Trends and Outcomes of Early and Late Palliative Care Consultation for Adult Glioblastoma Patients: A SEER-Medicare Retrospective Study


 PurposeGlioblastoma (GBM) carries a poor prognosis despite standard of care. Early palliative care (PC) has been shown to enhance survival and quality of life while reducing healthcare costs for other cancers. No study has investigated differences in PC timing on GBM patient outcomes.MethodsThis study used Surveillance, Epidemiology and End Results (SEER)-Medicare data from 1997-2016. Based on ICD codes, three groups were defined: (1) early PC within 10 weeks of diagnosis, (2) late PC, and (3) no PC. Outcomes were compared between the three groups.ResultsOut of 10,812 GBM patients, 1,648 (15.24%) patients had PC consultation with an overall positive trend over time. There were no significant differences in patient characteristics. There were significant differences in survival among the three groups (P<0.001), with early PC patients with the lowest mean time to death from diagnosis (3.99 ± 4.22 months). The early PC group had significantly lower overall cost of home health aid (1901 ± 3025, p<0.0001) and overall healthcare costs (82842 ± 52726, p<0.0001) compared to other groups. Conclusion We present the first investigation of PC consultation prevalence and outcomes, stratified by early versus late timing, for adult GBM patients. Despite an overall increase in PC consultations, only a minority of GBM patients receive PC. Patients with late PC had the longest survival times. Early PC was associated with lower healthcare costs and resource utilization when accounting for the patients’ entire disease course. Prospective studies can provide additional valuable information about this unique population of GBM patients.


Introduction
Glioblastoma (GBM), the most common primary brain malignancy diagnosed in adults, carries a poor prognosis despite standard of care treatment. With an overall median survival of 16-21 months, GBM patients suffer from neurologic and cognitive symptoms, undergoing treatments that further affect their body and mind [1,2]. Palliative care (PC) is de ned by the World Health Organization as "an approach that improves the quality of life of patients and their families facing…life-threatening illness, through the prevention and relief of suffering…" in a broad manner [3]. Literature about PC and its bene ts for patients with advanced cancers is relatively developed, stating improvements in not only survival but also symptoms, mood, quality of life and healthcare costs [4,5]. In particular, early PC was investigated as an intervention in a well-known clinical trial, revealing survival bene t for non-small cell lung cancer (NSCLC) patients receiving PC consultation within 8 weeks of diagnosis [6].
Early PC for GBM patients has not been studied in detail and warrants additional attention given the neurocognitive symptoms associated with both the disease and treatment course [7]. 7 Our retrospective study based on a nationwide cancer registry characterizes the landscape of early and late PC for adult GBM patients.

Data source
This study used the Surveillance, Epidemiology and End Results (SEER)-Medicare Linked Database. SEER is a national program of cancer registries collecting clinical, demographic and cause of death information for persons with cancer. Medicare is a Center for Medicare and Medicaid Services health insurance program covering the elderly (65 or older), people with certain disabilities or end stage renal failure from eligibility to death. Medicare data is composed of claims as enrollees navigate through the covered healthcare services.
We used SEER-Medicare 1997-2016 with claims from the ve following les: (1) Medicare Provider Analysis and Review les, which includes all Part A short stay, long stay, and skilled nursing facility with one summarized record per admission (2) Carrier Claims les, which includes collected physician/supplier (Part B) bills for 100 percent of all claims (3) Outpatient les, which contain Part B claims for 100 percent for each calendar year from institutional outpatient providers (4) Home Health Agency, which contains 100 percent of all claims for home health services (5) Hospice les which contains claims data submitted by providers.
PC encounter was identi ed using International Classi cation of Diseases, Ninth Revision (ICD-9) code V66.7 and ICD-10 code Z51.5 from Medicare data [8,9]. For those with >1 PC encounter, only the rst instance was included. Three analysis groups were de ned: (1) the early PC group whose rst PC encounter occurred within 10 weeks of diagnosis, (2) the late PC group whose rst PC encounter occurred after 10 weeks of diagnosis, and (3) the no PC group. The timing of 10 weeks was selected based on previous publications citing a range of timing for implementing early PC for other advanced cancer diagnoses [6,10].

Patient characteristics
Patients' characteristics include age at diagnosis, comorbidities, GBM location and treatment related information. The Elixhauser score was used to account for the burden of comorbidities [11]. We used the adaptation to ICD-9-CM codes developed by Quan et al [12].
Comorbidities were evaluated during 3 months before diagnosis.

Outcomes of Interest
We assessed the following measures of health care utilization and cost from diagnosis to death, within 30 days and 6 months leading to death: (1) number of emergency room (ER) visits, intensive care unit (ICU) admissions, outpatient visit, hospital admission, home health aid (HHA) use and hospice use; (2) total number of days for ICU stay, and total of days for hospital stay; (3) the Medicare payment to outpatient, hospital, HHA and hospice use, respectively; (4) Total Medicare payment. Payments were in ation adjusted to 2016 US dollars using the medical component of the consumer price index (accessible through the United States Bureau of Labor Statistics website) [13]. The analysis of outcomes within the last 30 days and 6 months of life were performed in subgroups of those who survived at least 30 and 60 days after diagnosis, respectively. We also evaluated survival.

Statistical method
Demographics were compared using Kruskal Wallis test for continuous variables and Chi-square test for categorical variables among the groups. To obtain comparative groups by accounting for bias due to observed confounders, inverse probability of treatment weight (IPTW) technique was used [14]. A propensity score was calculated using a logistic regression model where the group was the dependent variable and all patient characteristics were included as independent variables. For each person, the weight was calculated as the sample size adjusted inverse of the propensity of getting the PC that they actually got. Post IPTW covariate balance was evaluated. Outcomes were compared between the three groups using IPTW-weighted generalized linear regression, negative binomial regression for the number of healthcare utilization encounters and linear regression on log-transformed values for payments. All tests were 2-sided and a p-value < 0.05 as considered statistically signi cant. Statistical analyses were performed with SAS 9.4 (SAS Institute Inc., Cary, NC).

Patient Characteristics
A total of 10,812 patients over age 66 who were diagnosed with glioblastoma (GBM), had Medicare insurance and records, and were deceased prior to 2017 were included ( Figure 1). 1,648 (15.24%) GBM patients had PC consultation at any point during their disease course prior to death. Early PC involvement was de ned as consultation within 10 weeks of GBM diagnosis. Temel et al. enrolled and randomized their early PC patients with metastatic lung cancer within 8 weeks of diagnosis, whereas another group de ned early PC as within 12 weeks of diagnosis for adult patients with solid malignancies [6,10]. 676 (6.25%) GBM patients in our study met criteria for early PC consultation. 972 (8.99%) GBM patients had late PC consultation.
Following IPTW matching, there were no signi cant differences in demographics characteristics among groups of patients who either did not receive or received early or late PC. The average age across the three subgroups (no PC, early PC, and late PC) was 76 ± 6 years, and the majority of GBM patients were male (no PC: 54.7%; early PC: 51.9%; late PC: 54.%) ( Table 1). The Elixhauser Index is a measure of mortality based on 31 co-morbidities, and most GBM patients within each subgroup had scores over 3 [15]. There were no signi cant differences in tumor location among the PC subgroups. Furthermore, GBM patients within each PC subgroup underwent similar rates of biopsy, surgical resection, radiation therapy, and chemotherapy.

Patient Survival among Palliative Care Subgroups
Following IPTW matching, there were signi cant differences in the survival curves among GBM patients (P<0.001) ( Figure 2). GBM patients who received early PC had a mean time to death from diagnosis as 3.99 ± 4.22 months, while GBM patients who received late PC had the longest mean time to death from diagnosis as 11.72 ± 13.20 months (Supplemental Table 1). Those who did not receive any PC during their disease had overall mean survival time of 7.76 ± 9.23 months.
Healthcare Utilization among Palliative Care Subgroups Various aspects of healthcare utilization were categorized for each PC subgroup of GBM patients and divided into the last 30 days of life, the last 6 months prior to death, and overall rates occurring in the time from GBM diagnosis to death ( Table 2).
In some areas of healthcare utilization, the subgroup of GBM patients who received late PC had the highest healthcare resource utilization over the entire disease course. This group demonstrated signi cantly higher average numbers of ER visits (3.46 ± 2.92, ICU admissions (1.50 ± 1.14), overall inpatient hospital admissions (4.48 ± 2.71), outpatient visits (19.38 ± 23.09), and length of stay in days (48.02 ± 48.73). On the other hand, patients in the late PC subgroup also had signi cantly greater use of hospice (2.25 ± 2.60) and HHA (1.32 ± 1.37).
In terms of Medicare payments and costs over the patients' disease courses, the early PC group had signi cantly lower overall cost of HHA (1901 ± 3025, p<0.0001), cost of outpatient visits (6033 ± 11779, p<0.0001), and overall healthcare costs (82842 ± 52726, p<0.0001) compared to both the no PC and late PC groups. Conversely, the late PC subgroup demonstrated signi cantly higher overall costs of inpatient admissions (74091 ± 51143), outpatient visits (17690 ± 28623), and overall healthcare (129236 ± 76892).
When accounting for healthcare costs in the last 6 months prior to death for GBM patients who had received PC, the late PC subgroup had signi cantly greater overall Medicare payments (62650 ± 41081) compared to patients with early PC (47215 ± 30923). Likewise, in the last 6 months prior to death, the late PC subgroup had signi cantly greater numbers of ER visits (1.99 ± 1.69), ICU admissions (0.51 ± 0.85), hospital admissions (2.14 ± 1.99), outpatient admissions (9.95 ± 8.81), overall days for length of stay (23.02 ± 32.04) compared to GBM patients without PC and those who received early PC.
Some of these differences between early and late PC were abrogated when examining the last month prior to death, such as numbers of ICU admissions, hospital admissions, days for length of stay as well as HHA and hospice use. The healthcare costs associated with PC are signi cantly greater than for those who did not receive PC in the last month of life.

Trends in Palliative Care Utilization
From 1997 to 2015, there was an overall increase in PC use for GBM patients with a positive trend from 2.64% in 1997 to 42.54% in 2015 ( Figure 3; Supplemental Table 2). For early PC consultation, the proportion of GBM patients meeting this criterion rose from 1.13% in 1997 to 20.63% by 2015. Likewise, the percentage of GBM patients who received late PC also increased from 1.51% in 1997 to 21.92% in 2015 (Supplemental Table 2). Similarly, for GBM patients who underwent biopsy or craniotomy, there was also an increase in PC use.

Discussion
Our retrospective study investigated early versus late PC use for 1,648 adult GBM patients from a cohort of 10,812 patients. Existing studies on PC for patients with primary brain malignancies are already rare, with those focusing on GBM even more so. A minority of adult GBM patients received PC consultations (34-40%) [16]. For those who received PC, patients and caregivers reported improved quality of life [17]. This study offers two unique features as it is the rst study examining the landscape of PC use and effects of early versus late PC consultation as well as the rst utilizing a national claims database to study PC for GBM patients.
In general, there was a positive trend in PC consultation over our study period. A separate study on patients with primary brain malignancies, which include GBM, found that rates of inpatient PC also rose from 2.3% in 2007 to 11.9% in 2011 [18]. The rise in PC use among GBM patients re ects a general trend of greater adoption in cancer care [19,20]. However, it is worth noting that, while historically regarded as end-of-life care, PC has been formally de ned as applicable earlier in a patient's disease course. And, just as GBM management has transformed over time to incorporate new protocols, palliative and supportive care have also changed [21,22].
Our analysis yielded no signi cant differences in patient or tumor characteristics among the three subgroups. Understanding prognosis is a key component-and bene t-of promoting timely PC implementation [23,24]. We would have assumed that certain patient or disease characteristics could in uence providers' decision-making in consulting PC early or at all. Several large-scale retrospective studies and predictive models have identi ed various factors, such as gender, functional status, and tumor location, as signi cant in prognosticating GBM [25][26][27]. However, no factor was necessarily correlated with PC use in our analysis.
Furthermore, there were no differences in biopsy versus surgery rates among the subgroups. Patients with brain metastases who underwent neurosurgery had signi cantly lower rates of PC consultation [28]. Another study on patients with primary brain malignancies indicated that those who received inpatient PC were less likely to undergo treatments, including surgery [18]. There was no analysis on timing of PC with regards to procedures for our study due to database limitations.
The difference among subgroup survival times is striking. Patients with late PC consultation had the longest median survival time, whereas the early PC group had the shortest. Other published studies on early PC and survival time for patients with advanced cancer showed variable results. A cohort of NSCLC patients who had PC within 8 weeks of diagnosis experienced better quality of life, received aggressive end-of-life care less often, and ultimately had longer survival [6]. 6 Early PC was also found to positively in uence quality-of-life and increased reported scores across a synthesis of seven randomized control trials on patients with incurable cancers [29]. However, the four studies that reported survival data did not reveal differences in e cacy. Quality-of-life was higher among patients with incurable solid tumor cancers who received PC consultation within 12 weeks of diagnosis [10]. But, no signi cant difference in survival time between the early PC and standard oncologic care groups was identi ed.
Early PC involvement may depend on opinions about optimal timing, which vary greatly among physicians. A focus group study involving patients with various cancers indicated that the best timing of PC implementation was not necessarily soon after diagnosis but after failure of curative treatment, disease recurrence, and other signs of progressive poor prognosis [30]. Perhaps, our study's early-PC patients were deemed among the most likely to have poor survival by their providers, despite no signi cant differences in patient characteristics.

Limitations
The retrospective nature of our study presents several limitations. Selection bias is inherent to retrospective studies, although we utilized IPTW-matching to control for covariates. The SEER database is limited by each hospital's coding practices, which may affect the sensitivity and speci city of patients identi ed in our study. We were also limited by the lack of standardization of PC. Furthermore, the database does not include data related to location of death, symptoms, or any patient-reported outcomes, which are also important outcomes to consider in PC studies [6]. Functional status and frailty metrics, in addition to other potentially salient information that could in uence providers' decision-making about early PC implementation, are also missing. The strengths of our study lie in the national sampling of GBM patients from a comprehensive, well-utilized database, which allows for investigation of PC effects on a rare but deadly cancer.

Conclusion
We present the rst investigation of PC consultation, strati ed by early versus late timing, for adult GBM patients. Despite an overall increase in PC consultations over time, only a minority of GBM patients receive PC despite the poor prognosis. Patients with PC, albeit after ten weeks from diagnosis, had the longest survival times. Early PC within 10 weeks of diagnosis did not necessarily indicate longer survival times but was associated with lower healthcare costs and resource utilization when accounting for the entire disease course.
Prospective studies and randomized controlled trials are required to provide valuable information about for the impact of PC for the unique population of GBM patients.

Declarations
Funding: The authors have no funding sources to report.
Con icts of Interest: The authors have no con icts to report.
Availability of Data: The authors will provide data upon request.
Code Availability: The authors will provide code upon request.