Study Design

This was a retrospective cohort study of persons aged 20 years and older who died from medical conditions considered amenable to palliative care. A linked deidentified extraction of all hospital admission records, death records, hospital ED visits, and community-based care records was obtained from the Data Linkage Branch, Western Australian Department of Health. The ED presentations from 365 days before the date of death until death (ie, the last year of life), as well as place of residence at each hospital admission and at death, were available for each decedent. Ethical approval was obtained from Curtin University and the Western Australian Department of Health Human Research Ethics Committees.

The final cohort was selected from death registration records. People who died with conditions amenable to palliative care in Western Australia from January 1, 2009, to December 31, 2010, and who were aged 20 years or older were identified. This was achieved by searching part 1 of the death certificate (underlying causes of death) for any mention of the most frequent conditions amenable to palliative care identified by Rosenwax et al, namely, neoplasia, heart failure, renal failure, liver failure, chronic obstructive pulmonary disease, Alzheimer’s disease, motor neuron disease, Parkinson’s disease, Huntington’s disease, and HIV/AIDS. The underlying antecedent cause of death took precedence over more direct causes of death when more than one condition amenable to palliative care was present. Because of small sample sizes, decedents with motor neuron disease, Parkinson’s disease, Huntington’s disease, and HIV/AIDS were excluded. Decedents with Alzheimer’s dementia were also excluded because we have published a detailed analysis of ED use and palliative care in this group elsewhere.

Community-based palliative care was defined as the palliative care provided by the nonprofit organization Silver Chain Western Australia, which provides more than 90% of in-home palliative care in Western Australia. Their community-based palliative care service area is largely based in the major metropolitan city of Perth, where approximately 75% of the state population resides. The community-based palliative care service is 24-hour in-home specialist nursing and, where required, medical service provided by primary care physicians and palliative care specialists. Care includes at-home physical care and practical support, symptom management (for example, pain and nausea), counseling, a respite option, and links to other community, spiritual, and government services. Silver Chain also provide a palliative nurse consultancy service to public and private hospitals and residential care facilities where patient care is managed by a registered nurse 24 hours a day. A palliative rural telephone advisory service offers specialist advice and knowledge to rural service providers in regard to the palliative care needs of specific patients.

Access to community-based palliative care requires referral from a medical practitioner or hospital for patients with active, progressive, terminal illness requiring symptom management. This generally occurs within 3 months of death but can be 12 months or more from death. It has been reported that patients with life-limiting conditions with slow or fluctuating trajectories that required longer-term palliative care may have more difficulty accessing it.

An ED visit was defined as the first presentation of the day to an ED whether it resulted in a hospital admission or not. Multiple presentations to the ED on the same day were considered a single ED visit. The ED visits that occurred as part of hospital transfers were excluded from the analysis. Available ED data included the triage category, hospital admission status, and presenting symptom. Coded presenting symptom information was available for major metropolitan public hospitals only (approximately 80% of ED visits). The number of previous ED visits that had already occurred in the last year of life at each ED visit was recorded.

Residential address was classified as private residence, residential aged care facility, other types of care facilities, or other/unknown. Marital status at death was classified as partnered or nonpartnered. Accessibility categories were based on the Accessibility/Remoteness Index of Australia (ARIA+) index, which measures accessibility to services by taking road distances to nearest service centers and population size into account. Socioeconomic status was estimated from the Index of Relative Social Disadvantage, which estimates the average social disadvantage of small geographic areas according to 17 measures, including education, income, occupation, and unemployment. Comorbidity was defined as the presence of 1 or more of the 30 Elixhauser conditions recorded during inpatient hospital stays after exclusion of the principal condition leading to death.

Primary Data Analysis

χ ² Tests were used to assess the equality of proportions. Linear regression models were used to assess equality of means across principal conditions, using neoplasia as the reference groups. Nonparametric K-sample tests for the equality of medians were used.

For the time-to-event (survival) analyses, a marginal-models approach was used and multiple ED visits per decedent were accounted for with robust variance estimators. Each day in the last year of life of each decedent was assigned as no care (no hospital- or community-based palliative care), inpatient hospital care, or community-based palliative care.

A decedent was not considered at risk of an ED admission while in the hospital, and his or her data were removed from the risk pool during each hospital stay (interval truncation). The Nelson-Aalen estimator of the cumulative hazard function was used to visualize the accumulated total average number of ED visits in the last year of life. Because decedents could switch between receiving palliative care and then not receiving it, the Figure does not necessarily represent the same group of decedents on any one day.

A , The percentage of the decedent cohort who were hospital inpatients, visited an ED, or received community-based palliative care each day during the last year of life. The percentage in the hospital on any one day increased up to 60% in the week before death and has been shown truncated here at 45 days before death to better fit the graph. B , The average cumulative number of ED visits during the last year of life if decedents had continuously received or not received community-based palliative care during that time. Estimated with the Nelson-Aalen cumulative hazard with the data structured that allowed decedents to switch between the care states according to actual service dates. Days spent in the hospital were removed from the risk period.

To easily visualize the baseline hazards, a flexible parametric proportional hazards model that used restricted cubic splines (Royston-Parmar models) was constructed. Time-fixed (eg, sex) and time-varying covariates (eg, ED history and community-based palliative care state) variables were included. Time-dependent effects, ie, when the hazard ratio of a covariate changed over time (nonproportional hazards), were modeled as cubic splines. Interaction terms were included in the final model if the joint test of the 2 interaction terms or the test of an interesting partial interaction effect resulted in P <.05. Baseline complexity was described with 7 internal knots for the main model and 2 for the time-dependent effects. Stata (version 13; StataCorp, College Station, TX) was used.

Study Design

This was a retrospective cohort study of persons aged 20 years and older who died from medical conditions considered amenable to palliative care. A linked deidentified extraction of all hospital admission records, death records, hospital ED visits, and community-based care records was obtained from the Data Linkage Branch, Western Australian Department of Health. The ED presentations from 365 days before the date of death until death (ie, the last year of life), as well as place of residence at each hospital admission and at death, were available for each decedent. Ethical approval was obtained from Curtin University and the Western Australian Department of Health Human Research Ethics Committees.

The final cohort was selected from death registration records. People who died with conditions amenable to palliative care in Western Australia from January 1, 2009, to December 31, 2010, and who were aged 20 years or older were identified. This was achieved by searching part 1 of the death certificate (underlying causes of death) for any mention of the most frequent conditions amenable to palliative care identified by Rosenwax et al, namely, neoplasia, heart failure, renal failure, liver failure, chronic obstructive pulmonary disease, Alzheimer’s disease, motor neuron disease, Parkinson’s disease, Huntington’s disease, and HIV/AIDS. The underlying antecedent cause of death took precedence over more direct causes of death when more than one condition amenable to palliative care was present. Because of small sample sizes, decedents with motor neuron disease, Parkinson’s disease, Huntington’s disease, and HIV/AIDS were excluded. Decedents with Alzheimer’s dementia were also excluded because we have published a detailed analysis of ED use and palliative care in this group elsewhere.

Community-based palliative care was defined as the palliative care provided by the nonprofit organization Silver Chain Western Australia, which provides more than 90% of in-home palliative care in Western Australia. Their community-based palliative care service area is largely based in the major metropolitan city of Perth, where approximately 75% of the state population resides. The community-based palliative care service is 24-hour in-home specialist nursing and, where required, medical service provided by primary care physicians and palliative care specialists. Care includes at-home physical care and practical support, symptom management (for example, pain and nausea), counseling, a respite option, and links to other community, spiritual, and government services. Silver Chain also provide a palliative nurse consultancy service to public and private hospitals and residential care facilities where patient care is managed by a registered nurse 24 hours a day. A palliative rural telephone advisory service offers specialist advice and knowledge to rural service providers in regard to the palliative care needs of specific patients.

Access to community-based palliative care requires referral from a medical practitioner or hospital for patients with active, progressive, terminal illness requiring symptom management. This generally occurs within 3 months of death but can be 12 months or more from death. It has been reported that patients with life-limiting conditions with slow or fluctuating trajectories that required longer-term palliative care may have more difficulty accessing it.

An ED visit was defined as the first presentation of the day to an ED whether it resulted in a hospital admission or not. Multiple presentations to the ED on the same day were considered a single ED visit. The ED visits that occurred as part of hospital transfers were excluded from the analysis. Available ED data included the triage category, hospital admission status, and presenting symptom. Coded presenting symptom information was available for major metropolitan public hospitals only (approximately 80% of ED visits). The number of previous ED visits that had already occurred in the last year of life at each ED visit was recorded.

Residential address was classified as private residence, residential aged care facility, other types of care facilities, or other/unknown. Marital status at death was classified as partnered or nonpartnered. Accessibility categories were based on the Accessibility/Remoteness Index of Australia (ARIA+) index, which measures accessibility to services by taking road distances to nearest service centers and population size into account. Socioeconomic status was estimated from the Index of Relative Social Disadvantage, which estimates the average social disadvantage of small geographic areas according to 17 measures, including education, income, occupation, and unemployment. Comorbidity was defined as the presence of 1 or more of the 30 Elixhauser conditions recorded during inpatient hospital stays after exclusion of the principal condition leading to death.

Primary Data Analysis

χ ² Tests were used to assess the equality of proportions. Linear regression models were used to assess equality of means across principal conditions, using neoplasia as the reference groups. Nonparametric K-sample tests for the equality of medians were used.

For the time-to-event (survival) analyses, a marginal-models approach was used and multiple ED visits per decedent were accounted for with robust variance estimators. Each day in the last year of life of each decedent was assigned as no care (no hospital- or community-based palliative care), inpatient hospital care, or community-based palliative care.

A decedent was not considered at risk of an ED admission while in the hospital, and his or her data were removed from the risk pool during each hospital stay (interval truncation). The Nelson-Aalen estimator of the cumulative hazard function was used to visualize the accumulated total average number of ED visits in the last year of life. Because decedents could switch between receiving palliative care and then not receiving it, the Figure does not necessarily represent the same group of decedents on any one day.

A , The percentage of the decedent cohort who were hospital inpatients, visited an ED, or received community-based palliative care each day during the last year of life. The percentage in the hospital on any one day increased up to 60% in the week before death and has been shown truncated here at 45 days before death to better fit the graph. B , The average cumulative number of ED visits during the last year of life if decedents had continuously received or not received community-based palliative care during that time. Estimated with the Nelson-Aalen cumulative hazard with the data structured that allowed decedents to switch between the care states according to actual service dates. Days spent in the hospital were removed from the risk period.

To easily visualize the baseline hazards, a flexible parametric proportional hazards model that used restricted cubic splines (Royston-Parmar models) was constructed. Time-fixed (eg, sex) and time-varying covariates (eg, ED history and community-based palliative care state) variables were included. Time-dependent effects, ie, when the hazard ratio of a covariate changed over time (nonproportional hazards), were modeled as cubic splines. Interaction terms were included in the final model if the joint test of the 2 interaction terms or the test of an interesting partial interaction effect resulted in P <.05. Baseline complexity was described with 7 internal knots for the main model and 2 for the time-dependent effects. Stata (version 13; StataCorp, College Station, TX) was used.