Mass General Brigham shows what home-based palliative care delivers at the end of life
Mass General Brigham's home-based palliative care team cares for seriously ill patients where they live. Working with Radial, it used AI to find the patients who need that care, then measured what its program delivers: against a matched group, more of its patients enrolled in hospice, fewer went back to the hospital in their last months, and medical spending outside hospice fell by millions.
Hospice use up from 62.7% to 76.3%
182 fewer ED visits and 96 fewer admissions in the last 3 months of life
$6.4 million less non-hospice spending in the last year of life
The challenge: the patients who need palliative care most are the hardest to find
The population is aging fast, and more people are living with several chronic conditions. Half of Medicare beneficiaries who live in the community are homebound in the year before they die. Just 5% of patients drive more than half of all health care spending, through 911 calls, emergency visits and hospital admissions, and much of that care is low value and out of step with what patients want.
Home-based palliative care answers that need. An interdisciplinary team of physicians, nurse practitioners, nurses and social workers visits seriously ill patients at home, manages their symptoms, teaches patients and caregivers what to do when things get worse, and helps them plan their care, including advance directives. Studies link timely palliative and end-of-life care to better quality of life, more time at home at the end of life, fewer avoidable hospitalizations and a better experience for bereaved families.
Yet health systems struggle to adopt it. They have trouble finding the right candidates in time, need institutional support to run and grow a program, and rarely have the bandwidth or technical skills to show its return to internal and external stakeholders. Mass General Brigham (MGB) partnered with Radial to tackle two of those barriers: finding the right patients with AI, and proving the program's value.
- 35%
- more patients enrolled in hospice
- 240%
- longer median stay on hospice
- 36.3%
- estimated cost savings across 11 studies
Source: Lustbader D, et al. The impact of a home-based palliative care program in an accountable care organization. J Palliat Med. 2017;20(1):23-28.
Why prognostic tools fall short
Clinicians have long used scores to judge who might benefit from palliative care, but most were built on patients with cancer, predict only short-term survival, and depend on subjective inputs. The Palliative Performance Scale misses decline in advanced dementia and frailty, and the Levine Prognostic Index, from a single site, doesn't measure function at all. Machine learning on claims and health record data screens patients automatically, and does markedly better.
| Tool | What it predicts | What it uses | Accuracy (AUROC) |
|---|---|---|---|
| Palliative Performance Scale | Median survival in days | Clinician ratings of mobility, activity, intake, self-care and mental status | 0.67 to 0.74 |
| Levine Prognostic Index | One-year mortality | Age, length of stay, prior nursing home discharge, metastatic cancer and other conditions | 0.70 |
| Machine learning (Brigham and Women's Hospital) | Six-month mortality | Medicare claims | 0.82 |
| Deep learning (Stanford) | Three- to 12-month mortality | Health record data | 0.87 |
The solution: AI that finds the right candidates for palliative care
Radial Care uses machine learning to find the patients most likely to benefit from palliative care, including the ones who aren't obvious, and explains why each one was flagged.
Read the baseline
Longitudinal claims show how well each patient's disease is controlled and where it is heading: utilization history, chronic conditions, frailty and patient-reported symptoms.
Catch the turning points
Live hospital feeds, health record data and social risk indexes (the Area Deprivation Index and Social Vulnerability Index) reveal inflection points in a patient's course, such as a new diagnosis or a hospital stay.
Ask nothing of clinicians
Hundreds of objective variables and several mortality end points are weighed automatically, in real time, with no data entry.
The partnership: from finding patients to proving value
MGB and Radial started small, followed the results, and changed course when MGB's needs changed.
- 1
Prove it in the hospital
At Newton-Wellesley Hospital, MGB tested AI-driven referral to inpatient palliative care. MGB's data showed that more than 10% of its MSSP ACO members died within 30 days of leaving a skilled nursing facility, and that 90-day mortality approached 30% for patients with three or more such stays. Inpatient palliative care consults rose 24%, outpatient palliative and hospice referrals rose too, and potential savings in total medical expenditure reached $850,000 for about 1,500 hospitalized patients. Clinicians used the surprise question more often, worked more closely with case management, and cited fewer barriers to involving palliative care.
- 2
Bring it home
MGB first looked at using Radial Care in the Massachusetts General Hospital emergency department, but its embedded palliative care team already had a strong triage process. MGB's leaders saw a much better fit with the home-based program. Radial built a palliative care score that ranked a random sample of 1,000 MGB ACO members by actionability, with the insights behind each score, so a stretched team could reach the most appropriate patients, including non-obvious ones in the community.
- 3
Prove the program's value
Turning patient identification into new workflows meant facing the program's own situation: its census had been stretching the team's capacity for a year, and a hybrid model with virtual visits was about to begin. MGB shared its goals openly with Radial, and Radial came back with a proposal for a thorough evaluation of the program.
"Many of the patients identified would benefit from home-based palliative care."
How the evaluation worked
Radial analyzed MGB's MSSP ACO claims from 2020 to 2025. The program group was the 690 home-based palliative care patients who died between January 2021 and May 2025 after at least one visit from the program in their last 180 days of life, or who died on hospice, each with at least a year of claims before they enrolled. They are about half of the program's patients: those outside the ACO, those who left it before death, and those dually eligible for Medicare and Medicaid were left out.
Each patient was paired with an ACO member who never enrolled in the program, matched on sex, five-year age band and disease group, and then on the closest claims-based frailty index.
| Patients | Home-based palliative care | Matched group |
|---|---|---|
| Age at death | 87.3 | 87.5 |
| Women | 62% | 62% |
| Claims-based frailty index | 0.2366 | 0.2361 |
| Advance care plan on record | 64.9% | 64.9% |
| History of falls | 66.8% | 64.2% |
| Do-not-resuscitate order on record | 84.2% | 77.2% |
| Prior MGB home care | 82.9% | 61.3% |
The results
Compared with the matched group, the program's patients enrolled in hospice more often and sooner, went to the hospital less in their final months, and cost far less outside of hospice.
| In the last year of life | Home-based palliative care | Matched group |
|---|---|---|
| Enrolled in hospice | 76.3% | 62.7% |
| Average days on hospice | 88.1 | 59.1 |
| Total days on hospice | 46,447 | 25,584 |
That is 20,863 more days of hospice care, and a median hospice stay 12 days longer than the national median. In the last three months of life, the program's patients had 96 fewer hospital admissions and 182 fewer emergency department visits, and 308 fewer emergency visits over the last year.
Spending on hospice rose by $3.5 million, as expected when more patients choose it, while all other medical spending in the last year of life fell by $6.4 million: $9,283 per patient, or about $120,858 for every month of the program.
- Hospice spending
- $3.5 million more$3.3 million more in the last 3 months of life
- All other medical spending
- $6.4 million less$4.6 million less in the last 3 months of life
What MGB learned
- An AI-driven approach can safely and effectively identify more of the right patients for palliative care referral.
- External partnerships can be powerful opportunities for innovation and growth when they are managed with transparent, two-way communication.
- Health systems should feel empowered to be honest about their needs, especially when those needs change over time.
- Partnerships work best when goals are clear and aligned: accelerating AI-enabled solutions, standardizing data-driven, evidence-based risk stratification, and filling internal gaps in proving a program's return and putting a solution into practice.
What's next
This was the most comprehensive and impactful evaluation in the 10-year history of MGB's home-based palliative care program. Its results are now the backbone of the slides the team uses to show the program's value to internal and external stakeholders. MGB is now designing a new home-based program for older patients with heart failure, and the work with Radial has opened several opportunities for growth and support.
As presented at the American Academy of Home Care Medicine (AAHCM) 2025 Annual Meeting
Mark Stoltenberg, MD, MPH, and Pamela Williams, CNP (Mass General Brigham), with Kerry Son, MSN, FNP, and Anant Vasudevan, MD, MBA (Radial)
Radial technology development was funded in part by the National Science Foundation and the National Institutes of Health.
A partner, not a portal
Find the patients who need palliative care, and show what your program delivers.
Bring one question about your population. We'll show you the platform on illustrative data, then size the opportunity on yours, with the team who would work beside you.
- A named clinical and technical lead
- Quarterly business reviews
- On-site visits and training