Nurse led telephonic palliative care versus specialty outpatient palliative care: pragmatic, randomised clinical trial

Abstract

Objective

To compare the effectiveness of nurse led telephonic palliative care versus specialty outpatient palliative care on quality of life, symptom burden, loneliness, and healthcare use, after attending the emergency department.

Design

Pragmatic, randomised clinical trial.

Setting

Emergency Medicine Palliative Care Access (EMPallA) randomised controlled trial enrolling participants from 18 emergency departments in 15 geographically diverse healthcare systems in nine US states, from 1 April 2018 to 30 June 2022.

Participants

Of 39 254 eligible patients, 1283 adults who visited the emergency department, were aged ≥50 years, who spoke English or Spanish, and had advanced cancer or end stage organ failure, were randomised to receive nurse led telephonic palliative care (n=639) or specialty outpatient palliative care (n=644).

Interventions

The nurse led telephonic palliative care arm consisted of weekly or biweekly calls over six months made by registered nurses certified in hospice and palliative care. For the specialty outpatient palliative care arm, patients had one visit each month for six months with a specialty trained hospice and palliative medicine provider.

Main outcome measures

The primary outcome was change in patient reported quality of life at six months, measured by the Functional Assessment of Cancer Therapy-General (FACT-G) questionnaire. Secondary outcomes were change in symptom burden and patient reported loneliness after six months, and healthcare use, measured as the number of emergency department revisits, inpatient days, and hospice use, from enrolment to 12 months.

Results

639 patients were assigned to nurse telephonic services and 434 (68%) engaged in care until death, or until they required hospice services or graduated from the programme. For specialty outpatient palliative care, 644 patients were assigned and 344 (53%) attended one or more visits, with an average of 2.7 visits. The mean change in FACT-G scores over six months for the nurse telephonic arm (n=418) was 3.7 (95% confidence interval (CI) 2.3 to 5.1) points compared with 3.1 (1.6 to 4.6) for those in the specialty outpatient care arm (n=409). In the model including all patients who survived to six months (n=1090), the estimated difference in average change in quality of life was 0.71 (95% CI −1.19 to 2.61) points higher in the nurse led telephonic palliative care arm. The analysis did not show any clinically meaningful differences in the change in quality of life between the treatment arms. Also, no important differences between groups were found for secondary outcomes or in subgroup analyses.

Conclusions

The results of the study provided no clear evidence that nurse led telephonic palliative care improved quality of life, or any secondary outcomes, relative to specialty outpatient palliative care.

Trial registration

ClinicalTrials.gov NCT03325985.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Palliative care interventions in the emergency department can identify patients at high risk at a time of crisis and improve patient centred outcomes

• Outpatient palliative care is a limited resource that improves patient outcomes, but the effectiveness of different palliative care models for patients with serious illnesses discharged from the emergency department has not been established

WHAT THIS STUDY ADDS

• Despite broader participant coverage, nurse led telephonic palliative care was not superior to specialty outpatient palliative care

• No meaningful differences in quality of life, symptom burden, loneliness, or healthcare use were found between the two palliative care models

• No intervention effect by disease type, functional status, race, ethnic group, covid-19 period, or per protocol was found

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE, OR POLICY

• Health systems could consider how nurse led telephonic palliative care can increase the reach of traditional specialty outpatient palliative care

• More research is needed to learn how these two models can work together to improve care for seriously ill patients and families

Introduction

Outpatient specialty palliative care improves symptoms and quality of life for patients with a broad range of serious illnesses.¹,⁴ Patients receiving palliative care services can often be cared for and supported at home, resulting in greater patient and family satisfaction and less prolonged grief and post-traumatic stress disorder for bereaved family members.^{12 4},⁷ Palliative care also lowers costs by reducing unnecessary hospital admissions, diagnostic and treatment interventions, and avoidable intensive care and emergency department care.⁸,¹¹ Nonetheless, the numbers of specialty palliative care providers struggle to meet the needs of the growing number of persons living with serious illnesses.^{12 13}

Half of older American patients visit the emergency department in the last month of their life, making the emergency department a key decision point where providers establish the subsequent course of care for patients.¹⁴ Palliative care interventions in the emergency department can both identify patients at high risk at a time of crisis and markedly improve patient centred outcomes.¹⁵,¹⁷ The effectiveness of different palliative care models for patients with serious illnesses discharged from the emergency department has not been established. By testing new, highly efficient models of care with few barriers to implementation after patients have been discharged from the emergency department to home, we can build on the effectiveness of specialty palliative care for patients admitted to hospital after attending the emergency department.¹⁶

Two of the models developed for the delivery of coordinated, community based palliative care are nurse led telephonic palliative care and specialty outpatient palliative care. Nurse led palliative care programmes have grown under Medicare Advantage plans, showing reductions in cost by targeting older adults with multiple chronic conditions.¹⁸,²² Most programmes are run by registered nurses who contact patients and families by telephone, which is less expensive, more accessible, and less burdensome for persons living with serious illnesses who may have difficulties accessing transportation.¹⁸,²³ Specialty outpatient palliative care is interdisciplinary by nature, based in brick and mortar facilities paired with telemedicine, with current access limited to a subset of academic and Veterans Affairs medical centres.^{24 25} We conducted an 18 site randomised, pragmatic, superiority clinical trial to compare these two established palliative care models: nurse led telephonic care and specialty outpatient palliative care. We evaluated patient centred outcomes, including quality of life, symptom burden, loneliness, and healthcare use.

In our comparative effectiveness superiority trial, we recruited geographically, racially, and ethnically diverse older adults living at home with advanced cancer or end stage organ failure, and their caregivers, on attending the emergency department. We hypothesised that patients randomised to receive nurse led telephonic palliative care would have greater improvements in quality of life, and lower healthcare use, loneliness, and symptom burden than those referred to specialty outpatient palliative care because nurse led telephonic care could flexibly meet patients’ needs and promptly deal with acute concerns in real time as they arose. Unlike in clinics where appointments are fixed and unplanned visits are difficult to accommodate, participants receiving treatment by telephone were not required to travel or have specific technology equipment (other than a working telephone) to access palliative care, and were followed up by their primary doctor or disease specific specialist.

Methods

Trial design and oversight

Emergency Medicine Palliative Care Access (EMPallA) was an investigator initiated, multicentre, pragmatic, two arm, randomised controlled trial enrolling participants from 1 April 2018 to 30 June 2022 at 18 emergency departments in 15 healthcare systems in nine states. The institutional review boards of each participating centre approved the study. Participants provided written or verbal consent for participation in accordance with institutional review board approved consent procedures (NCT03325985).

Patient population and recruitment

Eligible patients were adults aged ≥50 years and scheduled for discharge from the emergency department or observation unit, with a poor prognosis for advanced cancer (metastatic solid tumour) or end stage organ failure (New York Heart Association class III or IV heart failure, end stage renal disease with glomerular filtration rate <15 mL/min/m² or on dialysis, or global initiative for chronic obstructive lung disease stage 3, 4, or oxygen dependent). Patient data for sex were from assigned sex rather than from self-reported gender. Patients resided within the geographical area of the enrolment site, spoke English or Spanish, had access to a working telephone, and had health insurance accepted by the outpatient clinic. For sites with no observation unit, patients who were admitted for two midnights or less were eligible to participate. Patients were excluded if dementia was reported in the problem list of the electronic health record, or if they received hospice services, made two or more palliative care visits in the previous six months, or resided in, or were discharged to, a long term care facility. Patients who received a new diagnosis of cancer while in the emergency department were also excluded.

Research assistants checked the emergency department and observation unit electronic track boards as well as discharge reports at each site to identify patients with previously defined qualifying conditions. The research assistants then approached patients and conducted face-to-face or telephone interviews according to the study protocol to confirm eligibility criteria. For patients interested in enrolling in the study, the research assistants obtained signed, informed consent. The full study protocol has been previously published (original version, online supplemental file 3; final version, online supplemental file 4).²⁶

Randomisation, masking, and risk stratification

After completion of the baseline survey, patients were randomised to specialty outpatient palliative care or nurse led telephonic palliative care by a computer generated, web based randomisation system with masking of allocation. Patient level randomisation was grouped by emergency department site and disease (cancer v end stage organ failure) in varying block sizes of two, four, and six at a ratio of 1:1 assignment to each group. A biostatistician randomly pre-assigned the expected number of study participants to the intervention groups at each site, and research coordinators had no knowledge of group assignments. Given the nature of the interventions, patients and treating clinicians were aware of their assigned group, but outcome assessors, statisticians, and data analysts were masked to group assignment.

Trial interventions

The two distinct palliative care models varied by provider type (nurse led v provider led), location (centralised v regional), and mode of delivery (telephonic v in person or telehealth). The nurse led telephonic palliative care intervention was centrally located at the New York University Grossman School of Medicine, whereas specialty outpatient palliative care was conducted at clinics within each respective health system. Both comparators are established palliative care models. In-depth protocols and fidelity checklists for each arm were previously published.^{23 27}

The models also varied by frequency of visits. The nurse led telephonic palliative care programme consisted of weekly or biweekly calls over six months made by registered nurses certified in hospice and palliative care. Six nurses were trained over the period of the study, with a maximum of three nurses working at any given time. As well as certification, nurses were trained in motivational interviewing,²⁸ problem solving theory, Centre to Advance Palliative Care modules,²⁹,³¹ and Respecting Choices Last Steps.³² A standard operating procedure manual was developed based on the domains of the National Consensus Project for Quality Palliative Care.³³ The standard operating procedure manual included an initial needs assessment, identifying patient goals, discussion and documentation of advance care planning, and identifying a surrogate decision maker. All intervention details were reported in Epic. Nurses communicated with patients' primary specialist or usual doctor for prescriptions. To monitor intervention fidelity (ie, the degree to which an intervention is implemented as intended by its developers), two board certified hospice and palliative medicine physicians conducted weekly interdisciplinary team meetings and all patients were discussed at least monthly. Additional fidelity monitoring was conducted on nursing scope of practice by a PhD trained advanced practice nurse certified in hospice and palliative care.

The outpatient arm consisted of one visit each month for six months with a specialty trained hospice and palliative medicine provider. A standardised intervention checklist was developed based on the domains of the National Consensus Project for Quality Palliative Care: physical, psychosocial, bedside behaviour and intervention, and spiritual and quality or end of life planning. The checklist was completed at each visit to monitor intervention fidelity. The provider determined the sequence and inclusion of the components of the intervention for each visit; the checklist was reported in Epic and REDCap (research electronic data capture). Bimonthly meetings were conducted with all outpatient co-investigators to review any barriers to, and best practices for, scheduling and follow-up; individual site level meetings occurred on a monthly basis. Covid-19 related adaptations were required to continue the study throughout the pandemic (2019-21). The specialty outpatient palliative care arm was originally designed as in-person meetings with patients but was expanded to provide telehealth visits when considered appropriate. Online supplemental table A has more details on the components of the two interventions.

Outcomes

Study outcomes were assessed mainly by telephone, with the option to complete electronically, in person, or by postal mail if preferred by the patient. Research coordinators facilitated the questionnaires and attempted to reach patients by telephone, email, secondary contacts, mailing addresses, and the emergency department trackboard.

The primary outcome was centrally assessed as change in patient reported quality of life from enrolment to six months, calculated as the Functional Assessment of Cancer Therapy-General (FACT-G, version 4) score at six months minus the baseline score. The FACT-G questionnaire has been validated in other chronic conditions and in the general population, and has a defined minimal clinically important difference of four points.^{34 35} Secondary outcomes were patient reported change in symptom burden at six months, measured by the Edmonton Revised Symptom Assessment Scale (ESAS-r),³⁶ and change in patient reported loneliness, measured by the Three Item Loneliness Scale (range 3-9; higher scores indicate greater loneliness).³⁷ More detail on how outcomes should be interpreted have been previously published.³⁸ Loneliness was included as an outcome after our patient study advisory committee advised that loneliness was important to include in palliative care research, although no a priori belief that the interventions would affect this outcome was established. Patient reported outcomes were collected at baseline and at three, six, and 12 months.

Healthcare use outcomes were derived from electronic health records and therefore had less missingness, even for patients with incomplete patient reported data. Healthcare use was measured as the number of emergency department revisits (count), inpatient days (count), and hospice use (binary) from enrolment to 12 months. We recorded revisits to the emergency department and inpatient days by data extraction from the electronic health record. Hospice use was by self-report as well as data extraction from the electronic health record. Preplanned exploratory subgroup analyses were conducted to assess hypothesised differences in effectiveness defined by disease group (cancer and end stage organ failure), functional status, race, and ethnic group.

Sample size and power

We used simulation methods to determine the required sample size and power of the study (online supplemental table B). The target effect size was 4.0, considered to be the minimal clinically important difference, with a standard deviation (SD) of 6.0. Based on these assumptions, a sample size of 574 patients for each arm (1148 total) was estimated to provide 82% power to detect this effect. Also, the same sample size was estimated to provide 80% power to detect a difference of 2.0 points (SD 6.0) in the change in FACT-G scores from baseline to six months. Assuming a 15% dropout rate, the final target sample size was 1350 participants.

Statistical analysis

Baseline characteristics for each treatment arm for patients who survived to six months are described as mean (SD) for continuous variables and proportions for categorical variables. Standardised mean differences were calculated to describe differences between treatment arms. We plotted mean response rates at baseline and at six months for patient reported outcomes and created graphical representations for healthcare use outcomes. We evaluated missing data patterns to determine whether data were missing differentially across arms, and to determine whether multiple imputation should be incorporated in the final analyses (online supplemental figures A and B, and table C and online supplemental handling missing data).

All analyses were based on intention-to-treat analyses. Mixed effects linear models with a random site level intercept (to account for variation in changes in quality of life between sites) and fixed effects for disease group were used to estimate the difference in changes in quality of life between the two interventions. The prespecified primary analysis excluded patients who died before the six month follow-up,³⁹ because a six month change in quality of life is not possible under any scenario for these patients.³⁹ We conducted an additional conditional longitudinal analysis, however, that included all patients who died after the first three months as a sensitivity analysis. Based on our missing data investigations, we used multiple imputation with chained equations (creating 20 datasets) to account for patients who survived up to six months but were missing quality of life measures at baseline or six months.

Estimates and standard errors across imputed datasets were combined based on Rubin's rules to account for variability within and between imputations. As an additional analysis, we also fitted a model estimating six month quality of life scores adjusted for baseline quality of life, as suggested during peer review, to assess whether conclusions differed from the primary analysis. A similar approach was used for the secondary patient reported outcomes, although we used a mixed effects cumulative odds model for the loneliness outcome because a normal model was not justified; the effect estimated was a cumulative odds ratio.

We modelled the healthcare use outcomes with generalised mixed effects linear models. Models for the outcomes of emergency department revisits and inpatient days assumed a Poisson (count) distribution with an offset to adjust for the time in the study to estimate a rate ratio. Hospice use was modelled with a Bernoulli (binary) distribution to estimate an odds ratio. We found no missing data for healthcare use outcomes other than minimal hospice use missingness as a result of lack of data sharing with one site. The subgroup analyses were conducted with the same approach. These subgroup analyses were exploratory; the study was not powered to detect differences in treatment effects across subgroups. We estimated treatment effects separately within each subgroup but did not formally test for heterogeneity by including interaction terms in the model.

We conducted three sensitivity analyses. The first analysis was a preplanned per protocol analysis of the primary outcome that only included patients in the trial that accessed the interventions. Typically, this design is problematic because the decision to receive the intervention is not randomised. In this case, however, we assumed that patients choosing the intervention in each arm were comparable with each other. The second analysis was a conditional cohort analysis of the primary outcome measured at three, six, and 12 months that included all patients in the study except those who died before the three month follow-up period.⁴⁰ In the third analysis, we assessed the impact of the covid-19 pandemic on the primary outcome by including an interaction term between the covid-19 period and the nurse led palliative care intervention in the models, with the full study sample for each. All analyses were conducted with R, version 4.3.1 (R Foundation for Statistical Computing). The statistical analysis plan is included in online supplemental file 5.

Patient and public involvement

We assembled a study advisory committee of 18 members with a history of co-funding, co-publication, and collaboration, comprising three different categories: patients (n=3) with serious diseases or their caregivers; members of healthcare organisations related to the study illnesses or to palliative care, or both (n=5); and payers (n=6). Members were recruited during the study planning process in September 2017 from across the US. Online supplemental table D has a list of the names and affiliations of the members of the study advisory committee. Members of the study advisory committee were involved in all phases of the research, which included the development of the appropriate comparators and outcomes of interest, conduct and oversight of the study, and analysis and dissemination of the results. Our coauthored published papers with members of our study advisory committee have detailed information about the level of engagement at each phase of the study, from development and inception to evaluation and dissemination.^{41 42} All participants in the EMPallA study received a copy of the research findings. Also, the results of this study have been shared at academic conferences focusing on palliative care and emergency medicine.

Results

Participants

From 1 April 2018 to 30 June 2022, 39 254 patients were evaluated. Of these, 2843 (7%) met all of the inclusion and no exclusion criteria; 1350 were enrolled in the study and 1284 (3%) were randomised (figure 1). One patient was lost to follow-up after randomisation but before completing any baseline data collection and was excluded from the analysis. Thus the study sample comprised 1283 patients: mean age 66.8 (SD 10.1) years, 671 (52%) women, 404 (32%) black patients, and 142 (11%) Hispanic patients. A total of 671 (52%) patients required assistance to care for themselves. Baseline characteristics were similar for the nurse telephonic palliative care and specialty outpatient care arms (table 1), for patients in the two arms who survived to the six month follow-up (online supplemental table E), and for those who survived and those who died before six months. We found some differences by illness categories (online supplemental table F).

Figure 1. Consolidated Standards of Reporting Trials (CONSORT) flowchart of selection of study participants.

Table 1. Baseline characteristics of 1283 patients enrolled in the Emergency Medicine Palliative Care Access study and randomised to nurse led telephonic palliative care or specialty outpatient palliative care.

	Total No of patients (n=1283)	Nurse led telephonic care (n=639)	Specialty outpatient care (n=644)
Mean (SD) age (years)	66.8 (10.1)	66.7 (10.2)	66.9 (10.1)
Women	671 (52.3)	350 (54.8)	321 (49.8)
Race:
White	720 (56.1)	355 (55.6)	365 (56.7)
Black	404 (31.5)	210 (32.9)	194 (30.1)
Multi-racial	22 (1.7)	8 (1.3)	14 (2.2)
Asian	15 (1.2)	8 (1.3)	7 (1.1)
American Indian or Alaskan Native	9 (0.7)	4 (0.6)	5 (0.8)
Native Hawaiian or other Pacific Islander	8 (0.6)	3 (0.5)	5 (0.8)
Other	88 (6.9)	44 (6.9)	44 (6.8)
Missing	17 (1.3)	7 (1.1)	10 (1.6)
Ethnic group (missing=20):
Hispanic or Latino	142 (11.1)	59 (9.2)	83 (12.9)
Functional status:
Normal activity	347 (27.0)	159 (24.9)	188 (29.2)
Cares for self, cannot do normal activities	261 (20.3)	137 (21.4)	124 (19.3)
Requires occasional assistance	390 (30.4)	201 (31.5)	189 (29.3)
Requires considerable assistance	204 (15.9)	98 (15.3)	106 (16.5)
Disabled	77 (6.0)	40 (6.3)	37 (5.7)
Missing	4 (0.3)	4 (0.6)	0 (0)
Cancer	459 (35.8)	228 (35.7)	231 (35.9)
Marital status:
Married or living with a partner	549 (42.8)	270 (42.3)	279 (43.3)
Never married	259 (20.2)	124 (19.4)	135 (21.0)
Separated or divorced	258 (20.1)	137 (21.4)	121 (18.8)
Widowed	190 (14.8)	92 (14.4)	98 (15.2)
Other	16 (1.2)	9 (1.4)	7 (1.1)
Missing	11 (0.9)	7 (1.1)	4 (0.6)
Mean (SD) FACT-G (missing=6)	64.8 (18.3)	64.8 (18.3)	64.7 (18.4)
Mean (SD) ESAS-r (missing=7)	40.4 (13.0)	40.3 (13.0)	40.6 (13.0)
Mean (SD) Three Item Loneliness Scale (missing=9)	5.01 (1.94)	5.00 (1.97)	5.03 (1.90)

Data are number (%) unless indicated otherwise.

ESAS-r, Edmonton Revised Symptom Assessment Scale; FACT-G, Functional Assessment of Cancer Therapy-General.

During the study, 639 patients were assigned to nurse telephonic services and 434 (68%) engaged in care until death, they required hospice services, or graduated from the programme. For specialty outpatient palliative care, 644 patients were assigned and 344 (53%) attended one or more visits, with an average of 2.7 visits. Of the 19 specialty outpatient palliative care clinics, 18 offered in-person visits for the first appointment, and all 19 clinics allowed follow-up by telehealth. More details about nurse telephonic services and specialty outpatient palliative care have been published elsewhere.^{23 26 27} In the nurse telephonic arm, 122 (19%) patients were lost to follow-up during the first six months, 99 (15%) died, and 418 (65%) completed the six month follow-up. In the specialty outpatient arm, 141 (22%) patients were lost to follow-up during the first six months, 94 (15%) died, and 409 (64%) completed the six month follow-up (figure 1). Online supplemental figure C has a more detailed Consolidated Standards of Reporting Trials (CONSORT) diagram (the CONSORT checklist is in online supplemental file 6).

Primary and secondary outcomes

Figure 2 shows mean quality of life scores by treatment arm and disease group at baseline and six months for patients who survived to the six month follow-up and had complete data. Among these patients, the mean change in FACT-G scores over six months for the nurse telephonic arm (n=418) was 3.7 (95% confidence interval (CI) 2.3 to 5.1) points compared with 3.1 (1.6 to 4.6) points for those in the specialty outpatient care arm (n=409). In the model including all patients who survived to six months (n=1090; table 2), the estimated change in quality of life was 0.71 (95% CI −1.19 to 2.61) points higher in the nurse led telephonic palliative care arm. We could not conclude that the change in quality of life was different between the treatment arms for patients who survived to six months (table 2). Online supplemental table G has the full model results for the primary and secondary outcomes. The additional baseline adjusted analysis gave results consistent with the primary analysis, and we could not conclude that a difference existed between the two groups.

Figure 2. Distribution of quality of life scores for nurse led telephonic palliative care and specialty outpatient palliative care intervention arms by disease group (cancer and end stage organ failure) over 12 months. *Reported number of patients (n) does not match the analytical sample size because the figure shows only observed data whereas the analysis included patients with imputed values. FACT-G=Functional Assessment of Cancer Therapy-General score.

Table 2. Primary and secondary outcome measures: observed and estimated outcomes in patients randomised to nurse led telephonic palliative care or specialty outpatient palliative care.

Outcome	Nurse led telephonic care			Specialty outpatient care			Adjusted effect estimate†
	Baseline	Follow-up*	Mean change	Baseline	Follow-up*	Mean change
Quality of life at 6 months (primary outcome):
No of patients	634	418	418	643	409	409	1090
Mean (SD)	64.8 (18.3)	69.6 (19.4)	3.70 (14.6)	64.7 (18.4)	68.0 (18.9)	3.12 (15.0)	0.71 (difference)
95% CI	63.3 to 66.3	67.7 to 71.5	2.3 to 5.1	63.4 to 66.3	66.2 to 70.0	1.6 to 4.6	−1.19 to 2.61
Symptom burden at 6 months:
No of patients	636	413	413	640	409	407	1090
Mean (SD)	40.3 (13.0)	36.8 (12.7)	−3.03 (12.7)	40.6 (13.0)	37.9 (12.6)	−2.54 (12.9)	−0.39 (difference)
95% CI	39.4 to 41.4	35.6 to 38.1	−4.3 to −1.8	39.5 to 41.5	36.7 to 39.2	−3.8 to −1.3	−1.92 to 1.13
Loneliness at 6 months:
No of patients	636	411	410	638	408	405	1090
Median (IQR)	5 (2-5)	5 (1-5)	0 (0-2)	5 (2-5)	5 (2-5)	0 (0-2)	1.08 (cumulative odds ratio)
95% CI	NA	NA	NA	NA	NA	NA	0.83 to 1.41
Emergency department revisits at 12 months:
No of patients	NA	639	NA	NA	644	NA	1283
Mean (SD)	NA	2.5 (3.5)	NA	NA	2.7 (3.8)	NA	0.90 (risk ratio)
95% CI	NA	2.2 to 2.8	NA	NA	2.4 to 3.0	NA	0.78 to 1.04
Inpatient days at 12 months:
No of patients	NA	638	NA	NA	641	NA	1279
Mean (SD)	NA	8.1 (15.3)	NA	NA	8.6 (18.6)	NA	0.93 (risk ratio)
95% CI	NA	6.9 to 9.3	NA	NA	7.1 to 10.1	NA	0.74 to 1.16
Hospice use at 12 months:
No of patients	NA	635	NA	NA	641	NA	1276
Proportion	NA	0.16	NA	NA	0.18	NA	0.85 (risk ratio)
95% CI	NA	0.13 to 0.19	NA	NA	0.15 to 0.21	NA	0.62 to 1.16

Functional Assessment of Cancer Therapy-General (FACT-G) quality of life scores, symptom burden, and loneliness models used multiple imputation to adjust for participants with unreported outcome measures at either baseline or six months. Emergency department revisits, inpatient days, and hospice use did not require multiple imputation because data were collected from electronic health record data sources.

^*For quality of life, symptom burden, and loneliness, follow-up was at six months. For emergency department revisits, inpatient days, and hospice use, follow-up was at 12 months.

^†All models used multiple imputation with chained equations (creating 20 datasets) to account for patients who survived until six months but were missing baseline or six month quality of life measures. 1090 patients were used in the analysis of the primary outcomes, which included all patients who survived until the six month follow-up and imputing values for patients who did not respond.

CI, confidence interval; IQR, interquartile range; NA, not available; SD, standard deviation.

The change in symptom burden for the nurse led telephonic palliative care arm (n=413) was −3.0 (95% CI −4.3 to −1.8) compared with −2.54 (−3.8 to −1.3) for the specialty outpatient care arm (n=409) (table 2 and online supplemental figure D). The estimated difference between the two groups was −0.39 (95% CI −1.92 to 1.13) (table 2). Median change in loneliness scores for the nurse telephonic arm (n=411) was 0 (interquartile range −1 to 1) and 0 (interquartile range −1 to 1) for the outpatient specialty arm (n=408) (online supplemental figure D); the estimated cumulative odds ratio was 1.08 (95% CI 0.83 to 1.41) (table 2).

Figure 3 and figure 4 show healthcare use outcomes by treatment arm and disease group. The analyses indicated no differences in emergency department revisits, inpatient days, or hospice use between the two treatment arms (table 2 and online supplemental table G).

Figure 3. Healthcare use, measured as the number of emergency department revisits and inpatient days, for nurse led telephonic palliative care and specialty outpatient palliative care intervention arms and by disease group (cancer, solid lines; end stage organ failure, dashed lines) over 12 months.

Figure 4. Healthcare use, measured as hospice use, for nurse led telephonic palliative care and specialty outpatient palliative care intervention arms and by disease group (cancer and end stage organ failure) over 12 months.

Subgroup analyses

Subgroup analyses indicated no differences in the change in quality of life for the two treatment arms by functional status, race, or ethnic group after adjusting for disease group (online supplemental figures E-G and tables H-J). Similarly, we found no apparent differences in the change in quality of life across treatment groups within each disease group.

Sensitivity analyses

The per protocol analysis evaluated the change in quality of life for 661 patients who received their assigned intervention and survived until six months. The estimated difference across treatment arms was 0.37 (95% CI −2.76 to 2.03). The conditional cohort longitudinal analysis of quality of life did not identify any differences between the treatment arms (online supplemental figure H and table K). The covid-19 period also had no apparent effect on estimates of treatment effects (online supplemental table L).

Discussion

Principal findings

In this large randomised clinical trial, we compared two models of community based palliative care for older adults with serious, life limiting illnesses, discharged home after attending the emergency department. About half of eligible patients agreed to take part in our study, which is consistent with other studies that recruited patients with serious illnesses from the emergency department.⁴³ The population was sufficiently diverse, representing 15 health systems across nine states, with about a third of patients self-identifying as black and 11% as Latino, a third enrolled in Medicaid, and a third with household incomes <$25 000 (£18 700; €21 600). Also, 12% of participants were born outside the US and 37% had no further education beyond secondary education. More than a third of patients had advanced cancer and the other two thirds had advanced heart, lung, or kidney disease, making the generalisability of our findings applicable across these disease groups. Despite the increased participant engagement and broader coverage of the nurse led telephonic care, our hypotheses that patients randomised to this form of palliative care arm would have greater improvements in quality of life and lower healthcare use, loneliness, and symptom burden than those referred to specialty outpatient palliative care were not supported by the findings. Estimates for the changes in FACT-G were below the minimal clinically important difference,^{34 35} and there were no meaningful differences in primary or secondary outcomes between treatment arms. We found no differences between groups in quality of life, symptom burden, or loneliness from enrolment to six months, and no differences in healthcare use.

We observed no heterogeneity of treatment effect by disease status (advanced cancer v end stage organ failure), functional status, race or ethnic group (ie, white racial and ethnic participants v non-white racial and ethnic participants), or per protocol. More patients in the telephonic arm engaged in care (68%) than the outpatient arm (53%), suggesting that telephonic care may be more accessible to patients. Both arms showed small improvements in quality of life and decreases in symptom burden over time, despite disease progression. In studies of other palliative care interventions, these measures were stable in the treatment arm but worsening in the control arm, likely because quality of life is expected to decrease over time in a population with serious illnesses.^{2 44 45} The improvements we found in quality of life in both treatment arms could be because of a combination of both interventions working as intended, recovery of quality of life after attending the emergency department, or regression to the mean because those with worse quality of life were lost to follow-up or died.

Comparison with other studies

Despite the lack of previous comparative effectiveness studies, both models of community based delivery of palliative care have a robust evidence base. Nurse led telephonic palliative care programmes are appealing to payers and medical centres that participate in alternative payment models, and to policy makers looking for scalable outpatient palliative care models.^{46 47} Telephonic palliative care programmes have shown reductions in acute care and an increase in hospice use when targeting patients in the last months to years of life.^{18 20 48} Previous randomised controlled trials of nurse led primary palliative care interventions did not show similar effects on patient centred outcomes, likely because of the small sample size and the relatively lower dose of the intervention.⁴⁹,⁵¹ A Veterans Affairs study with a similar nurse or social worker telephonic model for patients with heart and lung failure and poor quality of life showed a major improvement in quality of life.⁵² Nurse telephonic care could increase outpatient palliative care to deal with shortages in the workforce. More research is needed to learn which patients would benefit and how provider clinic and telephonic nurse care could combine to expand access to palliative care.

Limitations of this study

Our study had several limitations. Firstly, about half of the patients who attended the emergency department who would otherwise have qualified for enrolment declined or could not participate in our study. Hence our patient cohort may have been healthier than those who otherwise met our broad inclusion criteria and therefore our findings could be less generalisable to the typical population referred to specialty outpatient palliative care. Secondly, fewer patients participated in either intervention than anticipated because of a combination of death, inability to attend in-person visits, inability to use telehealth technology, or lack of interest. Patients with stage III heart failure and chronic obstructive pulmonary disease, end stage renal disease, and some patients with metastatic cancer may have had stable disease or too few symptoms and therefore did not feel that they needed the interventions. Patients automatically referred based on illness criteria instead of physician initiated referral may have been less interested in palliative care or needed it less. Only 53% of patients randomised to specialty outpatient palliative care completed a visit and the average number of total visits was only 2.7. In contrast, 68% of those randomised to telephonic care engaged with the nurse intervention, suggesting that telephonic care may be more accessible to patients. The detailed protocol and patient, provider, clinic, and healthcare system factors associated with outpatient use have been previously published.⁵³ Despite lower engagement with the outpatient arm, the per protocol analysis showed no differences between the treatment arms.

Also, recruitment occurred during a global pandemic that required changing from mainly face-to-face visits in brick and mortar facilities for the specialty palliative care arm to telehealth delivered care. We conducted sensitivity analyses that showed no difference across treatment arms by pandemic status. Despite this analysis, we might have found a difference if patients had continued to be seen in face-to-face visits, although we consider this unlikely. Also, we assumed that the patient reported outcomes at all time points were missing at random, which may not have been the case. Although the healthcare use secondary outcomes included emergency department revisits, inpatient stays, and hospice use, including use of primary care services might have been useful. Finally, some may question whether we should have conducted a non-inferiority trial because the interventions had similar effects. Given the inherent weakness of a non-inferiority design, including difficulty defining the prespecified margin, we do not believe that this alternative design would have affected the relevance of our findings.

Conclusions

Although seriously ill older adults who received nurse led telephonic palliative care and those who received specialty outpatient care showed slight improvements in quality of life and small reductions in symptom burden over six months, no observed difference in the estimated change in quality of life between the two groups was found. Despite the broader coverage, nurse led telephonic palliative care was not superior to specialty outpatient palliative care based on quality of life or any secondary outcomes. Overall, our findings were negative. Given the previously shown benefits of both delivery models of palliative care, however, health systems should consider how centralised, telephonic nurse led palliative care can increase the reach of traditional specialty outpatient palliative care. More research is needed to learn how these two models can work together to improve care for seriously ill patients and families. Finally, new payment models are needed to reimburse care provided by nurses trained in palliative care and care provided by telephone.

Data availability statement

Data are available upon reasonable request.