A descriptive study was conducted using data from the Australian Palliative Care Outcomes Collaboration (PCOC), a national quality program funded by the Australian Government Department of Health and Aged Care [12].

The study population included people who: (1) were enrolled in palliative care services registered with PCOC; (2) needed palliative care due to a primary diagnosis of dementia, which included Alzheimer’s disease and other types of dementia; (3) died between 1 January 2013 and 31 December 2020; and (4) at least one assessment was conducted as part of their palliative care within the last two weeks of life before death.

The PCOC program obtains data from patients at three linked levels. Level 1 is the patient level at which various demographic information (sex, age, preferred language, diagnosis, place of death, and days before death) is collected.

The second level is information on each ‘episode of care’, defined as a continuous period of care for a patient in one setting. Two different types of episodes of care, hospital- and community-based palliative care, were investigated in this study. The hospital-based palliative care encompassed patients who were seen in designated specialist palliative care units and patients in non-palliative care designated beds seen by specialist palliative care consultants/teams. The community-based palliative care included patients who received specialist palliative care at private residences or residential aged care facility. The community palliative care teams in Australia are mostly led by palliative care nurses but also include other specialists, such as social workers. They can request support from palliative care physicians when necessary.

The third level of information relates to clinical assessment outcomes at each ‘phase of care,’ which describes the stage/clinical condition and palliative care needs of the patient within an episode. Inpatient palliative care services conduct these assessments at admission, every 24 h thereafter and at discharge to guide patient care. Community-based palliative services perform assessments on admission, during each subsequent patient contact and at discharge. These assessments results are submitted to PCOC biannually and PCOC processes these data for validation and quality assurance. Based on these data, the PCOC national office generates biannual reports on clinical performance for each participating service. These services can then benchmark their performance against national averages and industry-agreed standards, fostering an environment of continuous improvement in palliative care.

The same five standardized tools were used at each phase to assess patients’ urgency of palliative care needs, function, performance, and symptom burden and severity. Clinicians assess patients’ overall palliative care needs based on the holistic assessment of patients and their families using the non-sequential Palliative Care Phases (Stable, Unstable, Deteriorating, or Terminal) [13]. Inter-rater reliability of the Palliative Care Phases is substantial, with a kappa coefficient of 0.67.

Functional dependency is assessed using the Resource Utilization Groups-Activities of Daily Living (RUG-ADL) [14]. Clinicians use the RUG-ADL to measure the motor function of patients related to bed mobility, toileting, transfers, and eating. Total scores range from 4 (independent) to 18 (needs physical assistance by two persons). The RUG-ADL achieves a 56% variance explanation for resource usage.

The Australia-modified Karnofsky Performance Status (AKPS) evaluates patients’ performance relating to work, activity, and self-care [15]. The AKPS is a health professional-rated scale, ranging from 0 (deceased) to 100 (a normal function without evidence of disease), with the scores 10 to 100 captured in the PCOC dataset (as 0 signifies death). The Kappa coefficient for agreement between AKPS and the original KPS is 0.84.

The PCOC Symptom Assessment Scale (PCOC SAS) allows patients or proxies to describe the patient’s level of distress relating to seven common physical symptoms (insomnia, appetite, nausea, bowels, breathing, fatigue, and pain). The symptoms are assessed on a scale of 0 to 10 (0 = none;10 = worst possible). The symptom distress is categorized into four levels according to the scores on PCOC SAS: none (0), mild (1 to 3), moderate (4 to 6), and severe (7 to 10). The Cronbach’s alpha coefficient of PCOC SAS was 0.59 for patient ratings and 0.62 for patient and proxy ratings combined [16].

The Palliative Care Problem Severity Score (PCPSS) is a clinician-rated tool that facilitates the global assessment of four domains of palliative care: pain, psychological/spiritual, family/carer concerns, and other symptoms. Scores range from 0 (absent) to 3 (severe) with higher scores indicating higher levels of severity and difficulty to manage. The PCPSS has moderate levels of inter-rater reliability (Kappa) between 0.38 and 0.48 for the different domains [17].

Stata 17.0 (Stata Corp) was used to perform all analyses. For all analyses, p < 0.05 was used as the level of statistical significance. Patients were removed from our analysis if they had different diagnoses regarding their type of dementia at different episodes of care. Baseline demographic and clinical characteristics of patients were summarized using frequencies or medians (with interquartile range, IQR). We conducted comparisons between the two groups (hospital versus community) using Pearson’s chi-squared, Fisher’s chi-squared, or Wilcoxon rank-sum test depending on the data characteristics. We described the scores on clinical assessments (RUG-ADL, AKPS, PCOC SAS, and PCPSS) using means (95% confidence interval, CI).

We performed multilevel (level 1: individual patient level; level 2: phase level) mixed-effects logistic regression to estimate the differences in symptoms and function trajectories between the hospital versus community-based groups in the last two weeks of life reporting odds ratios (ORs). The hospital group was used as the reference group. Estimates were adjusted for sex, age, diagnosis, palliative care phase, and referral source. Observations with missing values were not included in the models. We assigned RUG-ADL, AKPS, PCOC SAS, and PCPSS as categorical variables by classifying them into different clinical levels as follows: (1) for RUG-ADL: 0 = monitors (corresponding to RUG-ADL = 4–5), 1 = requires one assistant (RUG-ADL = 6–10), 2 = requires one assistant plus equipment (RUG-ADL = 10–15), 3 = as above, pressures area risk, considers carer burden and MDT review (RUG-ADL = 16–17), and 4 = as above, requires two assistants; (2) for AKPS: 0 = completely bedbound (AKPS = 10–20), 1 = increasingly limited mobility (AKPS = 30–50), 2 = required occasional assistance with most care needs (AKPS = 60), 3 = symptomatic and ambulatory (AKPS = 70–80), and 4 = normal activity (corresponding to AKPS = 90–100); (3) for PCOC SAS: 0 = clinical absent (corresponding to PCOC SAS = 0–3), 1 = clinical distress (PCOC SAS = 4–10); (4) for PCPSS: 0 = absent and mild (corresponding to PCPSS = 0–1), 1 = moderate to severe distress (PCPSS = 2–3) [18].