What PC 3.0 Missed About Hospice

After I published Palliative Care 3.0, I started considering whether I made an argument that would hollow out hospice. The logic looked tidy enough. Move care upstream, tier it by competency, reach people earlier in the illness — and the organizations financed by hospice per diem end up with fewer days.

At the same time, I've spent two decades watching hospices launch home-based palliative care programs. Some of those programs are superb. Almost none of them have grown the way a working feeder would grow, and a fair number have closed without much of a eulogy. That is not the pattern you would expect from an upstream argument draining a downstream benefit.

Here's where I landed: those programs don't feed hospice because a good share of their patients were never upstream of hospice at all. Something sits between the two that my tiers don't describe.

The Feeder Thesis and Its Missing Growth Curve

The reasoning behind hospice-operated home-based palliative care is sound: find patients earlier. Build the relationship before the crisis. When the prognosis tightens and eligibility finally arrives, election becomes a conversation with someone the family already trusts instead of a cold pitch from a stranger in a bad week.

Two studies back it. Chris Kerr and colleagues studied Home Connections, a program built by a nonprofit hospice with two commercial payers, and found hospice entry at 70 percent against 25 percent in usual care, with median stay of 34 days against 9. Dana Lustbader and colleagues, working inside a Medicare Shared Savings ACO, reported a 35 percent higher hospice enrollment rate and median stay of 34 days against 10.

Those are real findings from real programs, and I'm not waving them off.

The largest study found nothing of the kind. Kaiser Permanente Southern California compared 7,177 decedents who received home-based palliative care against 25,102 who received hospice alone. Median hospice length of stay: 13 days against 12. Nothing.

And then the finding that isn't a study, which is the one I keep returning to. If hospice-run palliative care converted the way we've assumed, twenty years would have produced scale. Instead we get the same arc, over and over: a launch, a few strong years, a plateau, a board conversation.

Execution failure produces variance. Somebody, somewhere, cracks it and grows. What I see instead is a ceiling that sits in roughly the same place whether the program is excellent or merely adequate — which is what finally made me wonder whether the conversion problem is a conversion problem.

The Model That Was Not Upstream

The Medicare Care Choices Model gets cited as proof that upstream palliative care increases hospice use. Enrollees elected hospice at 83 percent against 65 percent in a matched comparison group and entered earlier. The evaluation also reports more hospice days per decedent, a figure that counts everyone who never elected as zero days and therefore blends the election rate with duration. Read carefully.

What that is glossing over is the eligibility criteria. MCCM required beneficiaries to already meet hospice eligibility: certified terminal illness, six-month prognosis, one of four qualifying diagnoses, no current election. The model enrolled people who already qualified and had said no, then paid participating hospices $400 a month to keep showing up.

That is a study of conversion among eligible decliners. It tells you very little about patients who are years from eligibility, which is precisely where my tier argument lives.

The generalizability is thinner still. Five hospices produced 46 percent of all enrollees. Fewer than one percent of eligible beneficiaries in participating markets ever enrolled. Participating hospices skewed large and nonprofit, and plenty of them told evaluators the payment didn't cover the work.

The Arithmetic of a Unit of Blood

Picture the patient with myelodysplastic syndrome who needs red cells every couple of weeks (and then every couple of weeks) to stay out of heart failure.

Hospices generally decline palliative transfusions, citing cost alongside logistics and, occasionally, philosophy. The arithmetic explains most of it. A unit of packed cells, plus typing and delivery and nursing time and monitoring, lands in the same neighborhood as a full day of routine home care payment. One transfusion eats a day. Two a month eats the patient's margin outright, and the hospice absorbing that loss is often the nonprofit already running in the red.

Transfusion-dependent Medicare beneficiaries with leukemia had a 51 percent shorter hospice length of stay than transfusion-independent patients. SEER-Medicare data show markedly lower hospice use among transfusion-dependent patients with myelodysplastic syndromes. Patients with hematologic malignancies who enroll average around 9 days against a national figure near 17; enrollment in AML runs somewhere between 23 and 41 percent; more than a quarter of AML patients who elect do so in their final three days.

We've also run the experiment in reverse, at small scale. BRUOG-407 paid for palliative transfusions for hospice-eligible patients with aggressive hematologic malignancies. Median stay came in at 13 days. Nobody enrolled for three days or fewer. Sixty percent stayed longer than nine days, and every single patient died outside the hospital. Ten patients, so I'm not going to overstate it — but the direction is clean.

I want to be careful here, because this is the point where a piece like this usually turns into a story about villains. It isn't one. The hospice that declines the transfusion is doing arithmetic that the MHB forced on it in 1983 and has never revisited, and the medical director who says "no" is often the same person who spent an hour on the phone trying to find someone who could say yes. Nobody in this story is behaving badly. That's what makes it structural.

These patients are not hospice-averse. They are hospice-excluded, by an arithmetic nobody designed and nobody owns. And a coverage exclusion is immune to everything my framework proposes: earlier contact, better prognostic communication, more skilled clinicians at every rung. This patient has a payment problem wearing a delivery problem's clothes.

The Second Population

The other group I meet who select home-based palliative care are the people who watched a parent's hospice go badly.

I can't support that with published work. I went looking. The literature on hospice-experienced families measures bereavement outcomes, not the election behavior of the relatives who survive them. What I have instead is twenty years of conversations that open with some version of we did that with my mother.

Both populations surface in the disparity data, though the numbers can't separate them from everything else living there. In 2024, 55.6 percent of white decedents used hospice, against 40.9 percent of Black decedents and 40.4 percent of Hispanic decedents. Among beneficiaries with end-stage renal disease: 31.4 percent. Dialysis sits behind the same wall as transfusion, for the same reason, which makes that ESRD figure the cleanest picture I can offer of a coverage exclusion masquerading as a preference.

A family failed once, in a community failed repeatedly, is running a forecast. We keep coding that as a knowledge deficit and building education programs to fix it. A topic for another day (and perhaps a guest author) I suppose.

What I Got Wrong About the Shape

Palliative Care 3.0 describes competency tiers, with patients ascending as their needs intensify. That picture assumes a continuous surface: add capability at each rung, calibrate specialist engagement to patient state, and people move up as their illness demands.

But now we see that there's a wall partway up (is this metaphor getting tortured...yes). It's built from what a per diem can't absorb and from what families remember, and some patients will never cross it no matter how much need they accumulate or how good the team below it becomes.

Which means home-based palliative care isn't a stage on the way to somewhere else. For a meaningful share of its patients, it's the destination — and financing it as a waiting room is a fairly complete explanation for the growth curve I've spent twenty years watching.

The correction to my own model is narrow, and I think it matters. Before we specify what each tier should be able to do, we have to specify which patients can reach it. Otherwise the framework keeps generating programs that perform beautifully and convert nobody, and people like me keep writing puzzled essays about why.

There's concrete work available here, and it isn't glamorous. We have never inventoried the exclusions. We know about transfusion because hematology got loud about it and produced a decade of data. Dialysis has a smaller literature and a louder patient community. Ventricular assist devices, prolonged parenteral antibiotics, some of the newer subcutaneous infusions — each has a handful of case series, a workaround somebody invented locally, and no national count. A field that wanted to know the shape of its own wall could assemble that inventory in a year if we still had a registry.

Counting the bricks would tell us something a taxonomy cannot: how many patients are sitting in home-based palliative care because it suits them, and how many are there because the door above them is bolted. Those two groups need different things from us, and right now we serve them identically and call it access.

Counterarguments

These programs failed on execution, not population. Possible. But execution failure shows up as variance, and in twenty years somebody would have solved it and scaled. The uniformity of the ceiling is what pushes me toward structure.

Transfusion-dependent patients are too small a group to hold this up. They are small. They're also the visible case of a general mechanism, and the wall has other bricks: dialysis, ventricular assist devices, prolonged parenteral antibiotics, certain infusions. Each has its own small literature and its own local workaround. Nobody has counted them together, which is itself worth someone's dissertation.

Hospices could cover transfusions if they chose. Some larger ones do, which tells you the constraint tracks margin rather than conviction. In 2023, for-profit hospices ran a 13.7 percent Medicare margin; nonprofits ran negative 1.3; hospital-based providers ran negative 25.6. The capacity to absorb the exception is thinnest at exactly the organizations whose structure most resembles what our field says it admires. I've written about that squeeze before.

This indicts Palliative Care 3.0 more than it indicts anything else. Partly, yes. But I think that the model survives with a correction to its topology and I'd rather publish the correction than defend the drawing.

Where I Might Be Wrong

The bad-experience population may be far smaller than I think. I'm building part of a structural claim on clinical impression, and impression is where a clinician's sample bias does its finest work.

Capital and contracting may explain the growth curve without any help from me. Home-based palliative care has no dependable payment source in fee-for-service, and a hospice board staring at a program that loses money every year has every reason to close it with no population mismatch involved. That explanation is simpler than mine, and I can't rule it out.

BRUOG-407 is ten patients. Lifting a coverage barrier might produce enrollment without producing benefit, and a larger trial could show transfusion-dependent patients entering hospice only to receive care that suits them no better than what they left.

And the wall may be lower than I'm describing. Hospices make quiet exceptions all the time, when the patient and the numbers allow. If that discretion is more common than anyone has measured, then what I'm calling structural is actually a matter of who your admissions nurse knows — which is a worse problem, and a different post.

Final Thoughts

There's a wall between PC 3.0 and hospice. Some of the work dismantling it falls to me to explain in my model and much of that falls to the health policy work of many to modernize the Medicare Hospice Benefit.

Ultimately though I don't think the wall comes down. I think it moves, and I think we lose the ability to see it.

More than half of Medicare beneficiaries are now in Medicare Advantage. Plans have wanted the last year of life for a decade, and the hospice carve-in they were handed ended in 2024 in part because nobody could say what was being bought. They'll be back, because the money is still sitting there. When they return, the exclusions that live inside a per diem today will live inside a network contract instead: negotiated, proprietary, varying by market, and considerably harder to notice than a line in the Federal Register.

That's the version where my tiers survive as vocabulary. Something for a brochure and a job posting, describing care that nothing in particular pays for.

There's a better path, and part of it is already built. HOPE did something this program has never done: it made an assessment finding trigger a required visit. An obligation attached to a patient's symptom instead of an agency's org chart. That's the mechanism Palliative Care 3.0 has been missing, and hospice got there first. Its assessment window closes at day 30. If that window is still shut three years from now, I'll have been wrong about which direction this was heading.

So we have an interval. My guess is about five years. Long enough to count our exclusions while they're still public and countable, short enough that if we spend it arguing about what to call ourselves, we'll end up negotiating that wall one contract at a time, in rooms our patients will never enter.

Rounds & Rants represents my views — not those of UCSD, AAHPM, CMA, AMA or any organization where I hold a role. I don't write on their behalf and they don't vet what I publish.