Outpatient palliative care and end-of-life care intensity: linking Massachusetts Cancer Registry with All-Payer Claims Database

Abstract

Background

Early palliative care is associated with better outcomes for patients with advanced-stage cancers. Using a novel data linkage, we assessed outpatient palliative care use before death and its association with end-of-life care intensity and variation across 8 provider networks in Massachusetts.

Methods

We linked Massachusetts Cancer Registry and the All-Payer Claims Database for individuals with commercial insurance, Medicaid, or Medicare Advantage diagnosed with colorectal, lung, prostate, and breast cancers from 2010 to 2013 who died by December 31, 2014. We characterized outpatient palliative care visits in the 6 months before death and identified end-of-life hospitalizations, emergency department visits, intensive care unit admissions, chemotherapy, no/late hospice enrollment, and in-hospital deaths. We used logistic regression to assess factors associated with outpatient palliative care and ordinal logistic regression with provider network fixed effects to assess the association of palliative care with a composite measure summing individual end-of-life intensity measures.

Results

Among 6279 decedents, 11.3% had at least 1 outpatient palliative care visit. Palliative care use varied across provider networks from 6.0% to 19.3%. In adjusted analyses, younger age, longer duration from diagnosis to death, death in 2012-2014 vs 2010, and provider network were associated with palliative care visits (all P values less than .05). End-of-life care intensity varied across provider networks. Patients with palliative care visits had lower adjusted odds of receiving intensive end-of-life care (adjusted odds ratio = 0.62 per additional measure of end-of-life intensity, 95% CI = 0.53 to 0.72).

Conclusions

Outpatient palliative care use varied substantially among regional provider networks and was associated with less intensive end-of-life care.

Introduction

In October 2010, the Massachusetts Expert Panel on End-of-Life Care released recommendations to ensure that Massachusetts residents receive high-quality, patient-centered care delivered by a system that ensures excellence and accountability for that care.¹ This report was released shortly after the publication of a groundbreaking randomized clinical trial demonstrating that early integrated palliative care delivered in the outpatient setting led to important improvements in quality of life and mood, less intensive end-of-life care, and longer survival, compared with standard oncologic care.² In this trial, early palliative care also led to less chemotherapy in the last 60 days of life and more timely hospice enrollment.³ Oncology practices began incorporating palliative care clinicians into their practices, and palliative care clinicians—who had historically provided palliative care to seriously ill patients in the inpatient setting—have increasingly provided palliative care in outpatient settings. A systematic review found that palliative care delivered in the outpatient setting and integrated with oncology care improves patients’ quality of life and cancer-related symptoms.⁴

In this study, we first examined the use of outpatient palliative care for patients with advanced cancers of 4 types at the time when the benefits of palliative care were becoming increasingly evident. Second, we assessed differences in the intensity of end-of-life care for cancer across 8 large Massachusetts provider networks that may differ in how they prioritized availability of palliative care services. Finally, we examined the extent to which access to outpatient palliative care across provider networks explained this variability. To achieve these aims, we conducted a novel linkage of data from the Massachusetts Cancer Registry with the Massachusetts All-Payer Claims Database (APCD), which may serve as a model to better understand the delivery of end-of-life care for individuals with cancer in Massachusetts and elsewhere covered by commercial insurance, Medicare Advantage, or Medicaid.

Methods

Data and study population

We obtained data from the Massachusetts Cancer Registry on all 78 518 cancers of the colon/rectum, lung, prostate, and female breast diagnosed in Massachusetts from 2010 to 2013. From the Massachusetts APCD, we obtained insurance data from 2010 to 2014 (Release 4.0), including medical and pharmacy claims and enrollment data for individuals aged 18 years and older with commercial insurance, Medicaid, and Medicare Advantage.^5,6 As described previously,⁷ we used the Centers for Disease Control and Prevention’s Link Plus probabilistic matching software to link cancer registry data with APCD data, yielding matches in medical eligibility files for 73 013 (93%) of 78 518 patient cases.

From these 73 013 breast, colon/rectum, lung, or prostate cancers diagnosed in 2010-2013, we identified patients who died by December 31, 2014 (ascertained from cancer registry data). To identify patients who were likely to have been treated for active cancer at the time of their death, we focused on patients who had advanced-stage/metastatic cancer (stage IIIB/IV for lung cancer and stage IV for other cancers) at diagnosis or had cancer listed as the cause of death, for a total of 7147 patients. We excluded 230 patients for whom no physician could be identified, 534 patients with no inpatient or outpatient visits, 104 patients with no cancer-related claims in the 6 months before death, and 17 patients missing ZIP codes, yielding 6262 decedents with advanced cancer. Because we lacked fee-for-service Medicare claims and thus could not observe hospice use for Medicare Advantage enrollees (which is billed under the fee-for-service Medicare benefit), analyses of hospice use were restricted to the 2947 (47.1%) patients aged younger than 65 years.

To understand care across provider networks in Massachusetts, we attributed patients to physicians and networks based on evaluation and management visits with physicians. We used a hierarchy to identify the physician most likely to be leading the patients’ cancer-directed care, assigning patients to physicians based on outpatient visits with a cancer diagnosis in the 6 months before death (or hospice enrollment among patients enrolled in hospice) in the following order: medical oncologist, radiation oncologist, urologist (prostate cancer patients only), primary care provider, medical specialist, surgical oncologist, other surgeon (general, thoracic); we used inpatient visits for 642 patients with no outpatient visits. If the patient had equal numbers of visits with 2 or more physicians in the specialty of interest, we selected the physician with visits closer to death. We used the Massachusetts Health Quality Partners Massachusetts Provider Database⁸ to assign 4285 of the 6262 patients to a network; the remaining 1977 patients were not attributed to a network and were considered “unassigned.” Sixty-one patients (<1%) had physicians who provided care in more than one network and were randomly assigned to one.

Palliative care visits

We identified outpatient visits with palliative care physicians in the last year of life using the specialty code on the visit (code 17 for hospice and palliative care) or if the visit was with a physician whose primary or secondary specialty was palliative care in the provider claims. We also considered visits with an International Classification of Diseases (ICD)-9 diagnosis code of V66.7 (“encounter for palliative care”) to be palliative care visits. We focused on outpatient palliative care visits because inpatient palliative care consultations are typically provided in response to an acute event.

Intensity of end-of-life care

We documented 6 measures of intensity of end-of-life care, including intensive care unit admission, hospitalizations in the last 30 days of life, 2 or more emergency department visits in the last 30 days of life, chemotherapy use in the last 14 days of life, no hospice enrollment or enrollment within 3 days before death, and death in a hospital. These claim-based measures were developed based on focus groups with patients and clinicians^9-11 and are associated with better family perceptions of care at the end of life¹²; they were previously endorsed by the National Quality Forum.¹³ In addition, after documenting correlations across measures, we created a patient-level composite measure by assessing the presence or absence of each of the 6 end-of-life intensity measures except hospice enrollment (which could not be ascertained reliably for individuals aged ≥65 years); this measure ranged from 0 (none of the measures of end-of-life intensity) to 5 (all measures of end-of-life intensity). In addition, we created a composite that included the hospice measure for individuals younger than 65 years, ranging from 0 to 6.

Covariates

We adjusted analyses for variables that we believed might be associated with intensity of end-of-life care. Specifically, we characterized each patient’s age at death; sex; race and ethnicity ascertained from medical record review by cancer registrars and categorized as non-Hispanic White (White), non-Hispanic Black (Black), Hispanic, other (Asian/Pacific Island, Native American, Alaska Native, “other”) race; marital status; cancer site; stage at diagnosis; number of months from diagnosis until death; and year of death based on cancer registry data. We characterized comorbidity in the year before death using claims data diagnosis codes based on the Klabunde version of the Charlson Comorbidity Index.¹⁴ We obtained ZIP code-level measures of the proportion of residents with a high school degree, median household income, and the proportion living in poverty from the American Community Survey. We also adjusted for provider network. Variables were categorized as in Table 1.

Table 1.

Study population characteristics and association with outpatient palliative care, n = 6262.

Variable	Study population characteristics	Unadjusted No. (%) with outpatient palliative care visits (or mean [SD] for patients with outpatient palliative care visits)	Un-adjusted P-value	Adjusted odds ratio (95% CI)a	Adjusted P-value
All who died, No. (%)	6262 (100)	712 (11.3)	—	—	—
Age at death, mean (SD)	68.0 (13.1)	63.5 (12.8)	<.001	0.98 (0.97 to 0.99)	<.001
Sex, No. (%)
Men	3073 (49.1)	345 (11.2)	.73	N/Ab	N/Ab
Women	3189 (50.9)	367 (11.5)		N/Ab
Race and ethnicity, No. (%)
Black	243 (3.9)	29 (11.9)		0.95 (0.62 to 1.46)	.82
Hispanic	113 (1.8)	22 (19.5)		1.48 (0.89 to 2.46)	.13
Otherc	110 (1.8)	17 (15.5)		1.12 (0.65 to 1.93)	.69
White	5796 (92.6)	644 (11.1)	.02	1.0	Reference
Marital status at diagnosis, No. (%)
Married/partnered	3040 (48.5)	378 (12.4)	.002	1.0	Reference
Separated/divorced	827 (13.2)	93 (11.2)		0.84 (0.66 to 1.08)	.17
Widowed	1049 (16.8)	85 (8.1)		0.89 (0.69 to 1.16)	.40
Single	1042 (16.6)	129 (12.4)		0.88 (0.70 to 1.11)	.27
Unknown/missing	304 (4.9)	27 (8.9)		0.77 (0.51 to 1.17)	.21
Year of death, No. (%)
2010	615 (9.8)	44 (7.2)	<.001	1.0	Reference
2011	1244 (19.9)	122 (9.8)		1.32 (0.92 to 1.90)	.14
2012	1552 (24.8)	184 (11.9)		1.52 (1.07 to 2.17)	.02
2013	1676 (26.8)	202 (12.1)		1.51 (1.06 to 2.16)	.02
2014	1175 (18.8)	160 (13.6)		1.57 (1.07 to 2.31)	.02
Number of months from diagnosis to death, mean (SD)	11.5 (11.1)	13.9 (10.2)	<.001	1.02 (1.01 to 1.03)	< .001
Cancer site, No. (%)
Lung	4466 (71.3)	508 (11.4)	.48	1.0	Reference
Breast	527 (8.4)	64 (12.1)		0.81 (0.60 to 1.11)	.19
Prostate	268 (4.3)	23 (8.6)		0.65 (0.41 to 1.02)	.06
Colorectal	1001 (16.0)	117 (11.7)		0.93 (0.74 to 1.17)	.55
Metastatic cancer at diagnosisd, No. (%)
No	1943 (31.0)	201 (10.3)	.09	1.0	Reference
Yes	4319 (69.0)	511 (11.8)		1.19 (0.98 to 1.44)	.08
Charlson Score, No. (%)
0	1709 (27.3)	229 (13.4)	<.001	1.23 (0.98 to 1.55)	.08
1	1672 (26.7)	208 (12.4)		1.25 (1.00 to 1.57)	.05
2	1058 (16.9)	113 (10.7)		1.15 (0.89 to 1.49)	.28
3+	1823 (29.1)	162 (8.9)		1.0	Reference
% High school graduation in ZIP code, mean (SD)	89.4 (7.6)	89.5 (7.2)	.77	1.01 (0.99 to 1.03)	.31
Median household income in ZIP code, mean (SD)	$70 681 ($24 758)	$71 875 ($26 326)	.23	1.05 (1.00 to 1.10)	.07
% Poverty in ZIP code, mean (SD)	9.6 (6.7)	9.8 (7.2)b	.39	1.02 (1.00 to 1.04)	.04
Provider networke, No. (%)
Partners Healthcare	1224 (19.5)	197 (16.1)	<.001	1.0	Reference
Lahey	191 (3.1)	25 (13.1)		0.81 (0.51 to 1.27)	.35
Atrius	402 (6.4)	24 (6.0)		0.37 (0.24 to 0.57)	<.001
Baycare	402 (6.4)	43 (10.7)		0.67 (0.47 to 0.97)	.03
UMass	436 (7.0)	84 (19.3)		1.32 (0.98 to 1.77)	.06
Steward	575 (9.2)	60 (10.4)		0.70 (0.51 to 0.96)	.03
BIDMC	635 (10.1)	56 (8.8)		0.50 (0.36 to 0.69)	<.001
NEQCA	420 (6.7)	33 (7.9)		0.47 (0.31 to 0.69)	<.001
Unassigned	1977 (31.6)	190 (9.6)		0.55 (0.44 to 0.68)	<.001

Abbreviations: BIDMC = Beth Israel Deaconess Medical Center; NEQCA = New England Quality Care Alliance; UMass = University of Massachusetts Medical Center.

a

Using logistic regression to adjust for all listed variables in the models.

b

Sex was not included in the adjusted model due to collinearity with cancer type.

c

Other race includes Asian/Pacific Island, Native American, Alaska Native, other.

d

Metastatic cancer at diagnosis includes stage IV cancers as well as lung cancers that are stage IIIB.

e

During the study period, Partners Healthcare was a health system with 2 large academic medical centers (Brigham and Women’s Hospital and Massachusetts General Hospital) as well as several community hospitals and a community-based network of primary care physicians primarily in eastern Massachusetts; Lahey was a health system that included a large hospital and community-based physicians primarily in eastern Massachusetts; Atrius was a large independent physician group in the greater Boston area; Baycare was an alliance of the medical staff and Baystate Health hospitals providing care in western Massachusetts; UMass was a health system in central Massachusetts that included an academic medical center (UMass Memorial Medical Center), community hospitals, and their physicians; Steward was a for-profit integrated delivery system that grew from the purchase of a network of Catholic hospitals in eastern Massachusetts and their physician network; BIDMC was a health system that included a large academic medical center (Beth Israel Deaconess Medical Center), several community hospitals, and a community-based network of physicians; NEQCA was a network of private practice and employed physicians throughout Massachusetts who contracted together with insurers as an independent practice association.

Statistical analyses

We used χ² tests for unadjusted analyses and logistic regression models to assess patient characteristics and provider networks associated with outpatient palliative care visits. We described each measure of intensive end-of-life care stratified by receipt or no receipt of outpatient palliative care and used patient-level logistic regression models to assess the association of palliative care with each end-of-life care measure, adjusting for all variables in Table 1 (except sex due to collinearity with cancer type) and including provider network fixed effects.

We next summarized each end-of-life measure and the composite measures across provider networks. Finally, we used ordinal logistic regression to assess the association of patient characteristics and provider networks with the composite measures of intensive end-of-life care, first among all patients (excluding the hospice measure, which could not be assessed for all patients), and then using the composite measure that included hospice enrollment among patients aged less than 65 years. We consider these models exploratory given the goals of providing a single summary measure; we found similar findings with alternative specifications of the composite measure, including Poisson and linear regression models as well as individual models for each end-of-life measure that contributed to the composite score. Results of these additional specifications were similar and are not presented.

We conducted several additional sensitivity analyses for our final ordinal logit model. First, we limited analyses to patients diagnosed with advanced cancers only (stage IV cancers and stage IIIB lung cancer). Second, to ensure equivalent follow-up time for all patients, we limited analyses to (1) only patients diagnosed in 2010-2013 who died within 1 year and (2) only patients diagnosed in 2010-2012 who died within 2 years. Third, to adjust for patient sex, we ran models after replacing cancer type with variables that considered both sex and cancer type: breast, prostate, female lung, male lung, female colorectal, and male colorectal. Missing data were rare; 304 patients with missing marital status at diagnosis were analyzed as a separate category; a small number of patients with missing race and ethnicity were categorized with other race. Two-sided P values of less than .05 were considered statistically significant. Analyses were conducted using SAS version 9.4. The study protocol was approved by Institutional Review Boards at the Massachusetts Department of Public Health and Harvard Medical School with a waiver of informed consent.

Results

Among the 6262 individuals in Massachusetts with commercial insurance, Medicaid, or Medicare Advantage who had cancer of the colon/rectum, breast, prostate, or lung diagnosed in 2010-2013 and died by the end of 2014, the median age was 66.0 years, 50.9% were women, and 92.6% were White (Table 1). These patients were attributed to 1442 physicians; 68.4% of patients were attributed to 1 of 8 provider networks in Massachusetts; the remaining 31.6% of patients were treated by an unattributed physician.

Overall, 712 patients (11.3%) had at least 1 outpatient visit with a palliative care specialist (Table 1); this utilization increased from 7.1% in 2010 to 13.6% in 2014 (P = .02). In unadjusted analyses, individuals who had palliative care visits were younger than those who did not. Hispanic patients (19.5%) were more likely to have palliative care visits than White patients (11.1%) and Black patients (11.9%). Outpatient palliative care visits were more frequent for patients with more recent deaths, less comorbidity, and longer time from diagnosis to death. Use of outpatient palliative care varied across provider networks, with proportions ranging from 6.0% to 19.2%. In adjusted analyses younger age, longer duration from diagnosis to death, death in 2012-2014 vs 2010, living in an area with higher poverty rates, Charlson Comorbidity Score of 1 (vs 3+), and provider network were statistically associated with having outpatient palliative care visits (Table 1).

The intensity of end-of-life care varied substantially across provider networks (Table 2). The proportion of patients with intensive care unit admissions ranged from 22.3% to 36.6%. The proportion with 2 or more emergency department visits in the last 30 days of life ranged from 10.5% to 21.4%. The proportion with 2 or more hospitalizations in the last 30 days of life ranged from 15.4% to 20.5%. The proportion of patients dying in the hospital ranged from 23.0% to 34.8%. The proportion with chemotherapy in the last 14 days of life ranged from 6.7% to 12.9%. Among individuals under 65 years of age, the proportion who did not enroll in hospice or enrolled within 3 days of death ranged from 39.8% to 55.7%. The composite scores ranged from 0 to 5; nearly half of patients (49.4%) had scores of 0 (Table S1). Among patients aged under 65 years, the score ranged from 0 to 6, with 33.2% having scores of 0 and another 21.1% having scores of 1 (Table S1). The composite score also varied across provider networks (Table 2).

Table 2.

Unadjusted proportions of end-of-life measures overall and by provider networks.

Provider network	Intensive care unit in last 30 days, No. (%)	2+ emergency department visits in last 30 days, No. (%)	2+ hospitalizations in last 30 days, No. (%)	Death in hospital, No. (%)	Chemo in last 14 days of life, No. (%)	Composite measure for all patients (excludes hospice measure), mean (SD)	No hospice or enrolled within 3 days of death (<65 years only)a, No. (%)	Composite measure for patients <65 years  (includes hospice measure), mean (SD)
Overall (n = 6262)	1708 (27.3)	1153 (18.4)	1129 (18.0)	1893 (30.2)	612 (9.8)	1.04 (1.27)	1342 (45.6)	1.65 (1.60)
Partners Healthcare, (n = 1224)	333 (27.2)	209 (17.1)	213 (17.4)	363 (29.7)	105 (8.6)	1.00 (1.22)	242 (40.9)	1.53 (1.52)
Lahey, (n = 191)	70 (36.6)	20 (10.5)	31 (16.2)	44 (23.0)	22 (11.5)	0.98 (1.24)	33 (39.8)	1.45 (1.51)
Atrius, (n = 402)	117 (29.1)	73 (18.2)	75 (18.7)	110 (27.4)	35 (8.7)	1.02 (1.30)	65 (45.1)	1.79 (1.71)
Baycare, (n = 402)	89 (22.1)	78 (19.4)	61 (15.2)	107 (26.6)	27 (6.7)	0.90 (1.22)	96 (48.7)	1.59 (1.57)
UMass, (n = 436)	113 (25.9)	74 (17.0)	76 (17.4)	118 (27.1)	55 (12.6)	1.00 (1.27)	91 (45.0)	1.58 (1.63)
Steward, (n = 575)	168 (29.2)	107 (18.6)	115 (20.0)	193 (33.6)	66 (11.5)	1.13 (1.26)	136 (55.7)	1.88 (1.59)
BIDMC, (n = 635)	190 (29.9)	122 (19.2)	127 (20.0)	221 (34.8)	74 (11.7)	1.16 (1.29)	129 (44.5)	1.62 (1.58)
NEQCA, (n = 420)	125 (29.8)	90 (21.4)	86 (20.5)	134 (31.9)	54 (12.9)	1.16 (1.38)	107 (48.9)	1.88 (1.77)
Unassigned, (n = 1977)	503 (25.4)	380 (19.2)	345 (17.5)	603 (30.5)	174 (8.8)	1.01 (1.26)	443 (45.6)	1.63 (1.59)

Abbreviations: BIDMC=Beth Israel Deaconess Medical Center; NEQCA=New England Quality Care Alliance; UMass=University of Massachusetts Medical Center.

a

Total n=2947 patients <65 years old at death.

Across all patients, those with outpatient palliative care visits had lower proportions of all measures of intensity of end-of-life care except chemotherapy in the last 14 days of life (Table 3). After adjustment for patient characteristics and provider network fixed effects, patients with palliative care visits had lower odds of intensive care unit admissions, frequent emergency department visits and hospitalizations in the last 30 days of life, in-hospital deaths, and no/late hospice use (P values all < .007). The adjusted odds of chemotherapy in the last 14 days of life also tended to be lower for patients with outpatient palliative care visits, although this finding was not statistically significant (P = .09).

Table 3.

Adjusted association of outpatient palliative care and individual end-of-life measures.

	Unadjusted proportion of each end-of-life measure among patients with or without outpatient palliative care			Adjusted oddsa  of outpatient palliative care visits associated with end-of-life measures
End-of-life measure	No outpatient, palliative care (%)	Outpatient palliative care (%)	P	Adjusted oddsa  of outpatient palliative care associated with end-of-life measures (95% CI)	P
Intensive care unit admission in last 30 days of life	28.0	21.6	<.001	0.70 (0.58 to 0.85)	<.001
2+ emergency department visits in last 30 days of life	19.0	14.3	.003	0.73 (0.59 to 0.92)	.007
2+ hospitalizations in last 30 days of life	18.8	12.6	<.001	0.63 (0.50 to 0.80)	<.001
Death in hospital	31.3	21.8	<.001	0.60 (0.49 to 0.72)	<.001
Chemotherapy in last 14 days of life	9.8	9.0	.47	0.79 (0.59 to 1.04)	.09
No/late hospice (<65 years old only)	47.3	34.7	<.001	0.67 (0.53 to 0.84)	<.001

a

Using patient-level logistic regression models, adjusting for all variables in Table 1 other than sex due to collinearity with cancer type, including fixed effects for provider networks.

Using ordinal logistic regression to examine the association of outpatient palliative care with the composite measures of intensity of end-of-life care, receipt of outpatient palliative care was associated with less intense end-of-life care for all patients for the composite measure that did not include hospice use (adjusted odds ratio [AOR] =0.62, 95% CI = 0.53 to 0.72) and among individuals aged <65 years for the model that included hospice use (AOR = 0.66, 95% CI = 0.54 to 0.81) (Table 4). The associations with other patient characteristics are presented in Table 4. Some variability across provider networks remained, although this variability changed little among models without and with inclusion of outpatient palliative care (Table S2). Results of sensitivity analyses that limited to advanced cancers only, that ensured similar follow-up time for all patients, and that included variables to adjust for sex by cancer type were similar to our primary findings (Table S3).

Table 4.

Ordinal logistic regression assessing the association of patient characteristics and provider networks with end-of-life composite measure.

	Composite outcome without hospice, all patients (n = 6262)	Composite outcome with hospice, patients <65 years (n = 2947)
Variable	Adjusted odds ratio (95% CI)	Adjusted odds ratio (95% CI)
Outpatient palliative care	0.62* (0.53 to 0.72)	0.66* (0.54 to 0.81)
Age at death, per year	0.98* (0.97 to 0.98)	0.98** (0.97 to 0.99)
Race/ethnicity
Black	1.36*** (1.06 to 1.74)	1.50** (1.14 to 1.97)
Hispanic	1.14 (0.82 to 1.63)	1.13 (0.78 to 1.64
Othera	1.40 (0.98 to 1.99)	1.51*** (1.02 to 2.24)
White	1.00	1.00
Marital status at diagnosis
Married/partnered	1.00	1.00
Separated/divorced	0.99 (0.85 to 1.14)	1.02 (0.85 to 1.22)
Widowed	0.78** (0.67 to 0.90)	0.89 (0.64 to 1.23)
Single	0.81** (0.71 to 0.93)	0.92 (0.78 to 1.09)
Unknown/missing	1.05 (0.84 to 1.31)	0.96 (0.69 to 1.32)
Year of death
2010	1.00	1.00
2011	0.83*** (0.69 to 0.99)	0.79 (0.61 to 1.04)
2012	0.90 (0.75 to 1.07)	0.78 (0.60 to, 1.01)
2013	0.88 (0.74 to 1.05)	0.87 (0.67 to 1.13)
2014	0.86 (0.70 to 1.05)	0.84 (0.63 to 1.13)
Number of months from diagnosis to death	0.97* (0.96 to 0.97)	0.96* (0.96 to 0.97)
Cancer site
Breast	1.21*** (1.01 to 1.46)	1.31*** (1.03 to 1.66)
Prostate	1.06 (0.83 to 1.36)	0.95 (0.65 to 1.39)
Lung	1.00	1.00
Colorectal	0.95 (0.83 to 1.09)	0.93 (0.77 to 1.14)
Metastatic cancer at diagnosisb
No	1.00	1.00
Yes	0.82* (0.74 to 0.92)	0.82* (0.69 to 0.96)
Charlson Score
0	0.80*** (0.70 to 0.92)	0.87 (0.72 to 1.05)
1	0.78* (0.69 to 0.89)	0.76** (0.63 to 0.92)
2	0.80** (0.69 to 0.93)	0.83 (0.66 to 1.04)
3+	1.00	1.00
% High school graduation in ZIP code	1.00 (0.99 to 1.01)	0.99 (0.98 to 1.01)
Median household income in ZIP code	1.02 (0.99 to 1.05)	1.04 (0.99 to 1.09)
% Poverty in ZIP code	1.01 (0.99 to 1.02)	1.01 (1.00 to 1.03)
Provider network
Partners Healthcare	1.00	1.00
Lahey	1.06 (0.79 to 1.41)	0.96 (0.63 to 1.46)
Atrius	0.95 (0.76 to 1.17)	1.15 (0.83 to 1.59)
Baycare	0.79*** (0.63 to 0.99)	0.94 (0.70 to 1.27)
UMass	0.97 (0.79 to 1.20)	0.99 (0.74 to 1.33)
Steward	1.19 (0.99 to 1.44)	1.34*** (1.02 to 1.76)
BIDMC	1.22*** (1.02 to 1.46)	1.07 (0.83 to 1.38)
NEQCA	1.15 (0.93 to 1.42)	1.29*** (0.98 to 1.72)
Unassigned	0.99 (0.86 to 1.14)	1.07 (0.89 to 1.29)

Abbreviations: BIDMC=Beth Israel Deaconess Medical Center; NEQCA=New England Quality Care Alliance; UMass=University of Massachusetts Medical Center.

a

Other race includes Asian/Pacific Island, Native American, Alaska Native, other.

b

Metastatic cancer at diagnosis includes stage IV cancers as well as lung cancers that were stage IIIB.

*

P < .001.

**

P < .01.

***

P < .05.

Discussion

Palliative care has become an important component of care for patients with advanced cancer. This study of outpatient palliative care using all-payer claims data linked to state cancer registry in Massachusetts had 3 key findings. First, among over 6000 patients with commercial insurance, Medicare Advantage, or Medicaid who died with advanced breast, colon/rectum, lung, or prostate cancer, use of outpatient palliative care nearly doubled from 7.1% in 2010 to 13.6% in 2014, although most patients did not receive this care. Second, among 8 regional provider networks in the state, unadjusted palliative care use varied more than 3-fold from 6.0% to 19.3%. Third, receipt of outpatient palliative was strongly associated with less intensive end-of-life care and greater use of hospice before death.

In 2012, the American Society of Clinical Oncology (ASCO) endorsed the use of palliative care for patients with metastatic lung cancer, who represented nearly three-quarters of our study cohort.¹⁵ Subsequent ASCO guidelines in 2017 and 2024 have reaffirmed and extended this recommendation to include all patients with advanced cancers.^16,17 Despite these recommendations, our study found that relatively few patients with advanced cancer in our later study period had evidence in insurance claims of outpatient palliative care, and these low proportions did not differ among the 4 cancer types studied.

Other studies have assessed palliative care use for individuals with advanced cancer using insurance claims data. In 2 related studies that analyzed Surveillance, Epidemiology, and End Results (SEER)-Medicare linked data, use of palliative care for traditional Medicare beneficiaries with metastatic lung cancer increased from approximately 3% in 2001 to 36.8% in 2015.^18,19 Another study of commercially insured patients aged 25-64 years with metastatic cancer also demonstrated increasing use of palliative care over time, with 18% of patients with a poor prognosis and 36% of patients with a very poor prognosis receiving palliative care services in 2016.²⁰ These studies included both inpatient and outpatient palliative care and did not assess variations across health-care systems.

Our study demonstrates the feasibility of linking cancer registry and APCD data to assess processes and outcomes of cancer care. Currently, more than 20 states require collection of all-payer claims, and APCDs are being considered in other states.²¹ Data linkages between state APCDs and cancer registries have the potential to greatly enhance our understanding of cancer care spending, utilization, quality, and outcomes. Nevertheless, our linkage was time consuming, limiting the recency of the data analyzed. The linkage required a new interagency agreement to allow 2 public agencies in the same state to share data. Also, although we expected that the APCD data would include traditional Medicare data, such data could not be obtained due to data sharing restrictions that were unresolved. Some other states, including Utah,^22,23 Colorado,²⁴ Rhode Island,²⁵ and Arkansas,²⁶ also have successfully linked cancer registry and APCDs. However, creating processes for sustainable linkages within and across states that are timely and include high-quality data will require additional funding and focused efforts by stakeholders, including policymakers, government agencies, payers, health-care organizations, clinicians, and patient advocacy groups.

Our study’s strengths includes the novel linkage of APCD data with commercial, Medicare Advantage, and Medicaid data—allowing population-based assessment of patients that have been infrequently studied. We evaluated the association of palliative care with end-of-life quality measures endorsed by the National Quality Forum. Moreover, our study period preceded the Supreme Court’s 2016 decision to exempt self-insured health plans²⁷ (which cover nearly two-thirds of commercially insured workers²⁸), and thus includes individuals in these plans.

Our study also has several limitations. First, we studied patients diagnosed with cancer in 2010-2013 who died by 2014, potentially limiting generalizability to patients more recently diagnosed with cancer. Nevertheless, evidence about the benefits of palliative care became more prominent during and since our study’s time period.⁴ Additional research will be important to understand recent trends. Second, we studied patients with commercial insurance, Medicare Advantage, and Medicaid in 1 state with low uninsurance rates.^29,30 Third, we used claims data and primary and secondary physician specialty codes to identify patients with outpatient palliative care visits, but these specialty codes may not have been fully accurate. Fourth, our composite measure of end-of-life care was not previously validated; however, findings assessing the association of outpatient palliative care with individual measures of end-of-life care were similar. Finally, we identified associations of palliative care with less intensive end-of-life care and greater hospice use, but our observational study design does not support causal interpretations of these findings. Moreover, intensive end-of-life care may be appropriate for some patients.³¹ As such, rates of these end-of-life care intensity measures should not be zero. Rather, these measures can be useful for comparing care over time or across health-care organizations.⁹

In conclusion, using a novel data linkage of cancer registry and APCD data for individuals with advanced cancer, we found that use of outpatient palliative care in Massachusetts rose steadily in the early 2010s, even as it varied substantially across regional provider networks. However, only about 1 in 7 patients who died with advanced cancer in 2014 received outpatient palliative care. Receipt of outpatient palliative care was strongly associated with lower intensity care at the end of life. Policymakers, health-care system leaders, and clinicians caring for patients with advanced cancer should ensure that all patients are offered and have access to palliative care services.

Acknowledgments

We acknowledge Alan M. Zaslavsky, PhD, for assistance with study design, Jennifer Cabrera Rodriguez for assistance with article preparation, and the Massachusetts Center for Health Information and Analysis as the source of the All-Payer Claims Database. The analyses, conclusions, interpretations, and recommendations drawn from these data are solely those of the authors and are not those of the Massachusetts Department of Public Health or the Massachusetts Center for Health Information and Analysis. The funder had no role in the design of the study, the collection, analysis, and interpretation of the data, the writing of the article, and the decision to submit the article for publication. N.L.K. and R.W. had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Author contributions

Nancy L. Keating (Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Supervision, Writing—original draft, Writing—review & editing), Joel S. Weissman (Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Writing—review & editing), Alexi A. Wright (Investigation, Methodology, Writing—review & editing), Robert Wolf (Data curation, Formal analysis, Methodology, Writing—review & editing), Susan Gershman (Data curation, Methodology, Project administration, Resources, Writing—review & editing), Richard Knowlton (Data curation, Methodology, Project administration, Writing—review & editing), and John Z. Ayanian (Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Supervision, Writing—original draft, Writing—review & editing).

Supplementary material

Supplementary material is available at JNCI Cancer Spectrum online.

Funding

This work was funded by the National Cancer Institute (R21CA172892) and the Herrick Trust for the Study of Cancer in Massachusetts.

Conflicts of interest

S.G. was previously employed (now retired) and R.K. is currently employed by the Massachusetts Department of Public Health. No other authors have any other potential conflicts of interest to report.

Data availability

The data underlying this article were analyzed under data use agreements with the Massachusetts Department of Public Health and the Massachusetts Center for Health Information and Analysis and cannot be shared with external investigators. However, other investigators can work with these organizations to obtain the original data sources.

Author notes

S. Gershman is retired.