Time to Move On From The Buckets

The generalist-specialist framework got us here. It can't get us where we need to go.

Last week I argued that the taxonomy is lagging the trials—that the generalist/specialist binary can't describe the work our most rigorous evidence now shows is effective. Stepped care, default referrals, tiered delivery: none of these organize the world into two boxes.

This week I want to go further: the binary wasn't just a descriptive framework that happened to become inadequate. It shaped what the field chose to build for a decade. And some of what we built was the wrong thing.

I've made versions of this argument before. In Palliative Care 3.0, I argued that we can't keep defending the specialty by keeping it small—we defend it by making it legible, structured, and scalable. In a piece that preceded it, I asked whether the field spent a decade chasing "everyone with serious illness needs palliative care" while ignoring the harder constraint: the payment architecture that makes real teams possible. This post connects those arguments to a specific mechanism that held us in place. The binary.

What the Binary Was Built to Solve

The setup was honest. By 2010, Lupu's Workforce Task Force analysis estimated the gap between supply and demand for HPM physicians at somewhere between 6,000 and 18,000 individual physicians. One hundred twenty fellowship slots per year against a denominator that Morrison and Kelley were about to widen from "the dying" to "the seriously ill." The math was brutal.

Quill and Abernethy's 2013 NEJM perspective was the field's answer to that math. They split palliative care skills into two tiers: primary (basic symptom management, goals-of-care conversations, care coordination, done by any clinician) and specialist (refractory symptoms, complex family dynamics, existential distress, done by the fellowship-trained team). The logic was clean: teach non-specialists the basics, reserve specialists for the hard stuff, stretch a small workforce across a large denominator.

This was a reasonable framework. I want to be explicit about that, because the critique I'm about to make can sound like I think the people who built the binary were wrong. They weren't wrong. They were working a math problem with the tools available in 2013, and the answer they produced was serviceable for its moment. PCHETA gets two funding tracks out of it. CAPC built a training architecture around it. The field's workforce advocacy found a clean story to tell Congress. Real policy work (some of it done by me). It matters.

But frameworks built for one problem have a tendency to persist into problems they weren't designed for. The binary persisted.

What the Field Built on Top of It

Once you have a binary, everything sorts into one box or the other.

The training investment went into "primary palliative care." CAPC's clinical training tracks, VitalTalk-adjacent communication curricula for non-specialists, the NCI's push for palliative care access metrics—all of it organized around the idea that the main lever for scale was teaching non-specialists to do basic palliative care skills. Workforce advocacy went into the specialist box: more fellowship slots, more academic positions, more protected time. Two tracks, two strategies, two funding pitches.

The access metric becomes the field's scoreboard. The question we track—in scorecards, in quality measures, in trial design—was: did the patient see palliative care, yes or no? That is a binary question for a binary taxonomy. It tells you whether the gate was opened. It tells you nothing about what happened on the other side.

And the research agenda followed. Trials tested "early specialist palliative care" (the Temel 2010 model) against usual care, with the implicit question being: does putting a specialist in the room improve outcomes? The answer was usually yes, but the trials couldn't tell us which part of specialist contact mattered, or whether the same components could be delivered differently, at a different dose, by a different team member at a different moment. The binary didn't let us ask those questions. (I'm going to be more fair here rather than below, because Temel tried to help us understand exactly what was happening behind the 2010 data.)

Here is what it cost us.

What the Binary Got Wrong

It treated substitution as the scale strategy, and the substitution evidence is weak. The CONNECT trial (Schenker et al., JAMA Intern Med 2021) tested an oncology nurse-led "primary palliative care" intervention for patients with advanced cancer. No improvement in patient-reported outcomes at 3 months. Ernecoff et al.'s comparative analysis (J Palliat Med 2020) put the evidence side by side: primary palliative care interventions were less comprehensive, less likely to show benefit on physical symptoms, and at higher risk of bias than specialist interventions.

These findings have received less airtime than they deserve. I think it's because the binary made substitution seem like the obvious answer—of course you train generalists, the specialists can't see everyone—and the null results feel like a threat to the whole scaffold. But the results don't mean non-specialists can't do palliative care work. They mean the binary's version of that work (training generalists in a diffuse skill set and hoping it translates to behavior) is the wrong implementation. The precision version—specific competencies targeted to specific clinicians at specific moments with supporting infrastructure—has yet to be adequately tested.

It created a referral-centric model that confuses the gate with the destination. BE-a-PAL (Parikh et al., JAMA Netw Open 2025) is the definitive illustration. Referral rates quintupled. Patient-reported outcomes didn't improve. The single initial consultation wasn't enough. The binary says: get the patient to the specialist. The continuum asks: then what? At what dose? For how long? With what intensity? Matched to what needs? Those are dose-response questions, and the binary has no framework for them.

The binary leaves critical competency failures unaddressed because they fell between the two boxes. 

Admane et al. (JCO Oncol Pract 2025) documented a five-fold decline in oncologist-prescribed opioid doses for cancer pain between 2016 and 2021 at MD Anderson. Corroborated at Yale (Baum et al., Cancer 2025). The 2022 CDC Guideline explicitly acknowledged that its 2016 predecessor had been misapplied to cancer and palliative care populations—populations the guideline was never meant to cover. A five-fold prescribing collapse in patients with cancer-related pain. The binary's answer—"refer to a pain specialist" or "train oncologists in primary palliative care"—misses both the scale and the specificity of the problem. The continuum's answer: this oncologist needs prescriber confidence, state-law literacy, and the nerve to call the pharmacy. The specialist team needs to be available for the genuinely complex cases. Different competencies, different moments, same patient.

Lai et al. (J Palliat Care 2023) found that internal medicine residents who provide recommendations in goals-of-care conversations learned the skill from observation on palliative care or critical care rotations. Not from didactics. Not from "primary palliative care training." The skill crossed the binary boundary through clinical proximity—the trainee saw it done, internalized it, and carried it into settings where no specialist was present. The binary can't describe that transfer. A competency continuum can.

The Tension

The binary is load-bearing for workforce advocacy, which has been my work for the better part of a decade. PCHETA's two-track funding model depends on a clean distinction between specialist and non-specialist training. The field's workforce projections (Kamal et al., Health Affairs 2019) are built on a demand model that assumes a particular version of specialist-centric delivery. If I'm arguing that the binary is inadequate, I need to be honest about what that means for arguments I've personally been part of making.

Here is what I think it means: the workforce arguments are stronger under a continuum framework, because a continuum specifies what specialists are uniquely for. Under the binary, the specialist tier is defined by subtraction: everything left over after you remove the basics. Under a continuum, the specialist tier is defined by expertise: refractory and complex symptom management, complex prognostic communication, team design and leadership, system architecture, training of other tiers, research. That's a cleaner argument to CMS, a more defensible argument against scope dilution (the Monster We Built problem), and an honest accounting of what fellowship training actually prepares you to do that observation and short courses do not.

I'm not arguing that specialist palliative care is unnecessary. I'm arguing that the binary category is not in service of the goal. When you can only describe the world as "specialist" or "generalist," you can't specify what the generalist tier actually needs to learn, you can't measure whether they learned it, and you can't build the infrastructure that makes competency reliable at any tier. The continuum gives you that.

What to Build Instead

The continuum framework I'm proposing has five operational components. None of them are original to me—they're assembled from what the trials and the international literature are already showing works.

1. Competency tiers, not clinician categories. Three or four tiers defined by specific, measurable competencies. The EAPC three-tier model (palliative care approach, general palliative care, specialist palliative care) is a starting template. TIER-PC is testing an operational version. What matters is that each tier has explicit competencies attached, explicit training requirements, and explicit infrastructure to make the competencies reliable.

2. Default-on infrastructure at every tier. BE-a-PAL proved that default referral works because the barrier is behavioral, not attitudinal. Apply the same logic within tiers: default GOC conversation templates in the EHR. Default opioid prescribing pathways for cancer pain with pharmacy escalation protocols. Default symptom screening at admission. Defaults, not training alone.

3. Dose-response specialist engagement. Temel's stepped care model is the proof of concept. Specialist contact calibrated to patient state over time, not to a fixed calendar. PRO-based triggers for step-up. This is how you stretch a small specialist workforce across a large denominator without pretending that substitution works as well as the real thing. (Start preparing your local oncologists now for the fact that your clinic will not be able to co-manage and prescribe for everyone moving forward.)

4. Competency transfer by proximity, not by curriculum. Lai's finding—that residents learn recommendation-giving through observation, not didactics—should reshape how we think about training the non-specialist tiers. Embed palliative care clinicians in the settings where other clinicians practice (ICUs, oncology clinics, heart failure programs) so that competency transfers through clinical proximity. That's what Palliative Care 3.0 looks like operationally: the specialist tier's job includes being present in the systems where other tiers work, not just receiving referrals from them.

5. Measurement that matches the continuum. The binary gave us a binary metric: did the patient see palliative care? The continuum needs a different scoreboard: was the right competency delivered at the right moment? We're not there yet. But PRO-based screening (Sedhom and Parikh's precision framework), sentinel-event triggers (Temel's stepped model), and EHR-embedded process measures can get us closer than a consult-yes/no checkbox.

Where I Might Be Wrong

The continuum framework is ahead of its evidence. No trial has compared a continuum delivery model to a binary delivery model head-to-head. TIER-PC will eventually contribute, but it hasn't reported yet.

I may be underweighting the binary's policy utility. "Specialist" and "generalist" translate into billing codes, accreditation standards, and legislative language. A continuum is harder to legislate. The pragmatic cost of better taxonomy may be political complexity that slows down advocacy we need right now.

The CONNECT null result does not generalize to all primary palliative care designs. The evidence base is heterogeneous enough that sweeping claims about primary palliative care failure are premature.

And the updated workforce demand model I'm implicitly calling for—one that incorporates stepped care, default-on infrastructure, and tiered delivery into the denominator math—doesn't exist yet. I'm asking the field to update its taxonomy before the modeling catches up. That's a risk.

But the alternative is continuing to build on a scaffold that the field's own trials are dismantling from the inside.

Quill and Abernethy gave the field what it needed in 2013: a clean binary that solved a workforce math problem and gave us a policy story to tell. Thirteen years later, the math has changed. The evidence has changed. The delivery models our best researchers are building don't use the binary. They use tiers, triggers, defaults, dose-response logic, and precision targeting.

I wrote in Palliative Care 3.0 that the mission hasn't changed: reduce avoidable suffering and maximize quality of life. Everything else is negotiable. The taxonomy is part of "everything else." It should describe the work. It should sharpen our advocacy. It should help us build better systems.

Right now, it's past its prime. The trials moved on. The international frameworks moved on. The patients were never in one of two boxes to begin with.

Time to catch up.

I am a palliative care physician, educator, and professional strategery expert known for turning rounds into rants and rants into teaching points.