Washington state death certificate

In the realm of end-of-life care for adults, much emphasis has been placed on where death occurs. During the past few decades, several population-based studies have shown that--despite most patients' expressing a preference to die at home (1,2)--the majority of deaths occur in a hospital or other institution. (3-8) These studies have documented that age, race, marital status, enrollment in hospice, the exact nature of the condition that caused death, and the pace at which the dying process occurred all potentially affect the site of death. Even after accounting for these individual-level attributes, however, geographical variation still exists in the proportion of deaths that occur in hospitals, which may be attributable to local health system characteristics or some other unmeasured factor. (9)

Far less is known about the locations where children die or what attributes influence the site of death. Given that a rising proportion of childhood deaths are attributable to complex chronic conditions (CCCs) (10,11) and that increasingly sophisticated pediatric medical care is being delivered in the ambulatory setting or at home, a trend toward a greater proportion of deaths occurring at home is a reasonable hypothesis that, if borne out, would have significant policy implications regarding adequate provision of community-based supportive care services.

To address these questions regarding the site of pediatric deaths, we conducted a retrospective case series of all deaths among children aged 0 days to 24 years in Washington state from 1980 to 1998. We specifically sought to test whether an increasing proportion of CCC-associated deaths were occurring at home.

METHODS

Classification of CCC

We defined CCCs as any medical condition that can be reasonably expected to last at least 12 months (unless death intervenes) and to involve either several different organ systems or 1 organ system severely enough to require specialty pediatric care and probably some period of hospitalization in a tertiary care center. On the basis of this definition and several published lists of International Classification of Diseases, Ninth Revision (ICD-9) codes from earlier studies that examined health service utilization for children with congenital or complicated medical conditions, we mapped specific ICD-9 codes to 9 CCC diagnostic categories (10,11): neuromuscular, cardiovascular, respiratory, renal, gastrointestinal, hematologic or immunologic, metabolic, malignancy, and genetic or other congenital defect conditions.

Data Sources and Case Selection and Classification

We used death certificate information for all deaths that occurred in Washington state from 1980 to 1998, which are contained on vital statistics computer files maintained by the Washington State Department of Health. Because several life-limiting pediatric conditions (eg, cystic fibrosis, Duchenne's muscular dystrophy) typically do not cause death until early adulthood, we selected all cases for whom death had occurred before the 25th birthday. All of the death certificates recorded sex, date of birth, date of death, and county of residence, as well as a single ICD-9 code specifying the "underlying cause of death." From 1988 to 1998, additional causes of death were specified on the death certificates, with as many as 11 additional diagnoses cited. Examining only the years 1988 to 1998, we determined sensitivity of the single "underlying cause of death" code to identify cases that had any CCC code in any of the diagnosis fields to be 88.8%. Given this high level of sensitivity, we included data from all years (1980-1998) in our analysis, using the "underlying cause of death" code as the basis for all classification of CCC status.

The death certificate files contained limited data regarding socioeconomic status of individual cases. To determine whether a higher level of socioeconomic status was associated with a greater likelihood that the death of a child with a CCC occurred at home, we turned to United States census data. Although such aggregated data encompasses many individuals, studies of health services utilization and outcomes have shown that census data (even aggregated at the level of zip codes) does usefully augment individual-level data when measures of socioeconomic status are otherwise missing. (12,13) We therefore used the 1990 Census Summary Tape File 3 to obtain the 1989 median household income by zip code, merging this data to the individual-level data from the death certificates based on the zip code recorded for each case, with a 99.8% successful match.

Statistical Analyses

After examining the proportion of deaths that occurred in all sites, we tested for change in the proportion of children who died at home over time using logistic regression, regressing the occurrence of home death on the year of death and examining the significance of the coefficient for year of death, which was modeled as a continuous variable. We then constructed a more complete logistic regression model that included adjustment covariates of year of death, male gender, nonwhite race, age (modeled as a linear spline with knots at 1 and 6 months and 1, 5, 10, 15, and 19 years of age), (14) quintiles of 1989 median household income in the zip code of residence, and indicator variables for each county of residence. Within this model, the predictor of interest--namely, the 9 CCC categories--was compared with leukemia-related cases. We used leukemia-related cases as the benchmark because palliative care for children has chiefly been considered with regard to children with fatal cancer, leukemia-related cases provide a relatively homogeneous group for comparisons, and many physicians at some stage in their training or career have had experience caring for patients in the terminal stages of leukemia. We also examined interaction between age and CCC categories, depicting these interactions graphically. Then using this logistic regression model, we determined whether the addition of county of residence added significantly to the model by using likelihood ratio testing. Assessment of this complete model yielded a Homer-Lemeshow goodness of fit P value of 0.78, with an area under the receiver operating characteristic curve of 0.80 (indicating excellent discrimination). Finally, we calculated the adjusted odds ratio (AOR) of home death for cases over 1 year of age for each county for geographical presentation. For all statistical tests, we deemed P < .05 as significant and calculated 95% confidence intervals (CI) for estimates of the AORs. Analyses were performed using Stata 7.0 (Stata Corp, College Station, TX).

RESULTS

From 1980 through 1998, a total of 31 455 infants, children, and young adults younger than 25 years died in the state of Washington. Of these, 64.8% were males, 80.0% were white, and 39.4% were younger than 1 year. Although the majority of deaths occurred in the hospital (52%), with 17.1% occurring at home, these proportions varied substantially depending on the cause of death (Table 1).

Throughout this period, as seen in Fig 1, infants who died as a result of a CCC did so predominantly in the hospital, ranging from 88.4% to 96.6% with a slight but significant downward trend (P < .01). Beyond the first birthday, however, CCC-related deaths increasingly occurred at home, rising from 21% in 1980 to 43% by 1998 (P < .01 for trend).

[FIGURE 1 OMITTED]

Focusing on only those deaths caused by a CCC, the adjusted likelihood of death occurring at home has increased by approximately 6% per year (95% CI, 3%-9%) and also increased as the local residential area in which the child had resided rose in terms of affluence (comparing the highest to the lowest quintile of median household income: AOR: 1.58; 95% CI: 1.06-2.34; Table 2). Neither the child's gender (male: AOR: 1.11; 95% CI: 0.92-1.34) nor race (nonwhite: AOR: 0.82; 95% CI: 0.63-1.06) were significantly associated with death occurring at home. The underlying cause of the child's death was strongly associated with the occurrence of death at home. Using leukemia-related deaths as a benchmark, deaths as a result of congenital, genetic, neuromuscular and metabolic conditions, and other forms of cancer all were more likely to have occurred at home (P < .001), whereas deaths as a result of respiratory conditions seem to have been less likely to have occurred at home (P = .015). Age, too, influenced the adjusted probability of death occurring at home, but this effect varied depending on the underlying cause of death (as illustrated in Fig 2 for the 3 CCCs that accounted for the most deaths; variation existed as well within the other CCC types).

[FIGURE 2 OMITTED]