Factors Associated with the Costs of Palliative Care: A Retrospective Cost Analysis at a University Cancer Hospital in Brazil

Introduction

Palliative care (PC) is an interdisciplinary specialty focused on improving quality of life for seriously ill patients and their families through better symptom management and coordination of care, as well as optimal communication among professionals, the patient, and the family.^1,2 Studies evaluating economic aspects have suggested that PC results in cost savings during inpatient care and during the end-of-life phase, especially for patients with cancer. ³ However, policymakers should interpret those findings with caution, because there are methodological differences among studies, and there are problems with the analysis, particularly in relation to the skewness of cost data. ⁴ The heterogeneity among such studies shows the challenges of conducting research in this area, such as defining which core components should be included in the economic evaluations. ⁵ To characterize patients with regard to resource use and complexity, it is necessary to consider the most important determinants of resource consumption. ⁶

Like other medical specialties, PC competes for limited and finite resources. Given the unique constraints in low- and middle-income countries (LMICs), payers and policymakers in such countries require evidence to validate the PC currently being provided and to explore ways that it could be made more cost-effective. ⁷

The majority of the studies evaluating the costs of PC have been conducted in the United States, depicting a privately funded health system. ⁸ Therefore, their findings cannot be generalized to a mixed private- and public-funded health care system, as exists in Brazil. The core cost components and the perspective from which analyses are undertaken are necessarily context and country specific. There have been few studies attempting to identify factors associated with a higher cost per day for PC in patients with cancer in LMICs, such as patient characteristics, diagnosis, and prognosis.

The objective of this study was to identify the sociodemographic and clinical variables that could explain the cost per day of PC in patients with cancer in Brazil.

Methods

Results

Costs

The distribution of the formal health care sector costs of the patients who received PC was positively (right) skewed (by 5.5) and overdispersed. The mean total cost per patient was $12,335, ranging from $517 to $262,265 (SD = 14,602; 95% CI = 11,803 to 12,851). After adjustment for LOS, there was no overdispersion (coefficient of variation, 75.7%) and the right skewness decreased (to 1.5). As given in Figure 1, the mean cost per day per patient was $325.50 (SD = 246.30; 95% CI = 316.60 to 334.30).

OPEN IN VIEWER

FIG. 1.

Distribution of total cost per patient (a) and cost per day per patient (b) at the São Paulo State Cancer Institute, 2010–2013.^{a a}Exchange rate, adjusted for purchasing power parity, in 2018: 1.00 U.S. dollar = 3.31 Brazilian reals; current exchange rate: 1.00 U.S. dollar = 5.36 Brazilian reals.

There was a statistically significant difference between one (or more) of the cost per day means for specific cancer sites and “other cancer sites.” Figure 2 provides box plots of the mean cost per day per patient according to the site of the cancer, respectively, showing the variation in the costs with right skewness. The mean cost per day was highest for the sites “other digestive tract locations” ($396.20), bladder ($394.40), blood ($388), and male genital organs ($354.10), whereas it was lowest ($193.00) for cancer located in the thyroid.

OPEN IN VIEWER

FIG. 2.

Box plot of cost per day by cancer site, in ascending order by mean cost. 1 (Thyroid) = $193.0; 2 (liver) = $272.9; 3 (other locations) = $303.2; 4 (prostate) = $309.7; 5 (esophagus) = $310.9; 6 (stomach) = $312.7; 7 (head/neck) = $313.3; 8 (colon/rectum) = $313.6; 9 (kidney) = $315.0; 10 (lung) = $321.6; 11 (brain) = $327.2; 12 (female genital organs) = $328.3; 13 (pancreas) = $330.0; 14 (neuromuscular site, not otherwise specified) = $343.8; 15 (melanoma) = $352.1; 16 (breast) = $353.7; 17 (male genital organ) = $354.1; 18 (hematologic) = $388.0; 19 (bladder) = $394.4; 20 (other digestive tract locations) = $396.2. The line in the box represents the median, the diamond corresponds to the mean, and the width of the box is associated with the sample size.

A univariate analysis showed that the following variables showed a statistically significant association with cost per day (Table 1): age, KPS score, referral to PC, number of hospital admissions, place of death, site of the cancer, and days in hospice. The multivariate analysis showed that sex, age, marital status, KPS score, referral to PC, number of hospital admissions, place of death, and days in hospice all had a significant effect on the cost per day (Table 2). Although the effect sizes were small, sex, age, KPS score, referral to PC, and days in hospice all had a desirable effect. The corresponding estimated parameters are negative, meaning that they can decrease the cost per day, although only slightly. The place of death variable had the greatest impact on the cost per day. When death occurred during hospitalization, the mean cost per day was higher by exp(0.446), corresponding to $1.56, compared with exp(0.605), corresponding to $1.83, when death occurred during hospice care.

OPEN IN VIEWER

Table 2.

Gamma Generalized Linear-Mixed Model with Log Link for Costs per Day

Variable	Log link (cost per day)
	Estimated value	95% CI	P
Intercept	5.661	5.509 to 5.83	<0.001
Male	Reference
Female	−0.060	−0.107 to −0.013	0.012
Age	−0.005	−0.007 to −0.003	<0.001
Married	Reference
Unmarried	0.048	0.001 to 0.095	0.045
KPS score	Reference	—
KPS score <50	−0.164	−0.229 to −0.099	<0.001
KPS score ≥50	−0.362	−0.437 to −0.287	<0.001
No hospital admissions	Reference
1–2 Hospital admissions	0.288	0.221 to 0.356	<0.001
≥3 Hospital admissions	0.220	0.088 to 0.352	0.001
Other referral to palliative care	Reference
Outpatient referral to palliative care	−0.481	−0.546 to −0.416	<0.001
Death in the emergency department	Reference
Death in an external setting	−0.362	−0.472 to −0.253	<0.001
Death in a hospital	0.446	0.340 to 0.551	<0.001
Death in hospice care	0.605	0.512 to 0.699	<0.001
Days in hospice	−0.002	−0.003 to −0.001	<0.001
ϕ	0.626
σb	0.079
Deviance	38,465.32
AIC	38,516.57
SBC	38,670.36

ϕ, Dispersion parameter; σ_b, standard deviation of the random effect; AIC, Akaike Information Criterion; SBC, Schwarz Bayesian Criterion.

Discussion

This study provides preliminary insights into the sociodemographic and clinical variables that explain the cost per day of PC in patients with cancer at a quaternary teaching hospital in Brazil. There was a statistically significant variation in costs depending on the site of the cancer. Multivariate analysis showed that the potential drivers of the cost per day in patients with cancer are the KPS score, the number of hospital admissions, referral to PC, and the place of death.

Our analysis highlights the significant role that the site of cancer plays in understanding variations in health care costs, given that the costs of PC are dependent on patient characteristics such as diagnosis. ¹⁶ However, previous studies have shown that the diagnosis does not significantly influence costs.^6,14,17 Nevertheless, some of those studies did not select diagnosis for inclusion in multivariate models explaining the cost per episode,^6,17 perhaps because of the influence that disparities in health care system and clinical practices have on health care resource utilization.

In accordance with our findings, a study on patient-related costs and factors associated with the cost of in-hospital specialist PC in Germany ⁷ identified factors explaining cost per episode, such factors including the KPS score, phase of illness (which could be related to the number of hospital admissions), and type of discharge (which could be related to the place of death).

Regarding the KPS score, previous studies have shown that extreme dependence was predictive of higher costs in PC units.^6,18 Multivariate models have shown that the phase of illness and KPS score are factors that explain cost per episode. ⁶ In this study, we found that costs were higher for patients with a KPS score <50 and that a reduction in this score was a statistically significant factor influencing cost per day in the GLMM.

Emergency department visits, hospital admissions, and, in particular, intensive care unit admissions are key drivers of resource use and expenditures in cancer care. ¹⁹ Similarly, integrating PC into a hospital setting could have a positive impact on the total budget by lowering the cost per day. A study evaluating a new PC inpatient unit program implemented at Johns Hopkins Medical Institutions showed that performing PC consultations for inpatients reduced the 30-day readmission rate from 15% to 10%. ²⁰ As in this study, these studies showed that number of hospital admissions is a key cost driver, related to intensity of treatment, which could be influenced by a PC program in the hospital setting. A previous cost analysis highlighted the importance of early PC, showing that cost savings from early PC accrue by shortening LOS and reducing the intensity of treatment, shorter LOS being associated with an estimated savings of 63%. ²¹

In this study, referral to PC was also a relevant cost driver, underscoring the importance of early PC. Our findings show that costs were lower for patients referred from outpatient care than for those referred from inpatient care. Patients referred to PC from outpatient had a reduced number of hospitalizations, and emergency visits, and ingresses earlier in PC service when compared with patients referred during hospitalization. Patient- and family-centered discussions on the goals of care and transition planning initiated by outpatient PC teams in the early phases of disease may be at least as important in driving cost savings as are reductions in the number of unnecessary tests and pharmaceutical use in the later phases. ²¹ Early PC consultations may be more related to care planning, which could clarify, for the referring physicians, the limited benefit associated with subsequent hospital admissions. Because late PC consultations may be more focused on symptom management and might not address care planning, their ability to influence the intensity of treatment and LOS is probably more limited. ²¹ Similarly, chemotherapy near death may reflect patient, family, or community preferences and the practice could be opposed by appropriate shared decision making by the treating clinicians, while discussing advanced directives, for instance. ¹⁹

Regarding the place of death, dying in a hospital is costly and is often not aligned with patient preferences related to end-of-life care.^22,23 One previous study showed that the LOS was longer and the cost per day was higher for those who died during hospitalization than for those who died during hospice care. ²⁴ In contrast, our data show that the mean cost per day was higher for the patients who died in hospice care than for those who died in the hospital. One possible explanation for this difference is late referral to hospice care, which means that patients were in a more advanced phase of the disease. Delayed referral can generate significant additional costs, such as administrative, food, and housekeeping costs, as well as those related to nursing, social counseling, and therapist staff, ⁸ which, at this phase of illness, are similar in both settings.

The main contribution of this study is that it provides information on the factors that may impact the costs of treating cancer patients in hospital PC units in LMICs. Therefore, the organization of PC services in these countries must consider aspects that are relevant for the management of hospital costs, such as early inclusion of patients in PC outpatient services, preventing hospital readmissions, and planning the place of death, regardless of the site of the cancer and the functional status of the patient.

At the same time, there has been increasing awareness that early discussion of goals of care, patient values, and preferences, as part of an advance care planning process (ACP), has the potential to reduce health care spending on inappropriate or futile interventions at the end of life.^25,26 Studies suggest that good planning can improve pain control for patients, allow them a greater possibility of dying at home, less chemotherapy, increase utilization of appropriate hospice services, correlate with fewer hospital admissions, and considerably lower health care costs.^27–32 Thus, oncologists and PC specialists must work toward finding the optimal time to discuss end-of-life care options for each patient to achieve high-quality ACP. In this sense, the cost drivers identified in this study may serve as outcomes to assess the impact of ACP implementation in the clinical setting.

Our study has some limitations. First, although it has been suggested that the overall cost of hospitalization is the most useful outcome measure for health managers and is the preferred dependent variable in a primary analysis, ²⁵ we did not evaluate the overall cost as an outcome measure. However, the chosen measure (cost per day) was adjusted for LOS to adjust for variability. Second, our analysis was focused on hospital costs. Future studies should explore costs from the perspective of patients and their families, in terms of informal care. Third, the degree to which the evidence presented is generalizable is unclear, because of the specificities of a quaternary hospital, which do not correspond to the reality at most cancer centers. Nevertheless, our study represents an important first step toward explaining the cost per day of PC at a quaternary teaching hospital in Brazil. It is also important to consider some methodological issues. Variables of interest with missing data, like KPS score information, could be better explored with predictive models. Future studies should also consider other type of models that might allow better understanding of various cost parameters. Because of the skewness and leptokurtosis, which are common to health care utilization data, it may be of interest to model other parameters. Therefore, a Generalized Additive Model for Location, Scale, and Shape (GAMLSS) could be an option.

Conclusion

Our data suggest that a KPS score <50, a higher number of hospital admissions, referral to PC from inpatient care, dying in a hospital, and dying in a hospice setting explained the higher cost per day of PC for patients with cancer at a quaternary teaching hospital in Brazil. To address health care costs, PC teams may consider focusing their efforts on integrating early end-of-life care with other specialties. In this regard, oncologists also have an important role in the delivery of primary PC (PC practiced by primary care clinicians, not specialized in palliative medicine), ³³ and in facilitating timely referral of patients to PC teams, to enhance symptom management, communication, decision making, and end-of-life care.

To allocate resources effectively, cancer hospitals must tailor integration-based interventions according to their resources, size, and the level of primary PC available. PC facilities in LMICs must emphasize early inclusion of patients in PC outpatient services by means of: integrating PC early in the disease trajectory; avoiding hospital readmission while discussing goals of care with patients and families; and planning the place of death, regardless of the site of the cancer and the functional status of the patient, by motivating advance directive completion.