SPC use was identified for 17% of the sample overall (3697 of 21,714), with higher use among the cancer cohort (30.0% or 3160 of 10,535) relative to the non-cancer cohort (4.8% or 537 of 11,179). The most prevalent underlying causes of death among the non-cancer cohort were heart failure (47.2%), stroke (15.8%) and dementia (14.9%) (Table 1). Among the cancer cohort, the most frequent cancer types included lung cancer (15.8%), haematological cancers (13.3%) and prostate cancer (10.6%). Relative to the non-cancer cohort, the cancer cohort was younger and had fewer health problems (in terms of comorbidity scores and days in hospital) prior to the final year of life (Table 1). A higher proportion of the non-cancer cohort were in residential aged care during the final year of life, while a higher proportion of the cancer cohort died in hospital (Table 1). A higher proportion of the cancer cohort used hospital inpatient services in the last year of life; 5% of the cancer cohort and 19% of the non-cancer cohort had no inpatient episodes in the final year.

The duration of SPC use differed between the cancer and non-cancer cohorts, where Cat3 (used for between 30 days and 6 months) formed the largest group of SPC users in the cancer cohort while Cat1 (used for less than 7 days) was the largest group in the non-cancer cohort. Among those in the cancer cohort using SPC, 15% fall into Cat1 (used for less than 7 days), 29% into Cat2 (between 7 and 30 days), 44% into Cat3 (between 30 days and 6 months) and 12% into Cat4 (between 6 months and 1 year). Among those in the non-cancer cohort using SPC, 36% fall into Cat1, 26% into Cat2, 26% into Cat3 and 12% into Cat4.

Overall, the cancer cohort incurred more costs than the non-cancer cohort over the final year of life; mean costs were AU$55,357 (SD 45,059) for the cancer cohort and AU$35,318 (SD 41,948) for the non-cancer cohort. Among the cancer cohort, the mean total costs for those using SPC at any time in the last year were AU$64,982 (SD 46,354) and AU$51,205 (SD 43,845) for those not using SPC. Among those dying from a non-cancer illness, the same mean costs were AU$59,975 (SD 47,852) for those using SPC and AU$34,074 (SD 41,245) for those not using SPC. The mean total costs of health care over the last year of life were highest for those with longer SPC use and lowest for those with no SPC use (Fig. 1) for both the cancer and non-cancer cohorts. The differences between the groups with different duration of SPC use diminish when we examine the shorter time periods to death (Fig. 1). Without risk adjustment, the group with longer SPC use (Cat4) had the lowest costs over the last month of life expressed as differences from the costs of those with no SPC use (Fig. 2). Although the costs for all groups increased toward the end of life, the mean total cost in the last month of life as a proportion of the mean cost over the last year of life was lowest for those with longer SPC use. These proportions (calculated from the raw mean costs over the last 1 and 12 months for each SPC category) were: Cat1 0.34, Cat2 0.27, Cat3 0.21, Cat4 0.17 and Cat5 0.28 for the cancer cohort and Cat1 0.36, Cat2 0.34, Cat3 0.19, Cat4 0.13 and Cat5 0.29 for the non-cancer cohort.

Mean total costs over the last 1, 3, 6, 9 and 12 months of life by time from first specialist palliative care (SPC) contact: raw mean costs for the cancer and non-cancer cohorts

Unadjusted mean total cost differences relative to the no specialist palliative care (SPC) group over the last 1, 3, 6, 9 and 12 months of life by time from first SPC contact: estimates for the cancer and non-cancer cohorts

Over the last year of life, the mean cost differences between each of the SPC groups and those not using SPC were substantially reduced by risk adjustment, particularly for the non-cancer cohort. However, relative to no SPC use (Cat5), they remained significantly higher for SPC use categories Cat2, Cat3 and Cat4 for the cancer cohort and for categories Cat1, Cat3 and Cat4 for the non-cancer cohort (Fig. 3). The differences were AU$6051 (95% CI 3047–9055), AU$11,351 (95% CI 8673–14,029) and AU$15,590 (95% CI 10,617–20,562) for categories, Cat2, Cat3 and Cat4, respectively, among the cancer cohort and AU$9029 (95% CI 3017–15,042), AU$16,456 (95% CI 8615–24,296) and AU$13,739 (95% CI 733–26,746) for categories Cat1, Cat3 and Cat4, respectively, among the non-cancer cohort. This changes for the costs over shorter time periods to death. After risk adjustment, the mean costs for those using SPC for more than 6 months were significantly lower than for those not using SPC over the last 1 and 3 months of life (Fig. 3). For the cancer group, the differences were − AU$2810 (95% CI − 3945 to − 1676) and − AU$2845 (95% CI − 5025 to − 665) over the last 1 and 3 months, respectively, and for the non-cancer group the same differences were − AU$4345 (95% CI − 6625 to − 2066) and − AU$6223 (95% CI − 11,016 to − 1431). Those using SPC for 1–6 months had significantly lower costs relative to the no SPC group over the last month of life with differences of − AU$2395 (95% CI − 3196 to − 1594) and − AU$2778 (95% CI − 4769 to − 787) for the cancer and non-cancer cohorts, respectively (Fig. 3). The relevant models are shown in the supplementary material (Online Resource Tables S1 and S2). Similar differences were found in the robustness checks reported in the supplementary material (Online Resource Tables S4 and S5). The results for the non-cancer cohort were not sensitive to the exclusion of those with cancer as a contributing cause of death or those receiving cancer treatment in the last year of life (Online Resource Table S3).

Risk-adjusted mean total cost differences relative to the no specialist palliative care (SPC) group over the last 1, 3, 6, 9 and 12 months of life by time from first SPC contact: regression estimates for the cancer and non-cancer cohorts. Models were adjusted for age, sex, residential aged care, previous year hospital days, comorbidity index, private insurance, household income, acute care episode if died in hospital, palliative care inpatient episode, decile of socioeconomic advantage, local health district of residence, year of death and chemotherapy or radiation therapy in the last year. Cancer model was adjusted for the type of cancer, and the non-cancer model was adjusted for the underlying cause of death

Inpatient hospital costs comprised the major portion of total costs, particularly over the shorter time periods to death. Over the last year of life, mean hospital costs were 69% of mean total costs for the cancer cohort and 77% for the non-cancer cohort. Over the last month of life, mean hospital costs were 89% of mean total costs for the cancer cohort and 90% for the non-cancer cohort. After risk adjustment, the differences in mean hospital costs showed a similar pattern to the mean total costs with Cat3 and Cat4 having the highest cost over the last year of life and the lowest cost over the last month of life (Fig. 4). Among the non-cancer cohort, lower hospital costs were identified over the last 6 months of life for those with longer SPC use (Cat4) relative to those with no SPC use (difference − AU$6157, 95% CI − 11,677 to − 637). Private medical services (Medicare) and pharmaceuticals (PBS) were considerably less costly than hospital services, which formed the major portion of total costs. The risk-adjusted cost differences showed a similar pattern to total costs for the cancer cohort but not for the non-cancer cohort where only the Cat3 group showed significantly lower Medicare costs than the no SPC group over the last month (Figs. 5 and 6). Among the cancer cohort, lower Medicare costs were identified over the last 6 months of life for those with longer SPC use (Cat4) relative to those with no SPC use (difference − AU$745, 95% CI − 1297 to − 194). The models are provided in the online supplementary material; see Online Resource Tables S6 and S7 for hospital costs, Tables S8 and S9 for Medicare costs and Tables S10 and S11 for PBS costs. Non-admitted emergency department attendances were less than 1% of total costs and are not shown separately.

Risk-adjusted mean hospital cost differences relative to the no specialist palliative care (SPC) group over the last 1, 3, 6, 9 and 12 months of life by time from first SPC contact: regression estimates for cancer and non-cancer cohorts. Models were adjusted for age, sex, residential aged care, previous year hospital days, comorbidity index, private insurance, household income, acute care episode if died in hospital, palliative care inpatient episode, decile of socioeconomic advantage, local health district of residence, year of death and chemotherapy or radiation therapy in the last year. Cancer model was adjusted for the type of cancer, and the non-cancer model was adjusted for the underlying cause of death

Risk-adjusted mean cost differences for private medical services relative to the no specialist palliative care (SPC) group over the last 1, 3, 6, 9 and 12 months of life by time from first SPC contact: regression estimates for cancer and non-cancer cohorts. Models were adjusted for age, sex, residential aged care, previous year hospital days, comorbidity index, private insurance, household income, acute care episode if died in hospital, palliative care inpatient episode, decile of socioeconomic advantage, local health district of residence, year of death and chemotherapy or radiation therapy in the last year. Cancer model was adjusted for the type of cancer, and the non-cancer model was adjusted for the underlying cause of death

Risk-adjusted mean cost differences for pharmaceuticals relative to the no specialist palliative care (SPC) group over the last 1, 3, 6, 9 and 12 months of life by time from first SPC contact: regression estimates for cancer and non-cancer cohorts. Models were adjusted for age, sex, residential aged care, previous year hospital days, comorbidity index, private insurance, household income, acute care episode if died in hospital, palliative care inpatient episode, decile of socioeconomic advantage, local health district of residence, year of death and chemotherapy or radiation therapy in the last year. Cancer model was adjusted for the type of cancer, and the non-cancer model was adjusted for the underlying cause of death

The use and duration of SPC was associated with a number of decedent socio-demographic and health characteristics. In both cohorts, residents in a non-metropolitan area had a lower probability of using SPC relative to those in the metropolitan area; this was the case for SPC of any duration in the cancer cohort and for Cat1 and Cat3 in the non-cancer cohort (Tables 2 and 3). In the cancer cohort, those with private health insurance and those residing in the most economically advantaged areas had a higher probability of using SPC of any duration, relative to those without insurance or living in the most disadvantaged areas (Table 2). Decedents in both cohorts who lived in residential aged care had a lower probability of late initiation of SPC (1 month or less before death) relative to those living in the community. Residential aged care was also associated with longer use of SPC in the cancer cohort (Table 2) but not in the non-cancer cohort (Table 3).

The health characteristics associated with the use and duration of SPC included indicators of prior poor health and underlying cause of death. Indicators of prior poor health were associated with a higher probability of earlier use of SPC in the final year of life; in the cancer cohort this was the case for those with a higher comorbidity index in the penultimate year of life (Table 2), while it was the case for those who had spent more days in hospital in the non-cancer cohort (Table 3). Among the non-cancer cohort, those dying from dementia or stroke were least likely to use SPC of any duration, while those dying from chronic obstructive pulmonary disease (COPD) or a neurodegenerative disease (other than dementia) were most likely to use SPC for more than 6 months before death (Cat4) (Table 3). Among the cancer cohort, decedents with a haematological cancer were least likely to use SPC for more than 1 month before death (Cat3 and Cat4) and those with liver cancer were most likely to use early SPC (6 months or more before death) (Table 2).

There were also some treatment factors associated with the use of SPC among the cancer cohort where treatment with radiation therapy or chemotherapy in the last year of life was associated with a higher probability of early use of SPC (Cat3 or Cat4) than for those who did not receive these treatments in their final year (Table 2). Among the non-cancer cohort, where the cause of death was not cancer, the use of radiation therapy in the last year of life was associated with an increased probability of using SPC for 1–6 months (Cat 3), while the use of chemotherapy was associated with a higher probability of using SPC for any duration (Table 3).

The unadjusted probabilities of dying in hospital were similar for those with no SPC (Cat5) and those with SPC for 6 months or more (Cat4); for the non-cancer cohort this was 0.44 and 0.43, respectively, and for the cancer cohort this was 0.60 for both Cat4 and Cat5. Those with shorter use of SPC (Cat1 and Cat2) had the highest probabilities of death in hospital (Table 4). After risk adjustment, those with longer use of SPC (Cat3 and Cat4) had a significantly higher probability of dying in their own home or a nursing home relative to the no SPC group (Cat5) for the cancer cohort. This was only the case for death in a nursing home for the non-cancer cohort. Among the cancer cohort, the probabilities of dying in hospital were 0.64, 0.56 and 0.71 for Cat3, Cat4 and Cat5, respectively, and 0.31, 0.33 and 0.47 among the non-cancer cohort for the same SPC groups. Those with late initiation of SPC (Cat1) continued to have the highest probability of death in hospital after risk adjustment, 0.84 for the cancer cohort and 0.66 for the non-cancer cohort (Table 4). Indicators of higher socio-economic status (income >AU$40,000 and the highest deciles of relative socio-economic advantage) were positively associated with death at home for the cancer cohort but this was not the case for the non-cancer cohort. The MNL models used to produce the adjusted probabilities in Table 4 are provided in the online supplementary material (Online Resource Tables S12 and S13).