Early Integration of Outpatient Palliative Care Among Adults With Advanced Cancer in a Safety-Net Health System: A Patterns of Care Analysis

Abstract

PURPOSE

Little is known about guideline-concordant, early integration of palliative care (PC) in the outpatient setting among patients with advanced cancer within a safety-net system. This study examined PC delivery patterns for patients seen in a large, urban safety-net system.

METHODS

Patients diagnosed with advanced-stage solid tumor and who had ≥1 outpatient oncology visit from January 2018 to July 2023 at Parkland Health were identified via electronic health record. Outcomes assessed included (1) receipt of PC referral ≤8 weeks after diagnosis, (2) receipt of any PC referral, and (3) PC visit completion. Multivariable logit models evaluated associations between key characteristics (age, race/ethnicity, gender, cancer type, preferred language, insurance, diagnosis year) and the outcomes.

RESULTS

Among 1,296 patients (44% female; 76% non-White), 55% received a referral. Of those referred, 46% patients were referred early (≤8 weeks). Two thirds of the referred patients completed a PC visit during the study period. In adjusted regression models, patients who were Black (v White; adjusted odds ratio [aOR], 0.52 [95% CI, 0.33 to 0.82]), Hispanic (aOR, 0.33 [95% CI, 0.18 to 0.59]), or had prostate cancer (v breast cancer; aOR, 0.27 [95% CI, 0.10 to 0.69]) had lower odds of receiving early referral. Ages 40-69 (v >80 years; lowest odds for 60 to <70, aOR, 0.41 [95% CI, 0.20 to 0.85]) and patients with gynecologic cancer (aOR, 0.14 [95% CI, 0.07 to 0.28]) had lower odds of receiving any PC referral. Females had higher odds of completing a PC visit (v males; aOR, 1.45 [95% CI, 1.01 to 2.08]).

CONCLUSION

Many patients did not receive an outpatient referral or received it late. Observed differences by race/ethnicity, cancer type, and age suggest the need for different interventions targeting PC delivery for underserved patients with cancer.

INTRODUCTION

Over the past decade, multiple studies have shown that early integration of palliative care (PC) services concurrent with active cancer treatment is associated with improved quality of life, reduced intensity of treatment, similar or improved survival, and cost savings.^1-5 PC in oncology began as hospice and end-of-life care and traditionally was available only to patients with cancer no longer receiving therapy with curative intent, late in the disease trajectory.^6,7 Although several oncology societies, including ASCO^8,9 and National Comprehensive Cancer Network,¹⁰ now recommend all patients with advanced-stage cancer receive early intervention by a PC team soon after diagnosis and during active cancer treatment, early PC use in cancer care in the outpatient setting remains low, often only occurring late in the course of disease.^11-15

CONTEXT

• Key Objective

• What are the factors associated with outpatient palliative care (PC) delivery (referrals and visits) for patients diagnosed with advanced cancer, cared for within a large, urban safety-net system?

• Knowledge Generated

• In this diverse sample of 1,296 patients (76% non-White), after adjusting for key characteristics including insurance type and non–English language preference, we found that subgroups such as Black and Hispanic patients were significantly less likely to receive an early PC referral (≤8 weeks after advanced cancer diagnosis). Patients diagnosed with pancreatic cancer were more likely to receive any PC referral during the study period, whereas younger patients and those diagnosed with gynecologic and prostate cancers were less likely to receive any PC referral.

• Relevance

• Findings demonstrate variability in early PC referral, and opportunities for interventions, such as clinical triggers, to support implementation of early PC connections in safety-net health systems.

Several studies identify even less use of outpatient PC among racial and ethnic minorities and individuals with lower socioeconomic status.^7,16 Safety-net health systems care for a disproportionate share of medically underserved individuals, including patients with advanced cancer.¹⁷ These patients traditionally experience challenges with access to care and lack of insurance and tend to present at later stages of disease.¹⁸

Few studies have characterized PC delivery in the outpatient setting among patients with advanced cancer within a safety-net health system. Previous studies are limited by outdated data, small sample sizes, and short time frames; only one has been conducted among patients with advanced cancer.^19-21 The purpose of this study was to examine patterns of PC delivery for patients with advanced cancer within a large, urban safety-net health system. We sought to characterize the receipt and timing of PC referrals as well as initial visit completion. We hypothesize that patients of racial/ethnic minority groups (eg, Blacks and Hispanics) would be less likely to access PC than White patients. Understanding this information will be critical for developing strategies that can optimally support clinicians to increase PC use and to maximize scarce resources, reduce disparities, and improve cancer outcomes in medically underserved populations.

METHODS

Study Setting and Population

We conducted a retrospective study among patients who received care at Parkland Health (Parkland), the sole safety-net cancer treatment provider for the under- and uninsured in Dallas County, Texas. All patient care is captured in the Parkland electronic health record (EHR). We identified patients age 18 years and older in the Parkland tumor registry, initially diagnosed with advanced-stage solid tumor at the time of diagnosis with a date of diagnosis between January 1, 2018, and December 31, 2022, provided by the registry, and who received their initial course of cancer treatment at Parkland. The tumor registry also reported cancer type and stage at diagnosis; we categorized stage into nonadvanced versus advanced (stage IIIb and higher for lung cancer, stage III and higher for pancreatic, stage IV for all others except for brain). Patients with hematologic malignancies were excluded because of widely varying treatment and symptom patterns depending on unmeasured changes in acuity after diagnosis,²² which make PC use difficult to compare with other cancers.

We then linked these tumor registry data with the Parkland EHR and included patients who had at least one outpatient oncology visit at Parkland between January 1, 2018, and July 31, 2023. Expanding the date range in this way allowed for at least seven full months after diagnosis to capture PC referrals and completed visits after diagnosis. As one of the main objectives of this study was to assess early PC use according to ASCO clinical practice guidelines,^8,9 we excluded patients who died ≤8 weeks of advanced cancer diagnosis to ensure exposure to the entire guideline period. We also excluded patients who had incomplete referral data (Appendix Fig A1).

Key Measures

Outcomes

The primary outcome of interest was timeliness of first PC referral after date of advanced cancer diagnosis. Using the EHR, we calculated time to PC referral as a binary variable (early v late) and defined early PC involvement as receipt of referral ≤8 weeks of advanced cancer diagnosis date in the subsample of patients who received a referral.^8,9 Additional outcomes of interest were receipt of any PC referral and PC visit completion during the study period. We defined receipt of any PC referral as the first referral ordered after diagnosis date of advanced cancer during the study time frame. We defined visit completion as the initial completed PC visit after outpatient referral.

Explanatory Variables

Sociodemographic and clinical characteristics were obtained from the tumor registry and included age at cancer diagnosis, race/ethnicity (Hispanic, Black, White, and other), sex (male, female), cancer type (breast, colorectal, pancreas, gynecologic, head/neck, kidney, lung, other GI, prostate, and other/unknown), and year of cancer diagnosis. The Parkland EHR contributed non–English language preference and insurance type at time of diagnosis (charity/self-pay/Medicaid, commercial/Medicare). Since eligibility for charity assistance and Medicaid is contingent on specific criteria, including household income and immigration status, we considered these already included and did not include those specifics again out of concern for collinearity.^23-25

Statistical Analysis

We first grouped PC referrals and visits into a patient-level data set, and the population was summarized with descriptive statistics. We used chi-square tests to assess bivariate associations with time to PC referral, receipt of any PC referral, and visit completion across sociodemographic and clinical characteristics. We used multivariable logit regression models to identify key sociodemographic (age, race/ethnicity, sex, non–English language preference, insurance) and clinical explanatory variables (cancer type, year of diagnosis) that may be associated with our primary outcome (timing of PC referral within 8 weeks of diagnosis) and secondary outcomes of interest (any PC referral and visit completion). For each sociodemographic and clinical explanatory variable examined, the remaining variables were adjusted for in the models. To improve interpretability, marginal effects were calculated to estimate adjusted predicted probability of PC referrals and visits across key variables of interest.

Sensitivity Analyses

We performed several sensitivity analyses to test the robustness of our findings. First, we reran our analyses on the outcomes of any PC referral and PC visit completed including patients who died within 8 weeks of advanced cancer diagnosis. Second, we performed additional analyses examining time to PC referral using ≤12 weeks of advanced cancer diagnosis date as the cutoff for early PC. Finally, we included a binary indicator for prepandemic (January 1, 2018-February 28, 2020) and postpandemic (March 1, 2020-July 31, 2023) time periods in the model to examine the impact of the COVID-19 pandemic on receipt of any PC referrals. All analyses were conducted using SAS 9.4 software and Stata/MP version 15.1. A two-sided P value of <.05 was considered statistically significant for all analyses. We did not adjust for multiple comparisons due to the exploratory nature of the analyses. The study was approved by the institutional review board at the University of Texas Southwestern Medical Center.

RESULTS

Sample Characteristics

A total of 1,296 eligible individuals were available for analysis (Appendix Fig A1). The mean age was 58.2 (range, 21-94), 43.9% were female, 42.2% were Hispanic, and 33.3% were Black. More than one third (36.7%) had non–English language preference and 75.9% were uninsured or enrolled in Medicaid at the time of diagnosis. The most common cancer type was lung (19.1%), followed by other GI (15.9%), colorectal (14.4%), prostate (10.3%), head/neck (10.0%), breast (9.3%), pancreatic (7.6%), and gynecologic (5.9%). Just over half (55%) of patients received any outpatient PC referral during the study period (Table 1).

TABLE 1.

Study Population

Variable	All Patients	Referred to PC, No. (%)	Not Referred to PC, No. (%)	P
Total, No. (%)	1,296	725 (55.0)	571 (45.0)	.52
Male	727 (56.1)	401 (55.2)	326 (44.8)
Female	569 (43.9)	324 (56.9)	245 (43.1)
Age, years, mean (SD)	58.2 (11.2)			.33
<40, No. (%)	86 (6.6)	57 (66.3)	29 (33.7)
40 to <50, No. (%)	161 (12.4)	88 (54.7)	73 (45.3)
50 to <60, No. (%)	450 (34.7)	253 (56.2)	197 (43.8)
60 to <70, No. (%)	419 (32.3)	225 (53.7)	194 (46.3)
70 to <80, No. (%)	134 (10.3)	73 (54.5)	61 (45.5)
≥80, No. (%)	46 (3.5)	29 (63.0)	17 (37.0)
Race/ethnicity, No. (%)				.43
Hispanic	547 (42.2)	292 (53.4)	255 (46.6)
Black	431 (33.3)	248 (57.5)	183 (42.5)
White	261 (20.1)	148 (56.7)	113 (43.3)
Other/unknown	8 (0.6)	5 (62.5)	3 (37.5)
Language preference, No. (%)				.12
English	820 (63.3)	472 (57.6)	348 (42.4)
Non-English	476 (36.7)	253 (53.2)	223 (46.8)
Insurance type, No. (%)
Charity/self-pay/Medicaid	985 (76.0)	561 (57.0)	424 (43.0)	.21
Commercial/Medicare	310 (23.9)	164 (52.9)	146 (47.1)
Cancer type, No. (%)				<.001
Lung	247 (19.1)	156 (63.2)	91 (36.8)
Colorectal	187 (14.4)	105 (56.1)	82 (43.9)
Prostate	134 (10.3)	52 (38.8)	82 (61.2)
Head/neck	129 (10.0)	60 (46.5)	69 (53.5)
Breast	121 (9.3)	71 (58.7)	50 (41.3)
Pancreas	99 (7.6)	78 (78.8)	21 (21.2)
Gynecologic	77 (5.9)	14 (18.2)	63 (81.8)
Kidney	51 (3.9)	33 (64.7)	18 (35.3)
Other GI	206 (15.9)	132 (64.1)	74 (35.9)
Other/unknown	45 (3.5)	24 (53.3)	21 (46.7)
Diagnosis year, No. (%)				<.01
2018	292 (22.5)	190 (65.1)	102 (34.9)
2019	260 (20.1)	152 (58.5)	108 (41.5)
2020	285 (22.0)	155 (54.4)	130 (45.6)
2021	254 (19.6)	128 (50.4)	126 (49.6)
2022	205 (15.8)	100 (48.8)	105 (51.2)
Days from advanced disease diagnosis to referral, No. (%)
Mean (SD)	NA (NA)	172.7 (247.98)	NA (NA)
Median (IQR)	NA (NA)	65 (23-231)	NA (NA)

NOTE. Other GI cancers = esophageal, gallbladder, gastric, liver, rectal.

Abbreviations: NA, not applicable; PC, palliative care; SD, standard deviation.

Outpatient PC Referrals

Among patients referred to PC, 332 (45.8%) patients were referred early (≤8 weeks of diagnosis), versus 393 (54.2%) patients who were referred late (>8 weeks after diagnosis). In all, 56.3% were referred within 12 weeks (Appendix Fig A2). The median time from date of advanced disease diagnosis to any PC referral was 65 days (IQR, 23-231 days). The bivariate analyses of timeliness of PC indicated that early versus late referral varied significantly according to race/ethnicity and cancer type (P < .05; Appendix Table A1). In the multivariable logit regression analysis, predictors of receiving a late PC referral (>8 weeks) were patient age 60 to <70 years (adjusted odds ratio [aOR], 0.37 [95% CI, 0.14 to 0.97]), Black patients (aOR, 0.52 [95% CI, 0.33 to 0.82]), Hispanic patients (aOR, 0.33 [95% CI, 0.18 to 0.59]), other/unknown race/ethnicity (aOR, 0.39 [95% CI, 0.18 to 0.88]), and patients diagnosed with prostate cancer (aOR, 0.27 [95% CI, 0.10 to 0.70]; Table 2). Sensitivity analysis of using ≤12 weeks as the cutoff slightly changed these adjusted ORs of PC referral, but no changes in statistical significance were found. Although the statistical significance for specific age groups varied somewhat across the models, the adjusted ORs remained in the same direction.

TABLE 2.

Sample Characteristics Associated With Early Referral, Any Referral, and Visit Completion (multivariable analysis)

Variable	Early Referral (≤8 wks of diagnosis; n = 725)			Any Referral (N = 1,295)			Visit Completion (n = 725)
	OR	95% CI	P	OR	95% CI	P	OR	95% CI	P
Female (ref = male)	0.94	0.66 to 1.33	.72	1.18	0.89 to 1.56	.24	1.45	1.01 to 2.08	.04
Age (ref = ≥80)
<40	0.42	0.14 to 1.29	.13	0.74	0.30 to 1.82	.51	0.52	0.17 to 1.63	.26
40 to <50	0.39	0.13 to 1.12	.08	0.43	0.19 to 0.98	.04	0.89	0.30 to 2.66	.84
50 to <60	0.38	0.14 to 1.05	.06	0.43	0.20 to 0.93	.03	0.55	0.20 to 1.51	.24
60 to <70	0.37	0.14 to 0.97	.04	0.41	0.20 to 0.85	.02	0.46	0.18 to 1.20	.11
70 to <80	0.68	0.26 to 1.75	.42	0.54	0.26 to 1.12	.10	0.56	0.22 to 1.43	.22
Race/ethnicity (ref = White)
Black	0.52	0.33 to 0.82	.004	0.96	0.69 to 1.34	.81	0.84	0.53 to 1.33	.45
Hispanic	0.33	0.18 to 0.59	<.001	0.86	0.57 to 1.32	.50	0.58	0.32 to 1.04	.07
Other/unknown	0.39	0.18 to 0.88	.02	1.26	0.65 to 2.42	.50	0.57	0.25 to 1.27	.17
Language preference (ref = English)
Non-English	1.15	0.69 to 1.94	.59	0.85	0.58 to 1.23	.38	0.98	0.59 to 1.63	.93
Insurance type (ref = commercial/Medicare)
Charity/self-pay/Medicaid	1.12	0.66 to 1.92	.67	1.40	0.95 to 2.06	.09	1.58	0.93 to 2.69	.09
Cancer type (ref = breast)
Lung	1.36	0.72 to 2.59	.35	1.31	0.80 to 2.14	.29	1.12	0.58 to 2.15	.74
Colorectal	1.23	0.63 to 2.42	.54	0.97	0.58 to 1.62	.91	1.10	0.55 to 2.20	.79
Prostate	0.27	0.10 to 0.70	.01	0.51	0.28 to 0.93	.03	2.11	0.85 to 5.23	.11
Head/neck	0.46	0.20 to 1.03	.06	0.65	0.37 to 1.13	.13	1.62	0.72 to 3.65	.25
Pancreas	1.65	0.81 to 3.38	.17	2.86	1.50 to 5.44	.001	1.43	0.68 to 3.02	.35
Gynecologic	0.32	0.08 to 1.31	.11	0.14	0.07 to 0.28	<.0001	0.63	0.19 to 2.09	.45
Kidney	1.65	0.68 to 4.05	.27	1.46	0.71 to 2.99	.31	1.20	0.48 to 3.03	.69
Other GI	1.74	0.92 to 3.29	.09	1.40	0.85 to 2.30	.19	1.19	0.62 to 2.29	.61
Other/unknown	1.54	0.56 to 4.22	.40	0.78	0.38 to 1.61	.50	1.06	0.38 to 2.97	.92
Diagnosis year (ref = 2022)
2018	0.92	0.55 to 1.54	.76	2.06	1.41 to 3.02	<.001	1.67	0.99 to 2.82	.05
2019	0.75	0.44 to 1.28	.30	1.60	1.09 to 2.36	.02	1.16	0.68 to 1.97	.59
2020	1.45	0.85 to 2.47	.17	1.33	0.91 to 1.93	.14	1.45	0.85 to 2.49	.17
2021	0.73	0.42 to 1.28	.28	1.23	0.84 to 1.82	.29	1.18	0.68 to 2.04	.56

NOTE. Model for early referral and visit completion was restricted to patients who had a referral. Statistical significance was set at P < .05. Significant ORs (95% CIs) are bolded. Participants with missing values were excluded from the models.

Abbreviation: OR odds ratio.

The bivariate analyses of any PC referral demonstrated significant differences in the proportion of patients receiving a referral by cancer type and year of diagnosis (P < .05; Table 1). In multivariable logit regression, likelihood of any PC referral was higher among patients diagnosed with pancreatic cancer (aOR, 2.86 [95% CI, 1.50 to 5.44]) and patients diagnosed before 2020 (2018: aOR, 2.06 [95% CI, 1.41 to 3.02]; 2019: aOR, 1.60 [95% CI, 1.09 to 2.36]). Predictors of not receiving any PC referral were patients age 40 to <70 years (ages 40 to <50: aOR, 0.43 [95% CI, 0.19 to 0.98]; 50 to <60, aOR, 0.43 [95% CI, 0.20 to 0.93]; 60-<70: aOR, 0.41 [95% CI, 0.20 to 0.85]), patients diagnosed with gynecologic cancers (aOR, 0.14 [95% CI, 0.07 to 0.28]), and patients diagnosed with prostate cancer (aOR, 0.51 [95% CI, 0.28 to 0.93]; Table 2). Sensitivity analysis of including patients who died ≤8 weeks changed the adjusted OR of any PC referral for patients diagnosed with prostate cancer from 0.51 (P = .03) to 0.59 (P = .08), which was no longer statistically significant. No other changes in statistical significance were found. Sensitivity analysis including a pre/post COVID-19 indicator demonstrated prepandemic patients were 1.45 times more likely to receive a PC referral than postpandemic patients (95% CI, 1.15 to 1.84).

Outpatient PC Visits

In the subsample of patients who received any referral to PC, 472 (65.1%) of 725 completed a PC visit during the study period. The median days between PC referral and PC visit completion was 51 days (IQR, 35-70 days). Bivariate analysis indicated that there was a significant difference in the proportion of patients completing a visit by insurance type (P < .05; Appendix Table A2). In our multivariable logit regression analysis, compared with males, female patients had a higher odds of completing a visit (aOR, 1.45 [95% CI, 1.01 to 2.08]). No other variables were significant predictors of visit completion (Table 2). Sensitivity analysis of including patients who died ≤8 weeks of diagnosis yielded no changes in statistical significance.

Marginally Adjusted PC Referrals and Visits by Race/Ethnicity and Cancer Type

Adjusting for other covariates, patients who were Black (47.0%), Hispanic (36.6%), or diagnosed with prostate cancer (17.2%) had lowest probabilities of PC referral within 8 weeks of diagnosis (Figs 1A and 1B). We observed patients diagnosed with pancreatic cancer having the highest adjusted probability of any PC referral (79%), whereas the cancers with the lowest adjusted probability of any referral were gynecologic (16%) and prostate (41%; Fig 1C).

FIG 1.

(A) Adjusted predicted probabilities of receiving early outpatient PC referral by cancer type. Model was adjusted using variables listed in Table 1. (B) Adjusted predicted probabilities of receiving early outpatient PC referral by race/ethnicity. Model was adjusted using variables listed in Table 1. (C) Adjusted predicted probabilities of receiving any outpatient PC referral by cancer type. Model was adjusted using variables listed in Table 1. PC, palliative care.

DISCUSSION

In this study of patients with advanced cancer treated at a large, urban safety-net health system, we identified key patterns associated with receipt and timing of PC referrals and visit completion. We found that just over half (55%) of patients with advanced cancer received a referral to the outpatient PC team at Parkland, but more than half of these patients had the referral ordered more than 8 weeks after diagnosis (54.2%). Of the patients referred, two thirds went on to complete a PC visit during the study period.

Overall, these findings confirm and extend previous reports that show patients with advanced cancer experience low rates of outpatient PC referral and delays in accessing PC.^19-21,26,27 For example, Lakhani found 61% of patients with advanced cancer treated within a comprehensive cancer center were referred to outpatient PC after 8 weeks after diagnosis.²⁶ Likewise, Dillon et al²⁷ reported that 47% of patients treated at a large multispecialty group were referred to PC, with about half referred late (>60 days after diagnosis). Although few previous studies have been conducted within safety-net health systems, we note similar patterns of outpatient PC referrals and timeliness with a more diverse mix of patient demographics and insurance coverage statuses.^19-21

In our multivariable analysis, we found certain cancer types with significant differences in PC referral. Previous studies using population-based samples or conducted in non–safety-net health systems have observed similar differences.^7,28-31 We found advanced prostate cancer was the only cancer with both a significantly lower probability of an early referral (17%) and receiving any referral (41%), relative to other cancers. Previous literature indicates patients with prostate cancer present with a longer and more chronic disease course and may not necessarily require PC services at the time of advanced diagnosis.^28,29 Likewise, we found patients with advanced pancreatic cancer were no more likely to receive an early PC referral, but they were significantly more likely to receive any referral. This is consistent with previous research indicating patients with pancreatic cancer are offered PC, but it often occurs when patients are at their sickest and late in the disease course.³⁰ Finally, we observed patients diagnosed with advanced gynecologic cancer had a very low probability of any PC referral (15%). Previous studies of patients with advanced gynecologic cancer have reported similar findings.^7,31 For example, a large sample of patients from the National Cancer Database found <7% of deceased patients with metastatic gynecologic cancer used PC.⁷ The lower PC referral rate among patients with gynecologic cancers could reflect the training pathway of gynecologic oncologists who do not undergo medical oncology fellowships and may have less exposure to PC during their training. We also note that at Parkland, as is the case in many institutions, patients with gynecologic cancer receive treatment from the gynecologic oncology department, separated from the other combinations of care teams for the other solid tumors. Previous work in other health care systems indicates these service lines operate differently, which has implications for PC use and warrants further quantitative and qualitative investigation.³²

Across all our models, there was consistent indication that patients age >80 years had greater odds of receiving PC referral. Although the literature is quite mixed about the association between age and PC service use,^12,28,33-35 our findings suggest that the older patients in our study may have exhibited greater need for PC, such as less favorable prognosis, greater comorbidity, or higher symptom burden. Younger patients with cancer who seek care in safety-net health systems may experience more logistical barriers such as the time burden of having another clinic appointment, other competing priorities (eg, work, childcare), less acceptance of PC, or perceive that PC is synonymous with end of life and hospice care. These factors may have contributed to a lower likelihood of PC referral for younger patients compared with older patients in our study.^36,37

In this safety-net population, we found race/ethnicity was a significant predictor of early PC referral even after controlling for non–English language preference and insurance type. Black and Hispanic patients had a <50% probability of referral to PC within 8 weeks of diagnosis. These patients may have less early PC referral due to unmeasured factors such as a fragmented health care system and implicit biases of providers toward racial/ethnic minorities.³⁸ Furthermore, Black and Hispanic patients may experience barriers to early PC referrals due to low health literacy and a lack of understanding of PC, stress, and lack of social support system.³⁹ Cultural beliefs and value systems may also factor into patients' decisions on when to pursue PC services after diagnosis, resulting in delayed referral.^38,39 In addition, Black and Hispanic patients with cancer seeking care at a safety-net health system may differ from White patients in terms of disease characteristics. For example, White patients may experience delays in accessing cancer care before being transferred to a safety-net system and present with more advanced disease.^40,41 Yet, we did not find a significant race/ethnicity disparity in the probability of receiving any PC referral or PC visit completion. This latter finding contrasts with the existing literature describing broader disparities in PC access among these medically underserved groups.^16,39,42

There are several limitations to this study. Due to incomplete data, we were not able to account for certain factors that may predict PC use, such as symptom burden, reason for PC referral, and physician propensity to refer. In addition, we could not determine why patients did not receive a referral or complete an initial PC visit (ie, patient went directly into hospice care or died after the provider placed the PC referral). Our study also represents the experience of PC at only one safety-net system, which may limit generalizability to other systems. However, these analyses produced the most data, to date, about PC referrals within a safety-net system. The large population-based sample of safety-net patients with cancer allowed for comparisons among a large number of uninsured who may have been unlikely to have obtained PC in other health care systems. Approximately 80% of the population cared for at Parkland is uninsured and about one third are covered by Medicaid. Parkland is the only safety-net cancer treatment provider for the under- and uninsured in Dallas County. The charity assistance coverage for low-income adults who are ineligible for Medicaid is a county-funded program organized by Parkland Health, which does not cover treatment in other hospitals. Although Medicaid and other types of coverage allow for access to other health care systems, many patients continue their care through Parkland.²² As a result, a major study strength is the comprehensiveness of the data, which allowed for tracking the sequential steps of PC delivery in the outpatient setting (ie, referrals, timeliness, and completed visits). However, it is possible some patients who sought care in another system may have been missed and is a potential limitation of the study.

In conclusion, despite guideline recommendations that all patients with advanced cancer receive a referral to PC concurrent with active cancer treatment, nearly half of patients in our study did not receive a PC referral, and of those who had a referral, most received it late. Subgroups such as Black and Hispanic patients were more likely to receive a late PC referral, whereas younger patients and those diagnosed with gynecologic and prostate cancers were less likely to receive any PC referral. These findings demonstrate a need and potential for interventions, such as clinical triggers, targeting PC delivery to support implementation in safety-net health systems to improve care for underserved patients with cancer.

APPENDIX

TABLE A1.

Bivariate Associations Between Patient Characteristics and Time to First Palliative Care Referral (n = 725)

Variable	≤8 Weeks From Advanced Cancer Diagnosis, No. (%)	>8 Weeks From Advanced Cancer Diagnosis, No. (%)	P
Total	332 (45.8)	393 (54.2)
Sex
Male	183 (45.6)	218 (54.4)	.925
Female	149 (46.0)	175 (54.0)
Age, years			.753
<40	26 (45.6)	31 (54.4)
40 to <50	36 (40.9)	52 (59.1)
50 to <60	116 (45.8)	137 (54.2)
60 to <70	101 (44.9)	124 (55.1)
70 to <80	37 (50.7)	36 (49.3)
≥ 80	16 (55.2)	13 (44.8)
Race/ethnicity			.007
Black	112 (45.2)	136 (54.8)
Hispanic	117 (40.1)	175 (59.9)
White	85 (57.4)	63 (42.6)
Other/unknown	18 (48.6)	19 (51.4)
Language preference			.123
Non-English	106 (41.9)	147 (58.1)
English	226 (47.9)	246 (52.1)
Insurance type			.383
Charity/self-pay/Medicaid	252 (44.9)	309 (55.1)
Commercial/Medicare	80 (48.8)	84 (51.2)
Cancer type			<.0001
Breast	29 (40.8)	42 (59.2)
Colorectal	49 (46.7)	56 (53.3)
Gynecologic	3 (21.4)	11 (78.6)
Head/neck	17 (28.3)	43 (71.7)
Kidney	17 (51.5)	16 (48.5)
Lung	81 (51.9)	75 (48.1)
Pancreas	43 (55.1)	35 (44.9)
Prostate	10 (19.2)	42 (80.8)
Other GI	69 (52.3)	63 (47.7)
Other—unknown	14 (58.3)	10 (41.7)
Diagnosis year			.078
2018	88 (46.3)	102 (53.7)
2019	61 (40.1)	91 (59.9)
2020	84 (54.2)	71 (45.8)
2021	51 (39.8)	77 (60.2)
2022	48 (48.0)	52 (52.0)

TABLE A2.

Bivariate Associations Between Patient Characteristics and Palliative Care Visit Completion (n = 725)

Variable	Visit Completed, No. (%)	Visit Not Completed, No. (%)	P
Total	472 (65.1)	253 (34.9)
Sex			.155
Male	252 (62.8)	149 (37.2)
Female	220 (67.9)	104 (32.1)
Age, years			.198
<40	37 (64.9)	20 (35.1)
40 to <50	65 (73.9)	23 (26.1)
50 to <60	170 (67.2)	83 (32.8)
60 to <70	139 (61.8)	86 (38.2)
70 to <80	41 (56.2)	32 (43.8)
≥80	20 (69.0)	9 (31.0)
Race/ethnicity			.080
Black	169 (68.1)	79 (31.9)
Hispanic	176 (60.3)	116 (39.7)
White	105 (70.9)	43 (29.1)
Other/unknown	22 (59.5)	15 (40.5)
Language preference			.056
Non-English	153 (60.5)	100 (39.5)
English	319 (67.6)	153 (32.4)
Insurance type
Charity/self-pay/Medicaid	379 (67.6)	182 (32.4)	.010
Commercial/Medicare	93 (56.7)	71 (43.3)
Cancer type			.892
Breast	47 (66.2)	24 (33.8)
Colorectal	66 (62.9)	39 (37.1)
Gynecologic	8 (57.1)	6 (42.9)
Head/neck	43 (71.7)	17 (28.3)
Kidney	21 (63.6)	12 (36.4)
Lung	98 (62.8)	58 (37.2)
Pancreas	53 (67.9)	25 (32.1)
Prostate	38 (73.1)	14 (26.9)
Other GI	83 (62.9)	49 (37.1)
Other—unknown	15 (62.5)	9 (37.5)
Diagnosis year			.229
2018	135 (71.1)	55 (28.9)
2019	94 (61.8)	58 (38.2)
2020	103 (66.5)	52 (33.5)
2021	81 (63.3)	47 (36.7)
2022	59 (59.0)	41 (41.0)
Days from referral to completed visit
Mean (SD)	65.43 (78.14)	NA
Median (IQR)	51 (35-70)	NA

NOTE. Bold indicates statistical significance.

Abbreviations: NA, not applicable; SD, standard deviation.

FIG A1.

Sample selection.

FIG A2.

Time to palliative care referral by week.

AUTHOR CONTRIBUTIONS

Conception and design: Lisa DiMartino, Celette Sugg Skinner, Timothy P. Hogan, Navid Sadeghi, Alva Roche-Green, Arthur S. Hong

Administrative support: Celette Sugg Skinner

Provision of study materials or patients: Alva Roche-Green

Collection and assembly of data: Lisa DiMartino, Vincent Merrill, Alva Roche-Green, Arthur S. Hong

Data analysis and interpretation: Lisa DiMartino, Vincent Merrill, Timothy P. Hogan, Navid Sadeghi, Winnie Wang, Arthur S. Hong

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

Early Integration of Outpatient Palliative Care Among Adults With Advanced Cancer in a Safety-Net Health System: A Patterns of Care Analysis

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/op/authors/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).