Outcomes of Home-based Primary Care for Homebound Older Adults: A Randomized Clinical Trial

Abstract

Background

Homebound older adults are medically complex and often have difficulty accessing outpatient medical care. Home-based primary care (HBPC) may improve care and outcomes for this population but data from randomized trials of HBPC in the United States are limited.

Methods

We conducted a randomized controlled trial of HBPC vs. office-based primary care for adults ages ≥65 years who reported ≥1 hospitalization in the prior 12 months and met the Medicare definition of homebound. HBPC was provided by teams consisting of a physician, nurse practitioner, nurse, and social worker. Data were collected at baseline, 6- and 12-months. Outcomes were quality of life, symptoms, satisfaction with care, hospitalizations and emergency department (ED) visits. Recruitment was terminated early because more deaths were observed for intervention patients.

Results

The study enrolled 229 patients, 65.4% of planned recruitment. The mean age was 82 (9.0) years and 72.3% had dementia. Of those assigned to HBPC, 34.2% never received it. Intervention patients had greater satisfaction with care than controls (2.26, 95% CI 1.46 to 3.06, p<0.0001; effect size 0.74) and lower hospitalization rates (−17.9%, 95% CI −31.0% to −1.0%; p=0.001; number needed to treat 6, 95% CI 3 to 100). There were no significant differences in quality of life (1.25, 95% CI −0.39 to 2.89, p=0.13), symptom burden (-1.92, 95% CI −5.22 to 1.37, p=0.25) or ED visits (1.2%, 95% CI −10.5% to 12.4%; p=0.87). There were 24 (21.1%) deaths among intervention patients and 12 (10.7%) among controls (p<0.0001).

Conclusion

HBPC was associated with greater satisfaction with care and lower hospitalization rates but also more deaths compared to office-based primary care. Additional research is needed to understand the nature of the higher death rate for HBPC patients, as well as to determine the effects of HBPC on quality of life and symptom burden given the trial’s early termination.

Registration

Clinicaltrials.gov (registration number NCT02965508).

Introduction

Over 2 million Americans 65 years and older are homebound,¹ and live with high levels of multimorbidity, physical and cognitive impairment, and socioeconomic vulnerabilities that place them at risk for poor outcomes,² including hospitalizations rates that are over 3-times that of non-homebound older adults.¹ Their risks are compounded by limited access to care, especially geriatric and palliative care. A study of Kaiser Permanente members found that more than one quarter of frail older adults had no outpatient visits over 12 months while experiencing high rates of inpatient hospitalization, suggesting that functional challenges compromise proactive chronic disease management.³

Home-based primary care (HBPC) circumvents many barriers to care for patients like these. HBPC typically involves multidisciplinary teams who provide care and the intensive care coordination required to keep severely ill adults at home and out of long-term care facilities.⁴ Observational studies suggest that HBPC improves outcomes and reduces spending,^5–8 as have three randomized trials in the U.S.^9–11 However, the randomized trials were conducted more than 25 years ago and two were from the Veterans Affairs system whose model of HBPC focused on patients transitioning from inpatient care to home.

To provide a contemporary evaluation of the effect of HBPC on homebound adults, we conducted a randomized controlled trial of HBPC versus office-based primary care and examined satisfaction with care, quality of life, symptom burden, and hospitalizations and emergency department (ED) visits.

Methods

Study Participants and Settings

We recruited homebound adults ≥65 years in New York City, NY who reported ≥1 hospitalization in the prior 12 months, need for assistance with ≥2 activities of daily living, and leaving the home ≤2 times each week. These criteria approximated the Medicare definition of homebound, which specifies that leaving the home requires a considerable and taxing effort. Participants or their proxies had to be willing to transfer care to a HBPC clinician.

Study methods were reported previously.¹² Patients were recruited from Mount Sinai Health System primary care practices, the New Jewish Home subacute rehabilitation facility, and the Visiting Nurse Service of New York. Research assistants contacted the primary or supervising physician to determine homebound status and for assent for recruitment. After telephone screening, consent was obtained in person from the patient or their legally authorized proxy when the patient lacked capacity. A baseline (first) interview was conducted in English or Spanish immediately after consent was obtained and prior to randomization. Follow-up interviews were planned for 6 and 12 months. Participants received $50 for each interview. Study methods were approved by the institutional review boards of the Icahn School of Medicine at Mount Sinai, the New Jewish Home, and the Visiting Nurse Service of New York. Patients were enrolled from February 8, 2017 to July 2, 2019, and interviews were completed in September 2020.

Home-based Primary Care Intervention

HBPC was provided by an established program serving approximately 1200 homebound adults in Manhattan. Details of the program are provided elsewhere.^12–14 A HBPC nurse unaware of the patient’s study participation called to assess their appropriateness for the program and determine urgency using a triage assessment, which considered life expectancy, clinical stability, home environment, social support needs, and the family’s burden of care. The nurse instructed all patients to continue office-based care until their first HBPC visit, since waiting times depended on the availability of the team serving the patient’s neighborhood and could be as long as eight months.

HBPC was provided by a team consisting of a physician, nurse practitioner, nurse, social worker, and administrative assistant. At the initial visit, the physician completed a medical history and physical exam and developed a plan of care. Additional assessments included symptoms, activities of daily living, and fall risk if indicated. Physicians discussed advance care planning and encouraged patients and caregivers to complete a Medical Outcomes of Life Sustaining Treatments (MOLST) form.¹⁵ Follow-up visits occurred at the team’s discretion. Telephone access was available 24/7 and urgent in-person visits by a HBPC team member on weekdays. Social workers conducted a psychosocial needs assessment for new patients and provided ongoing assistance as needed. The teams met twice a week for inter-disciplinary rounds. Independent entities provided phlebotomy and x-ray services, physical and occupational therapy, mental health services, and hospice, as needed. Specialty referrals were mostly office-based. Patients randomized to HBPC remained in the program after completing the study.

Usual Care

Patients randomized to usual care continued care with their usual provider. They were offered HBPC at the completion of study participation.

Randomization and Blinding

Participants were randomized 1:1 to intervention or control by computer-generated randomization in blocks of 6, stratified by cognitive impairment (none to mild, moderate, severe). The project manager informed participants of their assignments by telephone; both were unblinded. Research assistants who performed interviews and chart abstractions, all investigators, and HBPC program staff were blinded.

Outcome Measures

We used measures validated for older chronically ill adults. Quality of life was measured for all patients using the 13-item Quality of Life in Alzheimer’s Disease (QoL-AD; range, 13–52).^{16, 17} Higher score indicates greater quality of life and a change in score ≥3.9 is considered the minimal clinically important difference (MCID).¹⁸ Symptoms were assessed using the Edmonton Symptom Assessment System, which measures severity of physical and emotional symptoms with 9 items (ESAS; range 0–90). Higher score indicates greater overall symptom burden (MCID, ≥3).¹⁹ Satisfaction with care was ascertained with the 5-item Family Satisfaction with End-of-Life Care (FAMCARE), which assesses doctor’s attention to the patient, information about tests, attention to symptoms, quality of follow-up care, and doctor’s availability (range, 0–10; higher score indicates greater satisfaction).²⁰

Data on hospitalization events and emergency department (ED) visits not resulting in hospitalization occurring during the 12 months prior to date of randomization and 12 months post-randomization were obtained from the New York Statewide Planning and Research Cooperative System (SPARCS) database. Hospitals are mandated to report all in-patient and ED discharge, regardless of payer.

Other Measures

Cognitive functioning was assessed with the Montreal Cognitive Assessment (MoCA).²¹ Scores were adjusted for age and education and categorized as normal, mild impairment, and dementia.²² We measured activities of daily living (bathing, toileting and continence, dressing, eating, walking, shopping, preparing meals, completing housework) and instrumental activities of daily living (making phone calls, managing medications, managing money), and summed the number of partial (score = 1) or complete dependencies (score =2) for each. Social network was measured using the Lubben Social Network assessment.²³ Score <12 indicates risk of social isolation.

Deaths of study participants were reported by proxies or caregivers at follow-up interviews and confirmed by medical record review. Medical records were abstracted for death and process of care data by research assistants (RA) using a structured abstraction form. A second RA independently abstracted 20 charts. Agreement on all data elements exceeded 95%.

Early Termination of the Study

The Data Safety Monitoring Board halted recruitment on July 2, 2019, citing delays from randomization to first HBPC visit and excess mortality in the treatment arm. The funder instructed the study team to complete a final interview with all participants remaining in the study and bypass interim data collection. Thus, study participants who had not yet completed a six-month interview completed interviews only at baseline and 12 months. Patients receiving HBPC continued to receive it and control patients were offered enrollment in HBPC after completing the 12 month interview.

Adjudication of Deaths

To determine whether delays in receipt of HBPC or participation in HBPC contributed to death, two Mount Sinai hospitalist physicians trained in case evaluation for morbidity and mortality reporting and unassociated with the study independently reviewed the decedents’ medical records using a structured data abstraction tool (available on the online supplement). Disagreements were resolved by a third reviewer.

Analyses

The primary outcomes were quality of life, symptoms, satisfaction with care, hospitalizations and ED visits. Targeted enrollment was 350 (175 participants per arm) for 80% power to detect a 38% relative difference in hospitalizations for intervention and control patients (two-sided alpha of 0.05, assuming 30% attrition). The enrollment target provided >80% power to detect significant differences for all primary outcomes based on estimates from observational studies.

Self-reported outcomes were continuous measures analyzed at baseline and 12 months. Secondary analyses examined the physical and emotional symptom domain scores of the ESAS and the individual components of the FAMCARE. A data management error caused loss of 75% of the data on pain and 27% on quality of life at baseline. The ESAS total and physical symptom scores were calculated without pain. Hospitalizations and ED visits were reported as the number of patients with an event and the average number of events per patient in the 12 months prior to and following randomization.

Primary analyses were conducted on the intention-to-treat principle. We calculated within group differences from baseline to 12 months and between group differences at 12 months using linear mixed models for normally distributed outcomes (quality of life, symptoms, satisfaction with care) and generalized linear mixed models (GLMM) for hospitalizations and ED visits. The GLMM models employed a logit link function for the outcome of any event (binomial distribution) and log link function for number of events (negative binomial distribution). All models accounted for the correlations of multiple measures within individual patients and study sites (G-side random effects),²⁴ and were adjusted for time and baseline social isolation and hospitalizations because they differed significantly between groups. All outcomes were analyzed using PROC GLIMMIX in SAS (version 9.4, Cary, NC). For patients with missing data not due to death, outcome values were imputed using the Markov Chain Monte Carlo method. Hospital and ED event data were considered complete for all patients. Differences in scores and number of events and their 95% confidence intervals, and effect sizes, were calculated from the adjusted models. Confidence intervals for rates of events (hospitalization, ED visit) were calculated by bootstrap (n=2000 simulations) as described by Horwitz et al.²⁴ P <0.05 was considered statistically significant. P-values for the subdomains of symptoms and satisfaction of care were adjusted for multiple comparisons.²⁵

We used worst-rank score analysis to address the problem of data missing due to death, which we could not assume were missing at random.²⁶ Outcome data were transformed into ranks, with the highest rank for best scores and lowest rank for worst scores or death. The earlier the patient died during study participation the lower the assigned rank. P-values for difference in ranks between study arms were calculated using the Wilcoxon Rank Sum Test.

We also performed per protocol analyses by adjusting for timing of the first HBPC visit or treating intervention patients who never received HBPC as controls. Both sets of per-protocol analyses were performed using the methods described for the main outcomes analyses.

Results

Recruitment, Subject Characteristics, and Receipt of the Intervention

Recruitment reached 229 (65.4% of goal); 188 (82.1%) from primary care practices, 29 (12.7%) from a visiting nurse service, and 12 (5.2%) from a subacute rehabilitation facility. Three were withdrawn by the study team and 114 intervention and 112 control patients were included in intention-to-treat analyses (Figure 1).

Figure 1.

Clinical Trial Flow Diagram

The mean age was 82 years (standard deviation, 9.0); 25.7% were male, 35.0% were Black, and 32.3% Hispanic; 37.6% lived alone and 72.3% had Dementia (Table 1). Fewer intervention patients were socially isolated than controls (40.3% vs. 58.0%, respectively, p=0.004) whereas intervention patients had more hospitalizations (mean 0.9 [1.5] vs. 0.5 [0.8], respectively, p=0.02).

Table 1.

Baseline Characteristics of Patients by Study Group

	All Subjects N= 226	Intervention N= 114	Control N= 112	P
	N (%)	N (%)	N (%)
Proxy respondent	94 (41.6)	46 (40.4)	48 (42.9)	.70
Age, mean (sd)	82 (9.0) years	82 (8.8) years	81 (9.1) years	.64
Age, years				.53
65–74	54 (23.9)	26 (22.8)	28 (25.0)
75–84	83 (36.7)	39 (34.2)	44 (39.3)
≥85	89 (39.4)	49 (43.0)	40 (35.7)
Male	58 (25.7)	33 (29.0)	25 (22.3)	.25
Race/ethnicity				.27
White, non-Hispanic	64 (28.7)	34 (30.4)	30 (27.0)
Black, non-Hispanic	78 (35.0)	44 (39.3)	34 (30.6)
Hispanic	72 (32.3)	31 (27.7)	41 (36.9)
Education				.30
<High school	45 (20.2)	19 (17.0)	26 (23.4)
High school graduate	72 (32.3)	39 (34.8)	33 (29.7)
Some college	42 (18.8)	25 (22.3)	17 (15.3)
College graduate	64 (28.7)	29 (25.9)	35 (31.5)
Monthly household income <$1,350	92 (50.0)	45 (49.5)	47 (50.5)	.88
Has a paid caregiver	166 (73.4)	83 (72.8)	83 (74.1)	1.00
Married/partnered	54 (24.0)	30 (26.6)	24 (21.4)	.37
Lives alone	85 (37.6)	44 (38.6)	41 (36.6)	.75
Type of residence				.07
Private apartment	139 (61.0)	78 (68.4)	61 (53.5)
Public housing	68 (29.8)	28 (24.6)	40 (35.1)
Other	21 (9.2)	8 (7.0)	13 (11.4)
Social isolation (Lubben score <12)	111 (49.1)	46 (40.3)	65 (58.0)	.004
General health, poor-fair	143 (63.6)	68 (60.2)	75 (67.0)	.29
ADL impairment score, mean (sd)	15.7 (3.9)	15.7 (3.9)	15.7 (3.8)	.84
IADL impairment score, mean (sd)	5.4 (2.0)	5.4 (2.0)	5.3 (2.0)	.66
Use of assistive device
Cane	128 (59.8)	62 (57.4)	66 (62.6)	.47
Walker	170 (78.7)	82 (75.2)	88 (82.2)	.21
Wheelchair	103 (47.5)	48 (43.7)	55 (51.4)	.25
Motorized wheelchair/scooter	16 (7.4)	8 (7.3)	8 (7.5)	.95
Cognitive status				.98
Normal	26 (11.8)	13 (11.8)	13 (11.8)
Mild cognitive impairment	35 (15.9)	17 (15.5)	18 (16.4)
Dementia	159 (72.3)	80 (72.7)	79 (71.8)
# Hospitalizations, preceding 12 mos.
Total	162	107	55
Per patient, mean (sd)	0.7 (1.2)	0.9 (1.5)	0.5 (0.8)	.02
# ED visits, preceding 12 mos.
Total	127	67	60
Per patient, mean (sd)	0.6 (1.2)	0.6 (1.3)	0.5 (1.1)	.91

ADL, activities of daily living; IADL, instrumental activities of daily living; SD, standard deviation; ED, emergency department.

Hospitalization and emergency department data were obtained from the New York State Statewide Planning and Research Cooperative System mandatory discharge database; all other data were self-report.

Of those randomized to HBPC, 39 (34.2%) did not receive the intervention, including 17 who refused it (Figure 1). The mean number of days from randomization to first home visit was 86.3 (56.7) days.

Quality of Life, Symptom Burden, and Satisfaction with Care

Mean scores on the QoL-AD increased for HBPC patients from 30.6 (standard deviation 6.2) at baseline to 32.7 (7.1) at 12 months (mean difference, 1.76, 95% confidence interval 0.66 to 2.86, p=0.002), while control patients had no increase (Table 2). However, the difference between arms was not significant (adjusted mean difference, 1.25, 95% CI −0.39 to 2.89, p=0.13).

Table 2.

Quality of Life, Symptom Burden, and Satisfaction with Care at Baseline and 12 Months, Within and Between Group Differences

					Within Group Difference at 12 Months		Between Group Differences at 12 Months
	N	Baseline Mean (SD)	N	12 Months Mean (SD)	Δ, 95% CIa	P	Δ, 95% CIa	Model-based Pa	Worst Rank Score P
Quality of Life
Intervention	82	30.6 (6.2)	77	32.7 (7.1)	1.76 (0.66, 2.86)	.002	1.25 (−0.39, 2.89)	.13	.77
Control	83	30.2 (5.5)	88	30.4 (5.9)	−0.01 (−1.03, 1.02)	.99
Symptom Burden
Intervention	113	30.7 (12.4)	76	27.9 (11.3)	−3.53 (−5.53, −1.53)	.001	−1.92 (−5.22, 1.37)	.25	.48
Control	112	29.8 (12.1)	88	29.3 (10.9)	−0.69 (−2.59, 1.21)	.48
Physical domain
Intervention	113	20.8 (8.6)	76	19.6 (8.0)	−1.38 (−2.93, 0.17)	.08	−0.95 (−3.39, 1.39)	.79b	.37
Control	112	20.8 (8.7)	88	20.5 (7.9)	−0.50 (−1.97, 0.98)	.51
Emotional domain
Intervention	113	9.9 (5.6)	76	8.3 (5.5)	−2.10 (−3.10, −1.10)	<.0001	−0.89 (−2.46, 0.68)	.79b	.40
Control	112	9.0 (5.6)	88	8.8 (5.8)	−0.23 (−1.18, 0.72)	.64
Satisfaction with Care
Intervention	112	6.5 (2.7)	74	7.6 (2.8)	1.14 (0.53, 1.76)	.001	2.26 (1.46, 3.06)	<.0001	.007
Control	109	5.7 (3.1)	84	5.5 (2.9)	−0.22 (−0.81, 0.38)	.47
Doctor’s attention
Intervention	112	1.4 (0.6)	74	1.6 (0.6)	0.15 (−0.07, 0.38)	.18	0.27 (0.03, 0.52)	.002b	.14
Control	109	1.3 (0.7)	83	1.2 (0.7)	−0.04 (−0.28, 0.20)	.75
Information about tests
Intervention	110	1.4 (0.6)	71	1.5 (0.6)	0.08 (−0.15, 0.31)	.47	0.23 (−0.01, 0.48)	.001b	.28
Control	106	1.2 (0.7)	81	1.3 (0.7)	0.03 (−0.22, 0.27)	.83
Symptom assessment
Intervention	109	1.5 (0.7)	71	1.5 (0.7)	0.03 (−0.21, 0.26)	.82	0.25 (0.01, 0.51)	.0001b	.26
Control	103	1.2 (0.8)	79	1.2 (0.6)	0.01 (−0.25, 0.26)	.93
Follow-up
Intervention	110	1.3 (0.7)	71	1.5 (0.7)	0.21 (−0.03, 0.44)	.09	0.39 (0.13, 0.65)	.001b	.02
Control	105	1.2 (0.7)	80	1.1 (0.7)	−0.09 (−0.35, 0.17)	.49
Doctor availability
Intervention	110	1.1 (0.7)	74	1.6 (0.6)	0.35 (0.11, 0.60)	.005	0.48 (0.20, 0.73)	.001b	.006
Control	106	1.1 (0.7)	80	1.0 (0.7)	−0.05 (−0.32, 0.23)	.74

^aLinear mixed models adjusted for time and baseline social isolation and hospitalizations.

^bAdjusted for multiple comparisons.

Symptom burden scores declined more on average for intervention patients (−3.53, 95% CI −5.53 to −1.53, p=0.001) than controls (−0.69, 95% CI −2.59 to 1.21, p=0.48), but the difference was not significant (−1.92, 95% CI −5.22 to 1.37, p=0.25). For the sample of 229, the power to detect statistically significant differences between arms for quality of life and symptom burden was 72% and 21%, respectively.

Overall satisfaction with care was significantly greater in the intervention arm at 12 months in both the adjusted model (2.26, 95% CI 1.46 to 3.06, p<0.0001; effect size, 0.74) and worst ranked score analysis (p=0.007). Satisfaction was also greater for intervention than control patients in the follow-up and doctor availability domains of the FAMCARE on worst rank score analysis.

Per-protocol analyses that accounted for delayed or no receipt of HBPC by intervention patients had results similar to those of the primary analyses (Table S1).

Hospitalizations and Emergency Department Visits

From baseline to 12 months, 17 (14.9%) intervention and 18 (16.1%) control patients had one or more hospitalizations (Table 3). In adjusted analysis, fewer intervention patients than controls experienced an incident hospitalization (−17.9%, 95% CI −31.0% to −1.0%; p=0.001; number needed to treat 6, 95% CI 3 to 100), although the total number of hospitalizations per patient were not significantly different (−0.54, 95% CI −1.35 to 0.25; p=0.18). ED visits occurred for 18 (15.8%) intervention and 17 (15.2%) control patients and there were no significant differences in the proportions of patients with incident ED visits or number of ED visits per patient in the adjusted analyses. Results of the per protocol analyses for hospitalizations and ED visits were similar to those of the primary analyses (Table S2).

Table 3.

Hospitalization and Emergency Department Visits, Intention to Treat Analyses

	12 Months Pre-Randomization	12 Months Post-Randomization	Within Group Difference		Between Group Differencea
			Δ, 95% CI	P	Δ, 95% CI	P
Hospitalizations
Patients with any event
Intervention	54 (47.4%)	17 (14.9%)	−35.4% (−43.0%, −23.1%)	<.0001	−17.9% (−31.0%, −1.0%)	.001
Control	37 (33.0%)	18 (16.1%)	−17.5% (−28.2%, −6.2%)	<.0001
# Events per patient, mean (95% CI)
Intervention	0.94 (0.66, 1.22)	0.22 (0.11, 0.33)	−1.50 (−2.03, −0.98)	<.0001	−0.54 (−1.35, 0.25)	.18
Control	0.49 (0.34, 0.64)	0.20 (0.11, 0.29)	−0.94 (−1.51, −0.37)	.001
ED Visits without Hospitalization
Patients with any event
Intervention	32 (28.1%)	18 (15.8%)	−12.9% (−20.2%, −4.0%)	<.0001	1.2% (−10.5%, 12.4%)	.75
Control	32 (28.6%)	17 (15.2%)	−14.0% (−21.8%, −4.6%)	<.0001
# Events per patient, mean (95% CI)
Intervention	0.59 (0.36, 0.82)	0.26 (0.13, 0.40)	−0.85 (−1.36, −0.33)	.001	−0.03 (−0.77, 0.70)	.70
Control	0.54 (0.32, 0.75)	0.25 (0.11, 0.39)	−0.81 (−1.34, −0.28)	.003

ED, emergency department.

^aGeneralized linear mixed models adjusted for time and baseline social isolation and hospitalizations.

Process of Care

Among intervention patients, the median number of office visits between randomization and before the first HBPC visit was 2 (range, 0–5) and the median number of home visits was 6 (2–15) (Table 4). Compared to controls, intervention patients had more telephone and social work encounters, and assessments of cognition, physical function, and falls. At baseline, 9 (7.9%) intervention and 7 (6.3%) control patients had a signed MOLST form in the medical record and 28 (24.6%) and 13 (11.6%), respectively, at 12 months (p=0.01). Also at 12 months, DNR or DNI orders were documented for 46 (40.3%) intervention and 23 (20.5%) control patients (p<0.0001), and do not hospitalize orders were documented for 7 (6.1%) intervention and no control patients.

Table 4.

Process Measures, 12-month Period of Observation

	Assignment
	N	Intervention N=114	N	Control N=112	P
# Days to HBPC telephone triage, mean (sd)a	72	25.8 (23.1)	-	-	-
# Days to first home-based primary care visit, mean (sd)a	74	86.3 (56.7)	-	-	-
# Days in HBPC, mean (sd)b	74	254 (82)	-	-	-
# Days to any in-person clinical encounter, mean (sd)a	93	57.3 (56.2)	87	59.8 (59.8)	.95
# Office-based clinical visits prior to first HBPC visit, median (range)	74	2 (0–5)	-	-	-
# Home-based primary care visits, median (range)	74	6 (2–15)	0	0	-
# Office-based primary care visits, median (range)	58	2 (1–14)	86	4 (1–13)	.003
Among intervention patients with no HBPC visits	23	5 (1–14)	-	-	-
# Telephone encounters with primary care, median (range)c	87	6 (1–34)	64	3 (1–14)	<.0001
Had specialty care referral	63	55.3%	49	43.8%	.08
# Specialty care referrals, median (range)		2 (1–13)		2 (1–8)	.90
Had any social work encounter	84	73.7%	43	38.4%	<.0001
# Social work in-person encounters, median (range)	33	1 (1–6)	13	2 (1–7)	.19
# Social work telephone encounters, median (range)	82	7 (1–37)	38	3 (1–17)	<.0001
Assessments
Falls	88	77.2%	73	65.2%	.046
Cognition	45	39.5%	10	8.9%	<.0001
Physical function	57	50.0%	30	26.8%	.0003
Signed MOLST form in medical record
At baseline	9	7.9%	7	6.3%	.63
At 12 months	28	24.6%	13	11.6%	.01
Documentation of DNR or DNI order
At baseline	6	5.3%	8	7.1%	.56
At 12 months	46	40.3%	23	20.5%	<.0001
Documentation of Do Not Hospitalize order
At baseline	0	-	0	-	-
At 12 months	7	6.1%	0	-	-

^aFrom date of randomization.

^bFrom date of randomization to study completion (12-month follow-up, death, or drop out).

^cIncludes physician, nurse practitioner, or nurse encounter.

MOLST, medical outcomes of life sustaining treatments; DNR, do not resuscitate; DNI, do not intubate.

P-values were calculated per intention-to-treat analysis.

Deaths

During the study, 24 (21.1%) intervention and 12 (10.7%) control patients died (p<0.0001). The average time from randomization to death was 188 (114) days and 165 (115) days for intervention and control patients, respectively (p=0.39). In the intervention arm, decedents were more likely than patients who completed the study to be male, had more ADL and IADL impairments, and were more likely to use a wheelchair (Table S3). There were no differences in the number of days from randomization to first HBPC visit (70 [28] vs. 90 [60], respectively, p=0.60) or number of office-based visits prior to the first HBPC visit (2.2 [1.9] vs. 2.1 [1.4], p=0.91).

Among all patients, deaths occurred for 27.5% with and 9.8% without a DNR (p=0.0004). A MOLST with DNR was documented for 33.3% of intervention patients who died vs. 8.3% of control patients who died (p=0.10). Additional details of the associations of subject characteristics with death are shown in Tables S3–S6.

Independent physician reviewers identified no adverse events and attributed no deaths to participation in HBPC or to delay in care from randomization to first HBPC visit. There were insufficient data in the available medical records to adjudicate 3 cases.

Discussion

Homebound patients in HBPC had greater satisfaction with care and a lower hospitalization rate than those in office-based primary care, and higher rates of advance care planning. There were no differences in quality of life, symptom scores, number of hospitalizations per patient and ED visits. The impacts of HBPC are notable given the participants’ high level of illness burden and dementia, and trial under enrollment at 65% of goal. Overall, the study’s findings align with observational research that has demonstrated higher satisfaction with HBPC and fewer hospitalizations,^{6, 8, 27} with the exception of a recent matched control study that found no difference in hospitalization rates.²⁸

Our findings are also consistent with a 1985 randomized trial of HBPC that documented reduced costs for HBPC patients,⁹ and the Veterans Affairs (VA) Team-Managed Home-Based Primary Care randomized controlled trial, conducted from 1994 to 1998.¹⁰ The VA trial demonstrated better satisfaction with care and quality of life for HBPC patients. The VA study was different, however, as it recruited hospitalized patients. The current study recruited primarily from outpatient practices and the mostly female and minority demographics of its participants contrast with the predominantly male (96%) and white (63%) make-up of the VA study. Our study is more generalizable to the growing number of older adults who become homebound each year, 69% of whom are female and increasingly racially diverse.²⁹

A notable finding of this trial was the higher death rate in the intervention arm, which raises the possibility that delayed care or HBPC caused harm. With respect to delays, all patients continued to receive office-based care while they waited for HBPC to begin, there was no difference in time from randomization to follow up care between the two arms, and intervention patients had more clinical visits and telephone encounters overall. Moreover, independent reviewers found no deaths attributable to delays, or to care delivered in the home. Nonetheless, the higher death rate among intervention patients in the context of a randomized trial cannot be dismissed as artifact.

One potential explanation is less aggressive care for intervention patients, reflecting appropriate care. With an average 2-year mortality rate of 40%,³⁰ many homebound older adults, like others with advanced illness,³¹ may prefer less aggressive care. A common feature of HBPC is the effort of physicians to elucidate patients’ goals of care and provide goal-concordant care.³² Consistent with this observation, intervention patients were more likely than controls to have had newly documented advance directives after randomization. DNR orders have been associated with higher mortality in previous studies.^33–35 Alternatively, more deaths could have arisen from inappropriate care or under treatment independent of patient preferences. While HBPC was superior to office care for quality measures (assessments, advance directives) in this study, this study was not designed for a comprehensive comparison of quality between the two practice modalities.

At a time of growing awareness of inequities in healthcare delivery and outcomes, highlighted by the experiences of racial and ethnic minorities during the COVID-19 pandemic, HBPC represents a solution for many underserved and marginalized Americans. In addition to overcoming barriers to access, home visits provide clinicians a direct view of the conditions in which patients and caregivers manage great medical complexity, enabling them to engage community-based resources and coordinate care to address many social determinants of health encumbering this population. The successes of HBPC for the predominantly Black and Latinx patients enrolled in this study may provide lessons for care of other at-risk populations.

Congress recognized the promise of HBPC by authorizing the Center for Medicare & Medicaid Services Independence at Home (IAH) demonstration, a shared savings program for HBPC providers.³⁶ While representing only 6% of the Medicare FFS population, IAH-eligible adults account for 30% of Medicare parts A and B spending, 24% of hospitalizations, and 38% of long term care admissions.³⁷ IAH saved $25 million in its first year and earned bipartisan support for a 2-year extension. Nonetheless, HBPC remains a scarce resource. Only 11% of homebound adults received HBPC between 2011 and 2017 and there was no net growth in HBPC during this time³⁸ even as shared savings programs expanded. Access is especially limited in rural communities. While 24% of homebound adults reside in rural areas, 96% of HBPC recipients live in cities and suburban communities.³⁸ Meanwhile, the physician workforce skilled in the care of frail elders is concentrated in nursing homes.³⁹ Low reimbursements for HBPC are the principal impediment to its expansion, insufficient to support the necessarily lower per-provider patient volumes and higher intensity care management it requires. Various pathways to funding HBPC expansion have been proposed,³⁷ but expansion will be limited without better fee-for-service reimbursement as long as fee-for-service predominates.

This study experienced a number of challenges that limited conclusions about the effects of HBPC. It was underpowered and many patients never received the assigned HBPC intervention or received it after extended delays. There was a loss of data on quality of life and symptoms (pain) that limited analyses of those outcomes. The study eligibility criterion for prior hospitalization was determined by self-report, which could have been misreported. The single, urban, academic medical center-based HBPC program setting may limit generalizability, though care provided by this program is similar to other HBPC programs in the U.S.⁴ Finally, there were statistically significant imbalances in baseline hospitalizations and social isolation, although we controlled for these in outcomes analyses.

In conclusion, HBPC improved satisfaction with care for homebound older adults and reduced hospitalization rates, but it was also associated with more deaths, a finding that warrants additional evaluation. Because of the study’s early termination, additional research may also be needed to determine the effects of HBPC on quality of life, symptom burden and ED visits.

Key Points

• Homebound older adults in home-based primary care reported greater satisfaction with care and had lower hospitalization rates than patients receiving office-based primary care

• HBPC was also associated with higher rates of advance care planning

• The study was terminated early; some outcomes may need to be examined through new research

Why does this matter?

Home-based primary care improves satisfaction of care among homebound patients and reduces hospitalizations.