As a primary care physician at a safety-net clinic in San Francisco, I recently saw a patient I’ll call Maria — a 52-year-old woman with diabetes who cleans houses for a living. She’d been hospitalized twice in the past year for diabetic complications, racking up tens of thousands in hospital bills. Each time, she was discharged with instructions to “follow up with your primary care doctor” and “take your medications as prescribed.” Each time, she couldn’t afford the copays for her medications, couldn’t get time off work for appointments, and ended up back in the emergency room.

This is the cruel irony of American healthcare: We’ll spend $30,000 on a hospital stay but balk at spending $300 on the medications that would have prevented it. And now, under the 2025 Medicaid reforms in the One Big Beautiful Bill Act, Maria might lose her coverage entirely — not because she doesn’t work hard enough, but because navigating monthly reporting requirements while working multiple jobs is nearly impossible.

The legislation imposes 80-hour monthly work requirements, enhanced eligibility verification every six months, and reduced federal funding for Medicaid expansion. In a recent analysis my colleagues and I published in JAMA Health Forum, we projected that for every 100,000 people who lose coverage, there will be 13 to 14 excess deaths and over 800 preventable hospitalizations annually. By 2034, we estimate 101 rural hospitals will be at high risk of closure each year.

For every 100,000 people who lose Medicaid coverage, there will be 13 to 14 excess deaths and over 800 preventable hospitalizations each year.

The evidence from states that tried work requirements tells a consistent story. When Arkansas implemented them in 2018, 18,000 people lost coverage within months — but research published in the New England Journal of Medicine found no increase in employment. More than 95% of those affected were already working, disabled, or should have been exempt. The problem wasn’t the work requirement itself but the reporting mechanism: manual monthly verification that working people couldn’t navigate. 

These aren’t policy failures — they’re working exactly as intended. The administrative burden is the point, not a bug.

Georgia’s current program enrolled only 6,500 people by late 2024 — far short of the 100,000 projected — while spending $13,360 per enrollee, with three-quarters of the money going to administration rather than healthcare. The Congressional Budget Office estimates the federal requirements will cause 1.5 million people to lose eligibility annually.

These aren’t policy failures — they’re working exactly as intended. The administrative burden is the point, not a bug, using terms like “they’ll get over it” and “well, we’re all going to die.” 

But here’s what troubles me most as both a clinician and researcher: We’re having the wrong debate. The current conversation is about how to spend less federal money on Medicaid. The right conversation is about how to build a healthcare system that actually keeps people like Maria healthy.

What the Research Actually Shows

For the past decade, healthcare innovators pursued what seemed like an obvious idea: identify the sickest, most expensive patients — the “super-utilizers” with multiple hospitalizations — and provide intensive care coordination to keep them out of the hospital. It made intuitive sense. It also didn’t work.

The Camden Coalition, pioneers of this “hotspotting” approach, conducted a rigorous randomized trial of 800 Medicaid patients with two or more recent hospitalizations. These patients received comprehensive post-discharge support from nurses, social workers, and community health workers. The result? Zero difference in readmissions: 62.3% in the intervention group versus 61.7% in the control group (difference: 0.82 percentage points; 95% CI: −5.97 to 7.61). Both groups improved dramatically — but that’s because very sick patients naturally get better over time regardless of intervention, a statistical phenomenon called regression to the mean.

This null result doesn’t necessarily invalidate all high-risk patient interventions, but it does tell us something crucial: waiting until patients are in crisis and then trying to coordinate their way out doesn’t work. What does work is catching people before they spiral into crisis.

Contra Costa County tested exactly this approach, identifying “rising-risk” patients — people whose health is deteriorating but who haven’t yet become hospitalization frequent-flyers. These patients receive support from community health workers who help with medication adherence, transportation, and appointment scheduling. This upstream intervention significantly reduced avoidable hospitalizations.

Waiting until patients are in crisis and then trying to coordinate their way out doesn’t work.

The difference in approaches matters enormously. One waits until patients are in crisis and then tries to coordinate their way out. The other prevents the crisis from happening.

Three Interventions With Strong Trial Evidence

While researchers were learning what doesn’t work, they were also building an evidence base for what does. Three interventions stand out for having rigorous proof of effectiveness from multiple randomized trials. Importantly, all three can be implemented by health systems and community organizations — though, as I’ll discuss, implementation faces real barriers.

Community health workers have the most consistent trial evidence. These are frontline workers from the communities they serve who help patients navigate the healthcare system, coordinate care, and address social barriers to health. The IMPaCT model developed at Penn Medicine has been tested in multiple randomized trials showing dramatic results. In one study, 30-day readmissions dropped from 40% to 15.2% among patients with multiple prior hospitalizations, delivering a return on investment of $1.80 per dollar spent for Medicaid payers. In diabetes programs, meta-analyses show HbA1c improvements of 0.21 to 0.39 standard deviations — modest but meaningful reductions in blood sugar that prevent complications over time. These findings have been replicated across VA, FQHC, and academic practice settings, though we don’t yet know whether similar effects occur in rural areas or smaller health systems.

Pharmacist-led chronic disease management shows similarly strong results. In a randomized trial at a federally qualified health center serving Medicaid patients, pharmacist-led diabetes care achieved an additional 0.91 percentage point reduction in HbA1c beyond usual care. Another study of primarily Medicaid patients showed simultaneous improvements across multiple conditions: HbA1c dropped nearly a full percentage point, blood pressure fell 9.1 mmHg, and LDL cholesterol decreased 40.4 mg/dL. These are clinically meaningful effects — roughly equivalent to adding a diabetes medication — achieved by having a pharmacist spend time with patients adjusting medications, explaining side effects, and addressing insurance barriers that physicians rarely have time to handle.

Collaborative care for depression has accumulated the deepest evidence base — more than 70 randomized trials, though most studied Medicare or commercially insured populations rather than Medicaid specifically. The model pairs primary care physicians with care managers (often social workers) and consulting psychiatrists. In the landmark trial, 45% of patients in collaborative care achieved significant symptom reduction compared to just 19% in usual care. Even more compelling: The model improves physical health alongside mental health. Patients with both depression and diabetes who received collaborative care saw their HbA1c drop 0.58% alongside their mood improvements.

What makes collaborative care particularly powerful is that it solves a fundamental problem Maria faces: primary care doctors drowning in administrative work while patients with treatable depression slip through the cracks. The care manager handles the follow-up, the psychiatrist provides expertise without requiring a specialist appointment, and the primary care doctor stays in charge but isn’t doing it alone.

Why These Solutions Remain Rare

If the evidence is so strong, why aren’t these interventions everywhere? The answer reveals a hard truth about American healthcare: Evidence alone doesn’t overcome structural barriers.

Fourteen states currently reimburse some community health worker services through Medicaid, but coverage is often limited to specific activities like diabetes self-management education rather than comprehensive care coordination. The IMPaCT model requires sustained funding that managed care organizations operating under tight capitation rates are unlikely to provide voluntarily. Pharmacist-led disease management faces similar challenges: Collaborative practice authority varies by state, with some requiring individual physician agreements that create bottlenecks. Collaborative care billing codes exist in 26 states, but many require prior authorization or limit eligible providers.

In places most influenced by the market, those with the greatest health needs receive the least resources and attention.

Health systems and managed care organizations face a crucial tension: They can implement these interventions using value-based contract funding, but MCOs operating under 2-3% profit margins have limited incentive to invest beyond what states explicitly mandate in their contracts. This is Medicaid’s version of what primary care physician and researcher Julian Tudor Hart called the “inverse care law”: in places most influenced by the market, those with the greatest health needs receive the least resources and attention.

Scaling these interventions requires state Medicaid directors to mandate them in MCO contracts with adequate reimbursement rates, not just encourage voluntary adoption. It requires addressing workforce pipelines — many states lack CHW certification programs or training infrastructure. And it requires recognizing that implementation capacity varies enormously across the 56 different Medicaid programs (50 states plus DC and territories). What works in urban Oregon or Pennsylvania may face different barriers in rural Mississippi or Texas.

Learning From Oregon’s Success — and Its Limitations

Oregon’s Coordinated Care Organizations provide one model for how Medicaid delivery reform can work. They integrated physical, behavioral, and dental care under global budgets with accountability for quality and access. Research found a 13.4% reduction in low birthweight babies (though notably, no benefits were observed among Hispanic women or in rural populations — precisely the groups that arguably needed CCOs most). Inpatient spending decreased 14.8% while primary care spending increased 19.2%, suggesting improved access rather than just cost-cutting.

Oregon’s success comes with important caveats for national replication. The state is racially less diverse than most, has relatively high Medicaid provider participation rates, and benefits from a political culture that supports integration and global budgets. Several health plans have exited Oregon’s market citing inadequate rates, raising sustainability questions. Oregon shows what’s possible — but achieving similar results elsewhere will require adapting the model to different contexts.

The Mixed Evidence on Social Determinants

Beyond these clinical interventions, housing assistance, medically tailored meals, and produce prescriptions receive enormous attention. But the research reveals a gap between the enthusiasm and the evidence.

Housing First programs show clear benefits for reducing emergency department visits, but the Denver randomized trial found no significant effect on mortality over two years. Produce prescription programs show promising health improvements in observational studies — but these studies lack randomized control groups, making it impossible to know whether benefits come from the intervention or simply from enrolling patients already motivated to change. The largest trial of medically tailored meals found no effect on the primary outcome of 90-day readmission.

Produce prescription programs show promising health improvements.

This evidence doesn’t mean these interventions are useless. The research suggests they work best when directly providing services (housing, food) rather than just coordinating referrals. But the evidence for improving health outcomes remains weaker than for community health workers, pharmacist-led care, and collaborative treatment of depression.

The Primary Care Crisis Demands Real Solutions

As of mid-2024, nearly 75 million Americans lived in areas with inadequate primary care access. A recent Commonwealth Fund survey found that 43% of U.S. primary care physicians report burnout, with one-third planning to stop seeing patients within three years. AI scribes that transcribe encounters help on the margins, but they don’t address the fundamental problem: We’re drowning primary care in administrative work while Maria waits weeks for an appointment.

Here’s what good primary care actually is: suspecting cancer and ordering the workup before referring to oncology. Managing the diabetes that affects one in seven American adults through lifestyle counseling and medication adjustments. Diagnosing the ambiguous chest pain or persistent cough before it becomes an emergency. This is clinical judgment work that can’t be automated — but it requires time that our current system doesn’t provide.

Federally qualified health centers — community health centers that see patients regardless of ability to pay — offer one model for accessible primary care. Counties that lost health center sites saw higher mortality rates in observational research, though it’s difficult to separate the effect of FQHC closures from the broader economic decline that leads to such closures. What we know for certain: half of these centers operate with fewer than 90 days of cash on hand. We have infrastructure that works, but we’re starving it of resources.

This is clinical judgment work that can’t be automated — but it requires time that our current system doesn’t provide.

Meanwhile, new AI-powered tools promise shortcuts — patient-facing platforms to appeal insurance denials, claiming success rates of 70-80% compared to industry averages around 50%. But these figures come from company reports, not independent trials. We don’t know if these tools work equally well for Medicaid populations or whether they improve health outcomes rather than just administrative ones. Given healthcare’s troubled history with algorithms that systematically disadvantage vulnerable populations, we need rigorous evaluation before deploying AI tools widely.

We have infrastructure that works, but we’re starving it of resources.

The interventions with proven track records remain decidedly low-tech: human community health workers, pharmacists with time to counsel patients, and collaborative care teams.

What the Evidence Demands

The current Medicaid debate asks the wrong question: How do we spend less federal money? Cutting Maria’s coverage doesn’t make her healthier. It just shifts costs from her Medicaid plan to the emergency room, then to the hospital’s uncompensated care budget, then to higher premiums for everyone else.

The right question is: How do we build a system that keeps people healthy? The evidence provides clear answers.

State Medicaid programs must mandate community health worker services in MCO contracts with adequate reimbursement, not just encourage voluntary adoption. Pharmacists must be empowered to manage chronic diseases through collaborative practice agreements that don’t require individual physician approval for each patient. Collaborative care for depression must be implemented using billing codes that already exist in 26 states. Federally qualified health centers need sustainable funding rather than chronic under-resourcing. And we must learn from Oregon’s success integrating care under accountable arrangements — while recognizing that what works in Oregon will require adaptation in Texas, Mississippi, or rural areas nationwide.

These aren’t hypotheticals or pilot programs. These are evidence-based interventions with proven effect sizes from rigorous trials. But evidence alone won’t scale them — that requires deliberate policy choices to change MCO contracts, workforce development, and payment structures.

We know how to keep people healthy. We’re choosing not to. Every time we impose administrative burdens that strip coverage from working people, we choose crisis over prevention. Every time we underfund primary care and community health centers, we choose expensive emergencies over affordable maintenance.

We know how to keep people healthy. We’re choosing not to

In a system built on evidence, Maria would have a community health worker to help navigate medication copays and schedule appointments that don’t cost her a day’s wages. She’d have a pharmacist with time to adjust her medications and explain why they matter. She’d have access to collaborative care for depression that makes managing diabetes harder. She wouldn’t face monthly reporting requirements designed to make her give up.

Maria shouldn’t have to choose between her medications and her rent. The evidence shows we can do better. The question is whether we will.

Disclosure: The author is co-founder and chief medical officer of Waymark, a public benefit company that provides care management services under contract with Medicaid managed care organizations. The views expressed here are the author’s own and do not necessarily reflect those of Waymark Some of the studies cited in this article were co-authored by Dr. Basu.