a difference that was most pronounced among infants and least noticeable among adolescents and young adults (Fig 1). Adjusting for gender, year, principal payer, and the annual number of deaths in each hospital, cases with CCCs were significantly more likely to have been ventilated longer, with approximately 40% reduction in their hazard of having mechanical ventilation stopped (Table 6). With adjustment for the same factors, children with respiratory CCCs were likely to experience the longest period of mechanical ventilation (hazard ratio [HR]: 0.49; 95% CI: 0.44–0.55) while those with renal CCCs the shortest period (HR: 1.33; 95% CI: 1.11–1.59).
Length of Terminal Hospitalization
The overall median duration of terminal hospitalization was 4 days, while the mean was 16.3 days. Again, differences between cases with CCCs (mean: 21.4 days; 95% CI: 20.6–22.2) and without CCCs (mean: 8.6 days; 95% CI 8.0–9.2) were seen, in this instance among all age groups (Fig 2). After adjustment for age, sex, year, and payer, compared with patients with no CCC diagnoses, those with 1 CCC category had a significantly diminished chance (hazard) of dying on any given day of the hospital stay (HR: 0.60; 95% CI: 0.57–0.62), those with 2 CCC categories even lower (HR: 0.53; 95% CI: 0.50–0.57), and those with 3 or more categories the lowest hazard of rapid death (HR: 0.51; 95% CI: 0.46–0.56) (Table 7). The decline of the HR across these 3 levels of CCC categories was a significant trend (P < .001). Again, no association was observed between length of stay and either year or the annual number of children who died in each hospital. With adjustment for the same factors, respiratory CCCs were again associated with the longest terminal admissions (HR: 0.52; 95% CI: 0.48–0.56) and renal conditions with the shortest admissions (HR: 1.04; 95% CI: 0.92–1.18).
Annual Number of Deaths
We sought evidence of a volume-outcome relationship between patient case load and measures of clinical care, but found no significant association between the annual number of children who died at each hospital site and either the duration of mechanical ventilation (Table 6) or of the entire hospitalization (Table 7).
This retrospective study of administrative discharge data has found that National Association of Children’s Hospitals and Related Institutions-member children’s hospitals in the United States care for a substantial number of dying patients, who differ widely by age and medical conditions. Furthermore, we found evidence to support the hypotheses that children who die with CCCs experience both longer periods of mechanical ventilation and of hospitalization before death.
Although we can not determine from this study specific reasons for the association between CCC status and processes of care, several possible mechanisms—broadly grouped into pathophysiologic, diagnostic, and prognostic explanations—had motivated our hypotheses initially. Pathophysiologically, the underlying biological processes that cause death may unfold more slowly in children with CCCs than children without CCCs. Diagnostically, decisions to admit children with CCCs to the hospital may be based on different admission criteria than other children, so that children admitted with CCCs for what becomes a terminal admission may not be as ill at the time of admission as other children who die in the hospital, and consequently their hospital course is longer. If so, the longer duration of hospitalization or mechanical ventilation is attributable to differences in the underlying condition or how admission decisions are made rather than to differences in hospital treatment. Prognostically, foreseeing what is likely to happen to gravely ill children with CCCs may be more difficult than for those dying children without CCCs, and this uncertainty leads to prolonged courses of intensive therapy before care is limited or withdrawn.32 Additionally—in what we have come to think of as the Persephone syndrome— children with CCCs who die in hospital may have had several other nearly-fatal episodes, leading family and care providers to be more persistent in pursuing aggressive therapy for prolonged periods of time in hope of yet another astonishing recovery. Just as those who loved the goddess Persephone waited steadfastly for her annual return each spring, so too these caregivers may become more resolute in their commitment to life-prolonging therapy by dint of previous experience. All these conceivable mechanisms warrant further investigation.
Of note, we did not find an association between the volume of fatal cases cared for in each hospital and the duration either of mechanical ventilation or of the entire hospitalization. One possible explanation is that as the medical staff within a high-volume institution gain experience with terminally ill children, they may improve their outcomes both in terms of longer survival and improved processes to compassionately and ethically limit or withdraw care, with little net effect on length of mechanical ventilation or hospitalization. More detailed studies are required to see whether case volume does indeed have mixed effects on the experience of dying children.
This study was limited in several aspects. First, we used hospital discharge data collected through standard operating administrative practices. The data collection process likely differed across sites and introduced some degree of error into the data, but probably in a haphazard manner, thereby introducing no systematic bias into our estimates of proportions, but potentially obscuring significant associations. Second, our examination was limited to only children’s hospitals. We suspect—without any specific evidence beyond personal experience—that non-children’s hospitals may have a very different pattern, with fewer deaths and a greater proportion of these deaths occurring in neonates shortly after birth or among adolescents in the emergency department after trauma. Finally, we examined only 3 years of data, the most recent of which is now nearly 5 years old. For both of these last 2 caveats, we acknowledge that extrapolating beyond the range of the data are always suspect. Any inferences about patterns outside of the children’s hospitals included in this sample, or regarding current or future trends, are unwarranted.
These limitations notwithstanding, we believe that this study offers several implications for the development of palliative care programs. First, to the extent that patterns seen in this study do persist, palliative care programs based at children’s hospitals will be responsible for patients ranging from infancy to adolescence who die anywhere from hours to months after being admitted. Second, in children’s hospitals many of the older children who die will have cancer, but most children who die in the hospital will suffer from other disorders, so that palliative care services will need to accommodate a wide range of symptoms and coordinate with many subspeciality services. Third, on any given day, most of the children’s hospitals in this sample typically had between 2.2 to 10.6 patients who would subsequently die within their walls. Although identifying these children in a timely, prospective manner will pose substantial challenges, this method of estimating an average daily census could provide guidance as to the staffing required to meet the needs of just those patients who die in the hospital (and not necessarily the needs of patients who are discharged and die at home or who are never admitted). Finally, given the high proportion of all fatal cases that were mechanically ventilated, issues of limiting and withdrawing care will perforce be one of the challenges that palliative care teams will confront.
Dr Feudtner was supported by a Special Projects award from the Ambulatory Pediatric Association, and by grant K08 HS00002 from the Agency for Healthcare Research and Quality.
We thank Robin Adams and Cindy Larison for their assistance with data preparation, and Lyn Bassett for preparation of the manuscript.
- ↵Nuland SB. How We Die: Reflections On Life’s Final Chapter. New York, NY: Alfred A. Knopf;1994
- ↵American Academy of Pediatrics, Committee on Bioethics and Committee on Hospital Care. Palliative care for children. Pediatrics.2000;106 :351– 357
- ↵Goldman A. Recent advances in palliative care. Importance of palliative care for children is being increasingly recognised. BMJ.2001;322 :234
- Corr CA, Corr DM. Pediatric hospice care. Pediatrics.1985;76 :774– 780
- Goldman A. ABC of palliative care. Special problems of children. BMJ.1998;316 :49– 52
- Armstrong-Dailey A, Goltzer SZ, eds. Hospice Care for Children. New York, NY: Oxford;1993
- Fleischman AR, Nolan K, Dubler NN, et al. Caring for gravely ill children. Pediatrics.1994;94 :433– 439
- Goldman A, ed. Care of the Dying Child. New York, NY: Oxford;1994
- ↵Feudtner C, Silveira MJ, Christakis DA. Where do children with complex chronic conditions die? Patterns in Washington State, 1980–1998. Pediatrics.2002;109 :656– 660
- ↵Robinson WM, Ravilly S, Berde C, Wohl ME. End-of-life care in cystic fibrosis. Pediatrics.1997;100 :205– 209
- Goh AY, Lum LC, Chan PW, Bakar F, Chong BO. Withdrawal and limitation of life support in paediatric intensive care. Arch Dis Child.1999;80 :424– 428
- ↵Stein A, Forrest GC, Woolley H, Baum JD. Life threatening illness and hospice care [see comments]. Arch Dis Child.1989;64 :697– 702
- ↵Goldman A, Beardsmore S, Hunt J. Palliative care for children with cancer—home, hospital, or hospice? Arch Dis Child.1990;65 :641– 643
- ↵Feudtner C, Christakis DA, Connell FA. Pediatric deaths attributable to complex chronic conditions: a population-based study of Washington State, 1980–1997. Pediatrics.2000;106(suppl) :205– 209
- ↵Feudtner C, Hays RM, Haynes G, Geyer JR, Neff JM, Koepsell TD. Deaths attributed to pediatric complex chronic conditions: national trends and implications for supportive care services. Pediatrics.2001;107(6) . Available at: http://www.pediatrics.org/cgi/content/full/107/6/e99
- ↵Stata Corporation. Stata Statistical Software: Release 7.0. College Station, TX: Stata Corporation;2000
- ↵Christakis NA. Death Foretold: Prophecy and Prognosis in Medical Care. Chicago, IL: University of Chicago;1999
- Copyright © 2002 by the American Academy of Pediatrics