The association between varying levels of palliative care involvement on costs during terminal hospitalizations in Canada from 2012 to 2015

Background Inpatient palliative care is associated with lower inpatient costs; however, this has yet to be studied using a more nuanced, multi-tiered measure of inpatient palliative care and a national population-representative dataset. Using a population-based cohort of Canadians who died in hospital, our objectives were to: describe patients’ receipt of palliative care and active interventions in their terminal hospitalization; and examine the relationship between inpatient palliative care and hospitalization costs. Methods Retrospective cohort study using data from the Discharge Abstract Database in Canada between fiscal years 2012 and 2015. The cohort were Canadian adults (age ≥ 18 years) who died in hospital between April 1st, 2012 and March 31st, 2015 (N = 250,640). The exposure was level of palliative care involvement defined as: medium-high, low, or no palliative care. The main measure was acute care costs calculated using resource intensity weights multiplied by the cost of standard hospital stay, represented in 2014 Canadian dollars (CAD). Descriptive statistics were represented as median (IQR), and n(%). We modelled cost as a function of palliative care using a gamma generalized estimating equation (GEE) model, accounting for clustering by hospital. Results There were 250,640 adults who died in hospital. Mean age was 76 (SD 14), 47% were female. The most common comorbidities were: metastatic cancer (21%), heart failure (21%), and chronic obstructive pulmonary disease (16%). Of the decedents, 95,450 (38%) had no palliative care involvement, 98,849 (38%) received low involvement, and 60,341 (24%) received medium to high involvement. Controlling for age, sex, province and predicted hospital mortality risk at admission, the cost per day of a terminal hospitalization was: $1359 (95% CI 1323: 1397) (no involvement), $1175 (95% CI 1146: 1206) (low involvement), and $744 (95% CI 728: 760) (medium-high involvement). Conclusions Increased involvement of palliative care was associated with lower costs. Future research should explore whether this relationship holds for non-terminal hospitalizations, and whether palliative care in other settings impacts inpatient costs. Supplementary Information The online version contains supplementary material available at 10.1186/s12913-021-06335-1.

Treatments that tend to drive these end-of-life costs are sometimes termed aggressive interventions, such as admission to intensive care units (ICU), mechanical ventilation, defibrillation, dialysis, percutaneous coronary interventions, and percutaneous feeding tube insertions. A recent study of decedents in Ontario demonstrated that, in the last 6 months of life, 19.4% were admitted to the ICU, and 13.9% received mechanical ventilation [4].
A meta-analysis on the association between palliative care consultations and direct hospital costs for adults with serious illness found that patients who received palliative care consultations had a statistically significant reduction in direct hospital costs (−$3237 USD; 95% CI, −$3581 to -$2893; P < .001) [5]. There are limitations to previous studies. Most studies focus on individual hospital or state-level costs. Most previous studies analyzed the relationship between palliative care and costs as a binary variable [6][7][8][9]. In reality there are differing levels of inpatient palliative care involvement. A binary variable does not give a good sense of how much palliative care people are receiving, and a more nuanced variable can help decision makers determine how to appropriately allocate resources.
Due to past studies' lack of nuanced palliative care variables and lack of national scope, we sought to understand the relationship between differing levels of inpatient palliative care and hospitalization costs in terminal hospitalizations. Our first objective was to describe patients who died in hospital, and their receipt of different levels of palliative care and active interventions.
Our second objective was to examine the relationship between terminal hospitalizations with varying levels of palliative care involvement and costs.

Study design, setting, and data sources
We conducted a national retrospective cohort study, using patient-level health administrative data from the Discharge Abstract Database (DAD) in Canada (excluding the province of Quebec 1 and the territories). The DAD is a national database that includes data from acute care institutions across Canada. It is mandatory for all provinces, except Quebec, to submit a DAD abstract for each hospital discharge. A trained Canadian Institute for Health Information data abstractor reviews the medical record for the encounter and makes determinations about the care received to populate the DAD. Analysis was conducted from January 2020-July 2020.

Study cohort
Our cohort included all adults who died in hospital between April 1st, 2012 and March 31st, 2015. We excluded people who were under age 18 or over age 105 at death, or who were non-Canadian residents. We excluded people with sex not identified as either male or female because there were few records and we could not meaningfully analyze the end-of-life experience of these individuals. We excluded people with missing postal code data, missing admission/discharge dates, or missing/zero inpatient resource intensity weight data, which results in missing/zero cost data. We excluded people from Yukon, Northwest Territories, and Nunavut due to small sample sizes and data unavailable for costs of a standard hospital stay in the Northwest Territories and Nunavut. We also excluded people with erroneous data entries concerning their admissions and death (e.g., multiple terminal admissions). Finally, we excluded people whose terminal hospitalization had a length of stay greater than 180 days, as these people are not representative of a typical acute care stay and may be capturing long-term care beds that exist within some Canadian hospitals.

Patient characteristics
We measured demographic and clinical variables including age, sex, rurality, province, comorbidities and chronic conditions defined as per the Elixhauser index [10] and the Van Walraven index (a weighted numeric score based on the Elixhauser classification system designed to predict in-hospital mortality) [11], represented as empirical quartiles of the score observed in our sample. We also measured ICU stay, receipt of major surgery [12] and "any active intervention", defined as mechanical ventilation, cardiopulmonary resuscitation (CPR), defibrillation, dialysis, percutaneous coronary intervention, percutaneous feeding tube insertion, blood transfusion, and bronchoscopy. We developed this list of active interventions based on consensus among our group of experts in palliative and intensive care.

Exposure
Our primary exposure was palliative care involvement, adapted from Webber et al. [13]. For each admission, the DAD captures up to 25 diagnosis codes based on the 10th revision of the International Statistical Classification of Diseases and Related Health Problems (ICD-10). Medium-high palliative involvement was defined as a hospitalization with palliative care as the most responsible service provider (i.e., the service that treated the patient for the majority of their hospital stay; this could include care delivered by a specialist palliative care provider or a generalist providing palliative care) and/or the most responsible diagnosis (i.e., when the patient is admitted solely to receive palliative care or when the patient received palliative care prior to hospital arrival but was not admitted for a reversible cause; using ICD-10 diagnosis code Z51.5). Low Palliative involvement was defined as a hospitalization with a secondary diagnosis of palliative care (i.e., using diagnosis codes 2 through 25), excluding most responsible provider. No involvement was defined as a hospitalization without a palliative diagnosis or most responsible service provider. In Canada, inpatient palliative care is largely provided as a consult service; here palliative care would not be noted as most responsible provider, but would likely be noted as most responsible diagnosis or secondary diagnosis. Inpatient palliative care might also include palliative care units, which would have the palliative care designation as most responsible provider.

Outcome
The total costs of a terminal hospitalization were calculated using validated costing methods [14][15][16][17]. Each acute care hospital admission is associated with an inpatient resource intensity weight (RIW), which measures the intensity of resource use associated with medical, diagnostic and surgical procedures, adjusted for patient and institutional characteristics (e.g., case-mix adjustments). Inpatient hospital costs are calculated by multiplying the RIWs by a cost-of-standard-hospital-stay, which is measured at the provincial level. We also adjusted costs to 2014 using the Statistics Canada Health Care Consumer Price Index (CPI) [18]. All costs are in Canadian dollars.

Statistical analysis
For Objective 1, wherein we describe patients who died in hospital and their receipt of palliative care, we presented descriptive statistics as mean (standard deviation (SD)), median (interquartile range (IQR)), and n(%). For Objective 2, wherein we examined the relationship between varying levels of palliative care involvement and costs, we fit a generalized estimating equation (GEE) model, assuming our response variable (inpatient hospital costs) followed a gamma distribution. We used a log link function. We accounted for clustering by hospital using a compound symmetric working correlation structure. We include the natural logarithm of patient length of stay as an offset in our gamma GEE models. Our unadjusted model had palliative care involvement as the only predictor and costs as the outcome. Based on clinical expertise and estimated bivariate associations from our unadjusted models, we built a multivariable gamma GEE model controlling for age, sex, province, and Van Walraven index. We present regression coefficient estimates (representing log cost-ratios), 95% confidence intervals (CIs) and p-values from the fitted gamma GEE models. For both unadjusted and adjusted models, we present the conditional expected value of cost (and its 95% CI) at each level of palliative care intensity, controlling for other variables in the model.
We conducted two sensitivity analyses. First, we estimated stratified multivariable models by province because palliative care delivery [19] and the cost of a standard hospital stay differs across provinces. Second, we estimated stratified multivariable models by Elixhauser comorbidity categories [10], specifically metastatic cancer, heart failure, major neurocognitive disorders, chronic obstructive pulmonary disease (COPD), kidney failure, human immunodeficiency virus/acquired immunodeficiency syndrome (HIV/AIDS), and liver failure. We focused on these diseases categories as they are known to play a role in the relationship between palliative care receipt and associated costs [5,20]. We report adjusted costs from the stratified gamma GEE model fits. All analyses were performed using SAS version 9.4 (SAS Institute, Cary, North Carolina).
This study was approved by Institutional Ethics Review Boards at Sinai Health and the Ottawa Health Services Network. While data sharing agreements prohibit the research team and Canadian Institute for Health Information from making the dataset publicly available, access may be granted to those who meet pre-specified criteria for confidential access. The underlying analytic code are available from the authors upon request, with the understanding that the programs may rely on coding templates or macros that are unique to Canadian Institute for Health Information.

Sensitivity analyses
Adjusting for age, sex, and Van Walraven index, we found that higher palliative care involvement was associated with lower costs in all provinces (Table 4) Adjusting for age, sex, province, and Van Walraven index, we found that higher palliative care involvement was Any active intervention is defined as any of the following: mechanical ventilation, CPR, defibrillation, dialysis, percutaneous coronary intervention, feeding tube, blood transfusion, and bronchoscopy  associated with lower costs across Elixhauser comorbidity categories (Table 5).

Discussion
In this study, we demonstrated increased involvement of palliative care was associated with lower costs.. However, a substantial proportion (38%) had no inpatient palliative care involvement in their terminal hospitalization and a minority (24%) received medium-high palliative care involvement. In addition to cost, these findings are also reflective of the care experience of Canadians who die in hospitals; our analysis found that the proportion of decedents with any active interventions decreased with increasing levels of palliative care involvement. This relationship between palliative care involvement and costs was sustained while stratifying across provinces and disease groups. The relatively low level of inpatient palliative care receipt was also observed in a national study of decedents who were hospitalized in their last year of life, wherein 18.7% received inpatient palliative care consultation [21]. In our study, the majority of people received little to no palliative care in their terminal hospitalization. This finding is in line with our past work on terminal hospitalizations in Ontario from 2012 to 2015, which found that 36% never had access to palliative care [22]. These low levels of palliative care receipt are problematic, as the Canadian Institutes for Health Information estimates that 94% of patients who died in hospital were hospitalized for a non-sudden death condition that could potentially have benefitted from palliative care involvement [19]; however, it is unclear how that report operationalizes potential benefit.
Our findings reinforce past studies demonstrating that inpatient palliative care is associated with lower inpatient costs [5][6][7][8][9]23]. A systematic review of the economic evidence on specialist palliative care consultations found that there were statistically significant savings of 9-25% associated with palliative care consultation [23]. However, ours is the first national study to use a tiered palliative care variable demonstrating the relationship between palliative care and lower costs. Importantly, our findings are restricted to terminal hospitalizations and future studies would need to analyze whether this relationship holds for other hospitalizations.
These lower costs associated with palliative care might be a consequence of patients receiving palliative care undergoing fewer active interventions and having fewer ICU stays, which is likely aligned with the patients' underlying preferences for care. Other studies have found that palliative care consultations are associated with lower intensity of hospital treatment, less expensive procedures, shorter length of stay [23][24][25][26][27][28][29][30], decreased use of chemotherapy near death, and lower risk of ICU admission, multiple emergency department visits, and multiple hospitalizations near death [31]. Hsu et al.'s national, Canadian study found that, in an adjusted analysis, receipt of palliative care was associated with lower odds of ICU admission in the last 30 days of life (OR: 0.31, 95% CI: 0.30-0.32) [21]. Importantly, while palliative care is associated with a reduced likelihood of receipt of these treatments, it may be because patients' preferences to receive less aggressive care is driving higher rates of palliative care and lower rates of these active interventions. Further, lifesustaining measures can cause or prolong suffering in a dying person, but the goals of palliative care and intensive care are not mutually exclusive. There are many models for integrating palliative care principles into the critical care environment [32], and one recent study found that among hospital decedents, ICU care was associated with higher family ratings of quality of end-of-life care than ward care [33].
The dose-response relationship between palliative care and costs was also observed across disease groups, including metastatic cancer, heart failure, major neurocognitive disorders, COPD, kidney failure, HIV/AIDS, and liver failure. These findings are consistent with diseasespecific palliative care economic evaluations. A metaanalysis of six studies on the economic impact of inpatient palliative care consultations, found that for studies focusing on cancer, palliative care led to a reduction in costs (−$4251 USD; 95% CI, −$4664 to -$3837). The same study found that for noncancer populations, there was also an association between palliative care and lower costs (−$2105; 95% CI, −$2698 to -$1511) [5]. May et al. conducted a pooled analysis of four economic evaluations of non-cancer diseases. They found that Major surgery is defined according to anatomic site: abdominal, cardiac, retroperitoneal, thoracic and vascular [12] inpatient palliative care was associated with a reduction in total direct costs for heart failure, neurodegenerative conditions, COPD, kidney failure, and liver failure. They did not find an association for HIV/AIDS [20].
The magnitude of costs varied across provinces, which is expected, as there are differences in provincial offerings of palliative care, and cost-of-standard-hospital-stay. However, some of these observed differences might be due to inter-provincial administrative coding differences. The mandatory elements to be coded vary by province, and there are noted inter-provincial inconsistencies in the format and coding of data elements, including diagnoses and procedures [34]. Notably, in our descriptive findings, the mean cost per day in Alberta was higher than any other province at $1250. This may relate to two factors, though further analysis is required to determine a causal link. First, Alberta has the highest per capita health spending of all the provinces, estimated at $7658 per person in 2019, compared to the national average at $7068 [35]. Second, Alberta has the lowest levels of inpatient palliative care of all the provinces at 15.2% [21]. When comparing people who died in acute care from 2016 to 2017, Alberta    [19]. Despite these variations, we observed that the dose-response relationship was maintained across provinces. Our paper has several limitations. First, costs were derived in a top-down approach using resource intensity weights, which is less precise than a microcosting approach; thereby limiting the generalizability of our findings. Second, there may be inconsistencies with how often and under what circumstances palliative care is coded across the provinces. Challenges with the subjective nature of the deeming palliative care as the most responsible diagnosis code and the risk of this designation being overused or misused was noted by Downar et al. in 2010 [36]; however, these issues persist. Third, the method we used for identifying receipt of palliative care, approximates receipt of palliative care, but further research is needed to verify that these methods are truly identifying encounters with palliative care providers and that the designation of low or medium/high is appropriate. This method also does not capture the intensity of palliative care receipt. Fourth, there are two main limitations with using a decedent cohort in terminal hospitalizations. We cannot draw causal inferences; however, we can examine the costs of care provided at the end of life. Further physicians do not necessarily know when a patient is admitted to hospital that he/she will die in hospital, so our findings cannot necessarily be leveraged for prospective decision making. While prognostication is inexact, future prospective studies may more precisely be able to examine the relationship between palliative care involvement, hospitalizations, and costs. Fifth, there are limitations to using the DAD; we are unable to measure the frequency of palliative care (i.e., how many times a patient was seen by palliative care specialists) or the quality of palliative care, when in the hospitalization palliative care services were received, what constitutes a palliative care service or provider, and concordance between preferences for care and services received. While in our adjusted models we controlled for variables we thought might confound the relationship between palliative care and costs, there may still be uncontrolled confounding regarding who receives palliative care, as well as the availability of hospital-based, specialist palliative care. Finally, our data was restricted to inpatient data, which precluded us from assessing the impact that outpatient palliative care might have on our results. Outpatient palliative care has been associated with decreased health utilization and increased likelihood of home deaths [37][38][39][40][41], which might lead patients receiving outpatient palliative care to be excluded from our cohort.

Conclusions
Increased palliative care involvement is associated with lower inpatient costs. This is the first national population-representative analysis on the impact of inpatient palliative care on costs using a tiered palliative care variable. Future research should explore whether this relationship holds for other hospitalizations towards the end of life, as well as whether palliative care in other settings impacts inpatient costs.