Community-based specialist palliative care is associated with reduced hospital costs for people with non-cancer conditions during the last year of life

This was a retrospective cohort study of the last year of life of all people who died in WA from 1 January 2009 to 31 December 2010 with a cause of death potentially amenable to receiving palliative care; namely cancers, heart failure, renal failure, liver failure, chronic obstructive pulmonary disease, Alzheimer’s disease, motor neurone disease, Parkinson’s disease, Huntington’s disease and HIV/AIDS. Approvals to conduct this study were granted by Human Research Ethics Committees at the WA Department of Health (#2012/76, 01/02/2013) and Curtin University (53/2012, 06/06/2012).

A de-identified linked extraction of death records, hospital records and community-based care records of all decedents spanning the last 2 years of life was obtained from the Data Linkage Branch at the WA Department of Health. Further details on the selection of this cohort and health service use in the last year of life have been reported previously[16, 17]. Briefly, it consisted of 12, 817 people who died during the 2 year study period in WA and who had mention on Part I of their death certificate of one or more of the ten disease conditions considered amenable for palliative care as defined by Rosenwax et al.[18]. When a decedent had more than one of the ten conditions of interest on Part 1 of their death certificate, then the most antecedent condition was assigned as the principal condition.

The community-based specialist palliative care in this study was provided by Silver Chain WA which supplied over 90% of referred community-based specialist palliative care in WA, although this was mostly restricted to major metropolitan areas [19]. WA is the largest state of Australia (2.5 million km²) with a population of 2.6 million that is concentrated in the capital city, Perth, and the south-west corner of the state. A multidisciplinary team of palliative care clinicians and nurses, allied health professionals and volunteers provide home nursing care, counselling, respite options, practical support and links to other services with the aim of enabling people with a life limiting illness to remain at home. A palliative nurse consultancy service was available to residential care facilities where client care is managed around the clock by registered nurses. This service included advice, assessment, staff education and telephone follow-up to meet the needs of specific clients. A palliative care rural telephone advisory service provided specialist palliative care advice to local rural service providers 24 h per day. Client dates of enrolment and disenrollment were used to define periods of time receiving community-based specialist palliative care. It was possible for decedents to have multiple periods of palliative care enrolment if their symptoms were of a relapsing and remitting nature. Clients were discharged from the service after 1 month if they were not available for regular review.

Each day of the last year of life for every decedent was assigned a day number ranging from 1 (365 days before death) to 365 (day of death). A hospital cost was applied to each day of the last year of life for every decedent with days not spent in hospital considered to have incurred zero costs. From these day-specific hospital costs, two cost-based outcome measures were defined. The first was the mean day-specific hospital cost averaged over all decedents whether they were in hospital or not. We refer to this mean day-specific hospital cost as being a cohort averaged (CA) hospital cost. This type of cost estimate is a reflection of both the decision to go to hospital and the actual costs once admitted. The second outcome measure was a mean day-specific hospital cost averaged over only decedents who were in hospital on that particular day. These costs are a reflection of hospital costs once admitted and these are referred to as inpatient averaged (IA) hospital costs.

The Australian Refined-Diagnostic Related Groups (AR-DRGs) system was used to estimate day-specific acute care hospital costs [20]. Component costs for acute episodes of care for each AR-DRG over the study period were obtained from National Hospital Cost Data Collection (NHCDC) reports [21, 22]. Fixed hospital costs included costs of operating room procedures, specialist procedure suites, critical care, pharmacy, pathology, imaging, allied health professionals and emergency department. Fixed costs were considered to have mostly been incurred during the initial period of hospital stays [23]. A daily fixed cost was estimated by fitting the fixed costs to a decreasing exponential distribution where the shape parameter was a function of the actual or national average length of stay such that 99.9% of fixed costs were distributed by the end of the hospital stay. Variable costs included staff salaries and on-costs, hotel costs, supplies and depreciation and were considered to have been incurred at a steady rate over the actual length of stay. Daily fixed and variable costs were summed to generate a day-specific hospital cost per decedent per day. For subacute care hospital stays, the WA average daily cost estimated from the NHCDC Round 15 (2010-2011) report and was assigned as $844 for rehabilitation, $1051 for palliative care, $815 for geriatric evaluation and management, $1162 for psychogeriatric care and $1012 for maintenance [24]. All costs are in Australian dollars ($A) and were adjusted for health inflation and standardised to the 2010-2011 financial year. On 31st December 2011, $A1 was equivalent to $US1 and Euro €0.79. The costs of hospital emergency department presentations were not included in this study.

Marital status was classified as partnered or not/unknown. Decedent’s residential locations were used to assign accessibility categories based on the Australian ARIA+ index that takes into account road distance measurements to the nearest service centres and population size [25]. Type of residence was classified as a private residence, residential aged care facility (RACF) or other. Changes in a decedents’ marital status, place of residence, accessibility to services and community-based specialist palliative care status over the last year of life were recorded as time-varying covariates.

Comordity was estimated from hospital inpatient medical records over the last 2 years of life by summing the number of the 31 medical conditions identified by Elixhauser [26] based on algorithms created by Quan [27]. The underlying cause of death was excluded from the calculation of the number of comorbid conditions. Comorbidity for decedents without hospital admissions in the last 2 years of life was estimated from the death certificate.

Data were structured so that each decedent had 365 observations representing their last year of life. Each day was assigned as being a day of a hospital stay and/or community-based specialist palliative care or not. Dates were permitted to overlap, for example, a day coded as a hospital stay could also be a community-based specialist palliative care day if it fell between palliative care enrolment and disenrollment service dates. Decedents with incomplete demographic information were excluded from further analysis. Locally weighted scatterplot smoothing was used to produce unadjusted curves of the mean daily hospital costs over the last year of life.

Cohort averaged daily hospital costs were modelled as a mixed discrete-continuous random variable with a mass point at zero. A two part model was fitted with the first part as a binary choice model of being in hospital or not (zero versus non-zero hospital costs) on any 1 day and the second part was conditional on being in hospital that day and used a generalised linear model with a reciprocal link and gamma distribution. The combined prediction was computed as the product of the probability of a hospital admission and the expected value of hospital costs conditional upon a positive cost.

Inpatient averaged mean daily hospital costs were modelled using generalised linear models with inverse Gaussian distribution and inverse quadratic link. Minimisation of the deviance, Akaike and Bayesian information criteria and the modified Park test were used to identify the best fitting distribution and link. Fractional polynomial transformations of variables were used when required. Additional explanatory power of interaction terms was assessed by the difference in model deviance values. Undue influence of individual observations on the model fit were tested and the model fit assessed by plotting Anscombe residuals against log variance. Standard errors were calculated using the clustered sandwich estimator to account for intragroup correlation and the Sidak correction was applied when estimating multiple marginal means. All analysis was performed using Stata 14 (College Station, TX).

The crude CA hospital cost averaged of the last year of life was A$108 (SD 414) per decedent per day. However, this varied with closeness to time of death and ranged from A$42 (SD 256) 1 year before death to A$602 (SD 714) per decedent per day at time of death. The greatest differences in crude CA hospital costs by community-based specialist palliative care status were observed close to the time of death (Fig. 1Ai). Decedents with liver failure had the highest unadjusted CA hospital costs close to the time of death while those with Alzheimer’s disease and Parkinson’s disease had the lowest (Fig. 1Aii). Higher crude CA hospital costs were also observed in decedents who were younger, male, partnered, had private health insurance, lived in a private residence and had multiple comorbidity (Table 1).

After adjusting for age, comorbidity and other potentially confounding variables in a two part main effects model, community-based specialist palliative care was associated with an average 27% reduction in CA hospital costs over the last year of life. This equated to a A$30 (95%CI 26-34) reduction per decedent per day relative to periods of time not receiving community-based specialist palliative care (Table 1). Decedents dying from heart failure, Alzheimer’s disease and Parkinson’s disease had significantly lower average CA hospital costs relative to decedents with cancer. Adjusted CA hospital costs for decedents with liver failure were A$14 per decedent per day lower than costs for decedents with cancer; a reversal of the earlier crude cost findings. Reduced CA hospital costs were also noted with increasing age at death, having a partner, not having private health insurance, living in a residential aged care facility and having less comorbidity.

However, a significant three-way interaction was observed between closeness to time of death, cause of death and community-based specialist palliative care which means that a single cost value cannot explain the associations adequately. Averaged over all causes of death, the association of community-based specialist palliative care with reduced CA hospital costs became evident at around 120 days before death (Fig. 2a). When stratified by cause of death, decedents with cancers showed this reduction earlier, at around 140 days before death (Fig. 2b). For decedents with heart failure, renal failure, chronic obstructive pulmonary disease and Parkinson’s disease, significantly reduced CA hospital costs associated with community-based specialist palliative care became apparent around 45 days before death (Fig. 2c to e, h). Decedents with liver failure had reduced CA hospital costs in the fourth to second last months of life with the hospital costs converging around death (Fig. 2f). Decedents with motor neurone disease (MND) showed little difference in CA hospital costs with or without community-based specialist palliative care (Fig. 2g).

The crude IA hospital cost was A$1114 per hospitalised decedent per day, although this varied with closeness to death and underlying cause of death (Fig. 1Bi and Bii). Being enrolled in community-based specialist palliative care at the time of hospitalisation consistently showed a lower crude IA hospital costs over the last year of life. The crude IA hospital costs stratified by the underlying cause of death showed a more variable pattern. IA hospital costs associated with cancer and Alzheimer’s disease were consistent over the last year of life at around A$1150 and A$1100 per hospitalised decedent/per day respectively regardless of closeness to time of death. In contrast, the three types of organ failure, Parkinson’s disease and COPD all showed increasing IA hospital costs with closeness to time of death. Higher crude IA hospital costs were also observed in decedents who were younger, male, partnered, had private health insurance, lived in a private residence and had multiple comorbidity (Table 2).

After adjusting for other decedent factors associated with IA hospital costs and averaged over the year, IA hospital costs for hospitalised decedents who were receiving community-based specialist palliative care at the time of admission was A$1030 per decedent per day over the last year of life. This was a reduction of A$96 (95%CI 84- 107) per hospitalised decedent per day, equivalent to a 9% (95%CI 7-10) reduction relative to decedents hospitalised during periods of time not receiving community-based specialist palliative care (Table 2). Reduced IA hospital costs were also noted for decedents who were older, female and with less comorbidity.

However, further modelling that included significant interaction terms showed that the reduction in IA hospital costs associated with community-based specialist palliative care varied significantly by the underlying cause of death and closeness to day of death and is best represented graphically (Fig. 3). Periods of time enrolled in community-based specialist palliative care were clearly associated with reduced IA hospital costs for decedents with cancers, heart failure, chronic obstructive pulmonary disease and Alzheimer’s disease (Fig. 3a–e). For decedents with renal failure, Parkinson’s disease and liver failure, the evidence of reduction in IA hospital costs with community-based specialist palliative care was only evident for parts of the last year of life (Fig. 3c–h). Decedents with motor neurone disease showed no significant difference in IA hospital costs with community-based specialist palliative care at any time over the last year of life (Fig. 3g).