End-of-Life Cost Trajectories in Cancer Patients Treated by Medicare versus the Veterans Health Administration

Abstract

BACKGROUND/OBJECTIVES:

To evaluate differences in end-of-life cost trajectories for cancer patients treated through Medicare versus by the Veterans Health Administration (VA).

DESIGN:

A retrospective analysis of VA and Medicare administrative data from FY 2010 to 2014. We employed three-level generalized estimating equations to evaluate monthly cost trajectories experienced by patients in their last year of life, with patients nested within hospital referral region.

SETTING:

Care received at VA facilities or by Medicare-reimbursed providers nationwide.

PARTICIPANTS:

A total of 36,401 patients dying from cancer and dually enrolled in VA and Medicare.

MEASUREMENTS:

We evaluated trajectories for total, inpatient, outpatient, and drug costs, using the last 12 months of life. Cost trajectories were prioritized as costs are not directly comparable across Medicare and VA. Patients were assigned to be VA-reliant, Medicare-reliant or Mixed-reliant based on their healthcare utilization in the last year of life.

RESULTS:

All three groups experienced significantly different cost trajectories for total costs in the last year of life. Inpatient cost trajectories were significantly different between Medicare-reliant and VA-reliant patients, but did not differ between VA-reliant and Mixed-reliant patients. Outpatient and drug cost trajectories exhibited the inverse pattern: they were significantly different between VA-reliant and Mixed-reliant patients, but not between VA-reliant and Medicare-reliant patients. However, visual examination of cost trajectories revealed similar cost patterns in the last year of life among all three groups; there was a sharp rise in costs as patients approach death, largely due to inpatient care.

CONCLUSION:

Despite substantially different financial incentives and organization, VA- and Medicare-treated patients exhibit similar patterns of increasing end-of-life costs, largely driven by inpatient costs. Both systems require improvement to ensure quality of end-of-life care is aligned with recommended practice.

INTRODUCTION

Intensive medical treatments are common at end of life, but not always beneficial.¹ There is increasing recognition, including from the American Society of Clinical Oncology and the National Quality Forum, that intensive medical care at the end of life may indicate poor quality care² and may supplant the focus on palliation and quality of life that the National Academy of Medicine notes should be the primary focus for patients at this stage.³ Past work has examined the intensity of care received by patients in their last month or days of life^4,5; however, without understanding patients’ earlier experiences of care, it is not clear whether intensive care in the last weeks of life is a continuation of previous intensive care or represents increased intensity of care. In this study, we evaluate care received in the last year of life to ascertain whether patients experience increasing or decreasing medical intensity as they approach death, as measured by costs of care incurred. We investigate differences in cost trajectories in veterans treated by Veterans Health Administration (VA) versus veterans treated through Fee-for-Service (FFS) Medicare. More than 90% of veterans aged 65 or older are enrolled in Medicare,⁶ with many using both VA and Medicare for their healthcare needs.⁷ Evaluating a dually enrolled cohort allows for analyses on the effect of provider type (VA or private sector) on care. VA is an integrated organization that promotes palliative care and is absent the financial incentives for overuse of care present in Medicare; it is thus possible that VA-treated veterans have different cost trajectories in the last year of life compared to veterans treated through Medicare.

We focus on cancer decedents, given their disproportionate effect on U.S. healthcare spending,^3,8–10 the low value of much end-of-life care for this population,^3,10,11 and the high financial burden experienced by many cancer patients.^12–15 Additionally, many cancer patients experience functional decline several months before death that objectively signals their poor prognosis.¹⁶ Our evaluation of cost trajectories provides insights regarding end-of-life care practices in health systems with substantially different financial incentives and organization as well as important policy implications as VA moves more toward purchasing of care in the community through the MISSION Act.

METHODS

Cohort

Our cohort consisted of veterans dying from solid tumors from FY 2010 to 2014, as assessed by underlying cause of death from National Death Index death certificate data.¹⁷ These data represent care provided prior to the passage of the Choice Act, which allowed VA-enrolled veterans to directly seek care in the community instead of at the VA. Patients were 66 years or older at death, continuously enrolled in Medicare Parts A and B and VA in the 12 months before death, and had a diagnosis of solid tumor for at least 365 days. We excluded the small minority of patients also enrolled in Medicaid; cost or utilization data were not available for these patients.

Data

We evaluated total costs of care in the last year of life from a health system perspective. Costs were obtained from Medicare claims data and VA administrative data. In Medicare data, other payer costs were subsumed in health system costs. To calculate VA health system costs, we used Managerial Cost Accounting (MCA) datasets for inpatient, outpatient, and pharmacy care. VA also contracts out some care to the private sector that it is infeasible for VA to provide; this is referred to as Fee Basis care. During the time horizon of our study, Fee Basis care represented approximately 11% of total VA costs.¹⁸ Therefore, in sensitivity analyses, we ran separate models including Fee Basis costs, which we subsumed as a part of VA costs.

Costs were allocated to categories of inpatient, outpatient, drug, or other. To properly characterize costs, the drug category included oral and intravenous/injection drugs administered on an outpatient basis, the latter of which represent a substantial portion of spending for cancer patients. In Medicare data, we allocated professional fees associated with inpatient care in the Carrier file to inpatient stays; excluding these costs from can substantially underestimate inpatient costs.¹⁹ In Medicare data, drug costs were identified using Berenson-Eggers Type of Service (BETOS), Healthcare Common Procedure Coding System (HCPCS), and International Classification of Diseases-9 (ICD-9) codes in Part D, Outpatient, Carrier, and Durable Medical Equipment files. In VA data, drug costs and utilization were identified using Current Procedural Terminology (CPT) and ICD-9 codes in Medical SAS Outpatient files, MCA Pharmacy files, and Fee-Basis files.

Data were compiled to facilitate comparisons of Medicare and VA. In VA, inpatient data include hospital, hospice, and skilled/long-term nursing home care. To properly compare these data with Medicare, we included only hospital stays in VA inpatient costs. VA also provides services that Medicare does not reimburse, such as long-term nursing care, domiciliary stays and dental visits. Including such costs would make VA look more expensive due to its greater breadth of services. Conversely, Medicare reimburses care to which VA providers often refer but for which there is much less capacity to provide in VA, specifically, hospice, and home health. For example, Medicare pays for 90% of hospice stays for veterans, even if the patient was enrolled in hospice care by a VA provider.²⁰ Including such costs would make Medicare look more expensive, even though services were often ordered by VA providers. As the provider ordering the care is unobservable in the administrative data, we dropped these categories of services from cost analyses to reduce bias in our Medicare and VA comparisons.

Costs were winsorized at the 99th percentile before being aggregated to monthly costs using day, system, and type of care.²¹ For example, Medicare-paid inpatient costs were winsorized at their 99th percentile per day, whereas VA-paid inpatient costs were winsorized at their 99th percentile per day. We adjusted all costs to 2014 dollars using the recommended Personal Consumption Expenditure Index.²²

Patients were allocated to the healthcare system upon which they were most reliant (Medicare, VA, or Mixed) for inpatient and outpatient care in the last year of life. Reliance was assessed as the number of days with a Medicare encounter divided by the sum of days with a VA or Medicare encounter and ranged from 0 to 1.0 inclusive.⁵ Patients 75% or more reliant on Medicare were characterized as Medicare-reliant. Patients 75% or more reliant on VA were characterized as VA-reliant. The remainder were characterized as Mixed-reliant. Reliance calculations were limited to services patients could reasonably receive in either system. Referring to the example above, VA provides long-term nursing care and domiciliary care that Medicare does not reimburse; such care was excluded from the reliance (main effect) calculation in addition to the cost (outcome) calculation. We tested reliance using different specifications, including reliance before the last year of life, and limiting reliance to outpatient care only; patient group assignment was highly stable across these specifications. We also conducted sensitivity analyses in which we limited the cohort to persons 100% reliant on VA and those 100% reliant on Medicare.

VA-to-Medicare comparisons are further complicated by the fact that Medicare cost data are reimbursements for care provided, whereas VA cost data are produced by an activity-based cost accounting system. For this reason, absolute costs across Medicare and VA are not directly comparable, as VA allocates indirect costs across the specific number of patients seen in that care environment, whereas total costs for one Medicare patient do not depend on the number of other patients seen by the Medicare provider. Thus, our health system cost analysis focuses on differences in cost trajectories, rather than differences in absolute costs.

Models

Three-level generalized estimating equations were used to evaluate differences in monthly cost trajectories of Medicare-, VA-, and Mixed-reliant patients. Modified Park and Box-Cox tests suggested a gamma family and a log link. Patients were clustered within geographic area of hospital referral region.²³ Hospital referral region was modeled a fixed effect to account for both geographic differences in practice of end-of-life care^24,25 and geographic differences in wages. We centered this variable to allow for interpretation at an average hospital referral region.²⁶ Models incorporated a quadratic term for month (time) based on the shape of unadjusted cost trajectories. We tested multiple models, the most complex of which included three-way interaction terms of: geographic region, system, and month; and geographic region, system, and month-squared. A comparison of models revealed these interaction terms did not add explanatory power. Thus, final models included two-way interaction terms for system-by-month and system-by-month-squared. The full model is presented in equation 1. Patient covariates included age, cancer type, race, rurality, VA priority group, and distance from a VA facility. The latter four variables were included in the model as they are predictive of selection into VA or Medicare.^27–30 For example VA priority group indicates the level of cost-sharing a veteran has if he receives care in the VA system; many veterans have no cost-sharing due to their income, tour of duty or presence of a service-connected disability; this may make these veterans more likely to use VA for their healthcare needs. An instrumental variable of standardized differential distance to address selection bias³¹ was explored but not found suitably strong for inclusion. Comorbidities were assessed but not included in models due to research showing VA data under-capture comorbidities compared to Medicare data for the same patient.^32–34 However, we conducted sensitivity analyses in which we included Elixhauser and Charlson comorbidities as right-hand side variables. We predicted adjusted costs from regression models for an average hospital referral region, with separate models produced for total, inpatient, outpatient and drug costs.

$$\begin{gathered} {\mathrm{cost}}_{ijk}=a+b1{(\text{system})}_j+b2{(\text{month})}_{ij}+b3{\left({\text{month}}^2\right)}_{ij} \\ +b4({\text{month*system})}_{ij}+b\mathit{5}{\left({\text{month}}^2*\text{system}\right)}_{ij} \\ +b\mathit{6}{(\text{hospital}\text{referral}\text{region})}_k \\ +b\mathit{7}{(\text{vector}\text{of}\text{patient}\text{covariates})}_j+e_{ijk} \end{gathered}$$ (1)

where $i$ = month; $j$ = patient; $k$ = hospital referral region.

RESULTS

Our cohort consisted of 36,401 veterans with an average age of 79 years (SD 7.74), of which 28.4% were VA-reliant, 20.7% were Mixed-reliant and 50.9% were Medicare-reliant. In our cohort, 15.3% percent of VA-reliant veterans were VA-copayment eligible, compared with 27.2% of Mixed-reliant and 47.9% of Medicare-reliant veterans (Table 1). VA-reliant veterans were more likely to be older; Medicare-reliant veterans were more likely to be younger. Veterans who were VA-reliant were more likely to be Black (14.2%) compared to those who were Medicare-reliant (4.7%).

Table 1.

Demographics

Variable	Total, n (%)	Medicare-reliant, n (%)	Mixed-reliant, n (%)	VA-reliant n (%)
Cohort size	36,401 (100)	18,542 (50.9)	7,518 (20.7)	10,341 (28.4)
Age
66–70	6,808 (18.7)	2,290 (12.4)	1,257 (16.7)	3,259 (31.5)
71–75	5,329 (14.6)	2,525 (13.6)	1,044 (13.9)	1,760 (17.0)
76–80	7,481 (20.6)	4,166 (22.5)	1,444 (19.2)	1,871 (18.1)
81–85	7,965 (21.9)	4,689 (25.3)	1,636 (21.8)	1,640 (15.9)
85–90	6,544 (18.0)	3,691 (19.9)	1,557 (20.7)	1,296 (12.5)
91 +	2,276 (6.3)	1,181 (6.4)	580 (7.7)	515 (5.0)
Race
American Indian/Alaska Native	109 (0.3)	40 (0.2)	35 (0.5)	34 (0.3)
Asian/Pacific Islander	336 (0.9)	157 (0.9)	78 (1.0)	101 (1.0)
Black	3,129 (8.6)	875 (4.7)	787 (10.5)	1,467 (14.2)
Mixed	207 (0.6)	73 (0.4)	48 (0.6)	86 (0.8)
White	32,363 (88.9)	17,278 (93.2)	6,511 (86.6)	8,574 (82.9)
Missing	257 (0.7)	119 (0.6)	59 (0.8)	79 (0.8)
Rural status
Highly urban	15,145 (41.6)	7,607 (41.0)	2,882 (38.3)	4,656 (45.0)
Rural	15,753 (43.3)	7,928 (42.8)	3,459 (46.0)	4,366 (42.2)
Urban	5,458 (15.0)	2,989 (16.1)	1,162 (15.5)	1,307 (12.6)
Missing	45 (0.1)	18 (0.1)	15 (0.2)	12 (0.1)
VA copayment eligible	12,350 (33.9)	8,878 (47.9)	1,892 (25.2)	1,580 (15.3)
Cancer cause of death
Bladder	2,489 (6.9)	1,381 (7.5)	479 (6.4)	629 (6.1)
Brain	509 (1.4)	310 (1.7)	107 (1.4)	92 (0.9)
Colorectal	3,567 (9.8)	1,746 (9.4)	668 (8.9)	1,153 (11.2)
Gastroesophageal	1,926 (5.3)	973 (5.3)	363 (4.8)	590 (5.7)
Head/neck	1,224 (3.4)	461 (2.5)	260 (3.5)	503 (4.9)
Hepatobiliary	1,385 (3.8)	610 (3.3)	347 (4.6)	428 (4.1)
Kidney	11,312 (3.6)	663 (3.6)	286 (3.8)	363 (3.5)
Lung	12,128 (33.3)	6,017 (32.5)	2,443 (32.5)	3,668 (35.5)
Melanoma	946 (2.6)	513 (2.8)	197 (2.6)	236 (2.3)
Pancreas	1,778 (4.9)	1,008 (5.4)	351 (4.7)	419 (4.1)
Prostate	9,137 (25.1)	4,860 (26.2)	2,017 (26.8)	2,260 (21.9)

Total Costs

Compared to VA-reliant veterans, Medicare-reliant and Mixed-reliant veterans had significantly different trajectories for total health system costs, as evidenced by significant values for the Mixed-reliant * Month, Medicare-reliant * Month, and Medicare-reliant Month² terms (P = .031, P < .001, and P < .001, respectively) (Table 2). However, visual inspection of adjusted health system cost trajectories reveals total costs were similar across all three groups of veterans until the last 3 months of life, when trajectories for VA-reliant increased faster than that for Medicare-reliant veterans (Figure 1). Across all three groups, total healthcare costs rose most sharply in the last 4 months of life, largely driven by inpatient costs.

Table 2.

Coefficients from Generalized Estimating Equation Models, Health System Cost Trajectories^a

Variable	Health system costs, beta coefficients (95% confidence intervals)
	Total	Inpatient	Outpatient	Drug
Month	−0.30** (−0.31, −0.29)	−0.51** (−0.54, −0.49)	−0.02** (−0.03, −0.13)	0.19** (0.17, 0.20)
Month2	0.02** (−0.02, −0.02)	0.03** (0.03, 0.03)	0.0003 (−0.0003, 0.0009)	−0.01** (−0.01, −0.01)
Mixed-reliant	−0.04 (−0.10, −0.02)	−0.11 (−0.24, 0.03)	−0.11** (−0.16, −0.07)	0.27** (0.17, −0.38)
Mixed-reliant * month	−0.02* (−0.04, −0.002)	0.27 (−0.02, −0.07)	−0.05** (−0.06, −0.03)	−0.08** (−0.11, −0.05)
Mixed-reliant * month2	0.001 (−0.005, 0.003)	−0.002 (−0.003, −0.006)	0.003** (0.002, 0.004)	0.005** (0.003, 0.007)
Medicare-reliant	−0.10** (−0.15, −0.06)	−0.12* (−0.22, −0.02)	−0.31** (−0.35, −0.28)	0.71** (0.64, 0.78)
Medicare-reliant * month	0.63** (0.05, 0.80)	0.06** (0.03, 0.09)	0.0009 (−0.01, 0.01)	−0.006 (−0.028, 0.150)
Medicare-reliant * month2	−0.004** (−0.005, −0.003)	−0.004* (−0.007, −0.002)	−0.0001 (−0.001, 0.001)	0.0004 (−0.001, 0.002)

^*P < .05.

^**P < .001.

^aPredicted results from models adjusting for age, race, distance from VA facility, rurality, enrollment priority, and type of solid tumor, and conditioning on geographic region. P-values of less than .05 for one or both of the interaction terms for health system × month or health system × month² indicate significant differences in cost trajectories compared with VA-reliant patients.

Figure 1.

Health system costs in the last year of life for VA-reliant, Medicare-reliant, and mixed-reliant patients.

Inpatient Costs

Medicare-reliant veterans and VA-reliant veterans had significantly different monthly inpatient cost trajectories, as indicated by significant p-values for the Medicare-reliant * Month and Medicare-reliant * Month² terms (P < .001 and P = .002, respectively) (Table 2). There were no significant differences in the cost trajectories of Mixed-reliant versus VA-reliant veterans. Nonetheless, visual examination of plotted data revealed similar trajectories across the three groups in much of the last year of life (Figure 1). In all three groups, inpatient costs remained relatively flat in the 12–7 months before death, began growing in the 7–4 months before death, and experienced a sharp increase in the 3 months before death.

Outpatient Costs

There were significant differences in the monthly outpatient cost trajectories of Mixed-reliant versus VA-reliant veterans, as indicated by the Mixed-reliant * Month and Mixed-reliant * Month² terms (both P < .001) (Table 2). There were not significant differences in the outpatient cost trajectories of Medicare-reliant versus VA-reliant veterans. Outpatient health system trajectories for VA-reliant and Medicare-reliant veterans were similar, whereas outpatient health system cost trajectories for Mixed-reliant veterans revealed a distinct pattern (Figure 1). For both VA-reliant and Medicare-reliant veterans, outpatient costs had a slow but steady increase in costs as patients neared death. However, for mixed-reliant patients, outpatient costs increased more in the 5 months before death than they did for the other two groups.

Drug Costs

There were significant differences in the monthly drug cost trajectories of Mixed-reliant versus VA-reliant veterans, as evidenced by the Mixed-reliant * Month and Mixed-reliant * Month² terms (both P < .001) (Table 2). There were not significant differences in the outpatient cost trajectories of Medicare-reliant versus VA-reliant veterans. A visual examination of plots reveals all groups experienced an apex of drug costs 7–8 months before death, with drug costs declining as patients neared death (Figure 1). Fluctuations in drug costs in the last year of life were most marked in Medicare-reliant patients.

Sensitivity Analyses

In models including Fee Basis costs as a part of VA costs, cost trajectories for total, inpatient, drug and outpatient costs were significantly different across Medicare-reliant versus VA-reliant veterans (all P < .001) and across Mixed-reliant versus VA-reliant veterans (all P < .001) (Supplementary Table S1). In models including comorbidities, although the magnitude of coefficients, and in some cases, the significance of certain coefficients, changed, the conclusions drawn regarding health system differences remain the same (Supplementary Table S2). Specifically, there remained significant differences in total cost trajectories across Medicare-reliant versus VA-reliant veterans and across Mixed-reliant versus VA-reliant veterans. For inpatient cost trajectories, there remained no differences between Mixed-reliant versus VA-reliant veterans, and significant differences between Medicare-reliant versus VA-reliant veterans. For trajectories outpatient and drug costs, there remained significant differences between Mixed-reliant and VA-reliant veterans, and no significant differences between Medicare-reliant and VA-reliant veterans. Lastly, in analyses where we limited the cohort to the 7,445 patients 100% reliant on VA or Medicare, we found no difference with respect to significance or direction of significant results (Supplementary Table S3, Supplementary Figure S1). Results for these sensitivity analyses are presented in the Supplementary Material.

DISCUSSION

Regardless of the health system in which they were treated, costs of care increased as cancer patients neared death, largely driven by steep increases in inpatient costs. Outpatient costs also increased as patients neared death, whereas intravenous and oral drug costs declined. The similar end-of-life cost patterns across Medicare and VA are notable given their dissimilar financial incentives and organizational structures. Traditional Medicare is a fee-for-service (FFS) environment with financial incentives for overuse of care, greater care fragmentation and limited availability of palliative care services. VA is an integrated system that during the time of our analyses was largely a direct provider of care. Unlike FFS Medicare providers, VA providers have no direct reward for over or underuse of care. Additionally, as an organization, the VA has made concerted efforts to improve quality of end-of-life care, including launching the Comprehensive End of Life Care Initiative and including receipt of palliative care before death as a performance measure for VA facilities. Nonetheless, we did not find evidence of reduced intensity of VA care as ascertained through cost trajectories for patients nearing death, although average spending levels (which are not assessable due to incompatible cost data) might differ.

That Medicare and VA both have rising cost trajectories in the last 12 months of life suggests that other, non-financial and non-health-system factors are at play. These may include challenges with providers ascertaining or communicating patients’ terminal status, patient/family preferences for intensive end-of-life care or patient/family lack of understanding of clinical information. However, there is increasing evidence that variation in end-of-life care is not well explained by patient preferences³⁵ or severity of illness²⁴ and is heavily driven by physician characteristics.²⁵ Our work contributes additional insight; our results showing similar cost trajectories across Medicare and VA suggests that end-of-life care is driven by factors other than payor type or financial incentives.

The body of evidence comparing VA and Medicare costs is scant, likely due to the challenges inherent in comparing costs across systems with a different basket of services and highly dissimilar cost estimation practices. The most recent peer-reviewed evaluations are also subject to these same limitations.^36–39 Indeed, the Congressional Budget Office has concluded that it could not properly compare absolute healthcare costs across the VA and the private sector.⁴⁰ However, our results echo other work on single-system evaluations. Medicare spending for cancer patients in Surveillance, Epidemiology, and End Results (SEER) hospitals increases at the end of life, with inpatient costs⁴¹ and aggressive care⁴² being the major drivers. Increasing inpatient cost trajectories at the end of life are also seen in commercially insured cancer decedents.⁴³ That our work replicates the findings of these older studies indicates the more-recent push toward quality measurement and reduction of intensive services in the last months of life for cancer patients leaves ample room for improvement. Other work comparing VA to Medicare with respect to quality of end-of-life care has found that VA outperforms Medicare; specifically, veterans treated by VA are less likely to receive intensive medical services in the last month of life that are considered to be a marker of overuse of care.⁵ Although previous work indicates VA outperforms Medicare when evaluating quality of care in the last 30 days of life, our results using a longer look-back period reveal that both systems increase the intensity of inpatient care as patients near death, a pattern which may be at odds with the focus on palliation that should characterize high-quality end-of-life care. Our cost trajectory models reveal that within the same patient, costs increase as patients near death; high end-of-life costs cannot be attributed to time-invariant characteristics such as patient race or geographic area. Our cost trajectory results contribute a new type of evidence to the literature and reveal opportunities for improvement in both systems that are otherwise not ascertainable through existing quality-of-care measures.

Across all three groups of patients, inpatient costs increased as patients neared death, rising in the last 3 months of life. Patients in this cohort had a diagnosis of solid tumor for 12 or more months before death; thus, in the last 3 months of life, a provider may have had some insight that his patient’s cancer was advancing¹⁶ and that aggressive care would not be helpful.^3,44 That inpatient costs instead rose sharply during this time for all three groups of patients is thus problematic. At the end of life, studies have found little correlation between intensity of care and care quality.^24,45 Increased use of inpatient services in the months before death does not only circumvent a focus on palliative and supportive care but can also exacerbate treatment-related financial burden, which is particularly prevalent for cancer patients.^35,46

As VA expands receipt of community care through the CHOICE and MISSION Acts, many veterans will shift from being VA-reliant to obtaining more care in the private sector, either in conjunction with or in place of VA care. Our work shows Mixed-reliant patients had increased outpatient costs before death compared to patients who received the majority of their care in one system. Further exploration is needed to understand why growth in outpatient costs in the last months of life is greater for these patients, and specifically whether this is due to patients’ care seeking behavior, duplication of services across care settings, or both. The effect of this increased outpatient care on care coordination also merits further exploration, especially as it relates to patients receiving preference-concordant care at the end of life.

Limitations

This work is subject to certain limitations. The largest limitation is that of selection bias, or systematic differences in patients seen by each healthcare system. We control for this in two ways: first, by limiting the cohort to veterans seeking care through Medicare versus veterans seeking care in VA; and, second, by adjusting for characteristics known to influence veteran care seeking in VA versus Medicare, including VA copayment status. Nonetheless, despite these adjustments, selection bias may remain, especially as we were not able to find a suitable instrumental variable. Second, this was a retrospective study of decedent veterans; providers treating these patients were not certain that these patients were going to die. To address this hindsight bias, we focused on cancer patients only, whose pattern of functional decline in the months before death is far steeper and more linear than that of other chronic, life-limiting illnesses,¹⁶ and limited the cohort to patients who died of solid tumor, for which it is easier to prognosticate death. We limited our cohort to patients who had cancer for at least 1 year and observe that most of the increases in cost appear in the 3 months before death; providers therefore had approximately 9 months to understand their patients’ disease trajectories before costs increased.⁴⁷

CONCLUSION

Our work indicates similar, and concerning, rising end-of-life cost trajectories in both VA and Medicare for dually enrolled veterans. Regardless of the environment in which veterans are treated, concerted efforts are warranted to reverse existing patterns of increasing spending at the end of life, largely driven by inpatient care. Improving end-of-life care in both VA and Medicare is needed to bring care in line with quality-of-care recommendations.

Supplementary Material

Supplementary Material

Supplementary Figure S1 Cost trajectories in the last year of life, for patients 100% reliant on VA or Medicare.

Supplementary Table S1. Coefficients from Generalized Estimating Equation Models, Health System Cost Trajectories, Subsuming Fee-Basis Costs into VA

Supplementary Table S2. Coefficients from Generalized Estimating Equation Models, Health System Cost Trajectories, Including Comorbidities

Supplementary Table S3. Coefficients from Generalized Estimating Equation Models, Health

System Cost Trajectories, for a Subset of Patients 100% Reliant on VA or Medicare (n = 7,445)

Additional Supporting Information may be found in the online version of this article.