Blog

Policy at the Capitol: Key Takeaways from the 3rd Annual Washington Health Economics Symposium

November 4, 2026

The 3rd Annual Washington Health Economics Symposium (WHES), held October 4–6, 2026 in Washington, DC, brought leading health economists and policy stakeholders together at the U.S. Capitol Building and The Morning Consult. Framed around the pairing of rigorous research and real-world policy, the two-day program featured seven research papers spanning payer policy, technology and innovation policy, and the use of AI in delivering primary care.

Together, the papers make a common point: how health care organizations are structured, paid, and regulated shapes what actually happens to patients and prices, often in ways the original policy design did not anticipate. Whether the topic was insurers meeting medical loss ratio targets through internal transfers, arbitrators with a financial stake in the disputes they decide, or AI software changing who gets a life-saving stroke procedure, the research shows that incentives and organizational form matter as much as the rules themselves.

The symposium opened with a welcome reception built around the question of whether “Medicare Advantage for All” is a 2030 possibility or fool’s gold. Listen to the full discussion On Background here.

On Monday, scholars presented their work in three sessions at the United States Capitol, and on Tuesday, each session was paired with a panel focused on advancing the research findings into public policy. Current and former policy experts who joined the paper authors to discuss their work included:

Douglas Holtz-Eakin, American Action Forum
Nathan Wilson, Econic Partners; former FTC Bureau of Economics
Steph Carlton, Deputy Administrator, CMS
Loren Adler and Matthew Fiedler, Brookings Institution
Randy Pate, former CCIIO Director

Scholars also participated in a conversation exploring how to bring the 340B Drug Pricing Program back to its original safety-net purpose without destabilizing the hospitals that rely on it. Proposals included better tracking of where 340B dollars go, pay-for-results benchmarks, and a transition toward Medicaid-based support, but participants acknowledged there is no easy fix since any replacement funding would fall on the federal budget. Listen to the 340B discussion On Background here.

Below are summaries of the seven research papers presented at WHES 2026. For more information about The HEAL Network’s annual symposiums and opportunities to participate, please visit our website here or contact info@thehealnetwork.org.

Paper 1: What Drives Premium Costs in Employer-Sponsored Health Insurance?

Salpy Kanimian, Brown University
Nate Dominitz, Rice University
Marah Short, Rice University
Vivian Ho, Rice University

Research Question

Why do premiums for employer-sponsored insurance differ so widely across employers in different local markets?

Main Findings

Data cover about 1,100 colleges and universities in 2025–2026.
Institutions offer three plans on average, ranging from one to twelve.
PPO/POS premiums, offered at 93% of institutions, range from $1,310 to $25,671, with a mean of $11,730.
The gap between the cheapest and most expensive plan on a single campus averages about $3,200.
The factors examined are purchasing-unit size, hospital concentration, insurer market power, having a medical school, actuarial value, and plan offerings.
Preliminary results show plan, market, and purchasing characteristics matter, while conventional institutional traits explain little.

Key Takeaways

Wide gaps within one industry: Among institutions in the same broad industry, PPO/POS premiums run from $1,310 to $25,671, a spread of nearly twentyfold. That points to local markets and purchasing arrangements rather than industry alone.
Plan design is a lever: Even on one campus, the cheapest and most expensive plans differ by about $3,200 a year on average. Decisions about how many plans to offer and how generous they are can move costs.
Market structure is in focus: By testing hospital concentration, insurer market power, and bargaining-unit size, the paper speaks to the questions states are asking as they experiment with ways to contain health care spending.
Who you are explains little: Conventional institutional characteristics explain relatively little of the variation. That suggests benchmarking against similar peers may miss what drives an employer’s premiums. The authors call these results preliminary.

Bottom line: Two employers in the same industry can pay very different prices for health coverage. Early evidence suggests the explanation lies less in who the employer is and more in the plans it offers, the hospital and insurer markets it buys in, and how it purchases. That is where employers and policymakers may find levers to slow premium growth.

Paper 2: The Private Option: Trade Shocks and Hospital Reliance on Private Healthcare Insurance

Cyrus Aghamolla, Rice University
Ly Nguyen, Rice University
David Zhang, Rice University

Research Question

How much do hospitals depend on private insurance reimbursement, and what happens to hospital finances, operations, and patient outcomes when a local shock erodes the share of residents with employer-sponsored coverage?

Main Findings

The authors use China’s 2001 entry into the World Trade Organization as a shock to manufacturing employment, and therefore to employer-sponsored insurance, in exposed communities.
The stakes are high: private insurers account for 37% of U.S. hospital-care spending despite covering only 24% of discharges, with payment rates averaging roughly 240% of Medicare prices.
Hospitals in exposed regions saw a sharp decline in privately insured admissions, while publicly insured admissions showed no comparable drop.
Because privately insured patients typically bring the highest reimbursement, the loss had a substantial negative effect on hospital finances.
Hospitals in manufacturing-dependent areas were more likely to close or merge with another hospital and showed higher mortality, particularly among riskier patients.
Effects were stronger at hospitals that relied more heavily on private reimbursement and in areas with greater outmigration and firm closure after trade liberalization.

Key Takeaways

Hidden costs of trade shocks: Job losses in manufacturing spill over into community access to health care through hospitals’ finances.
Cross-subsidization is fragile: The hospital system depends heavily on private payers to cover the cost of care for everyone else, so shocks to private coverage can threaten the viability of local hospitals.
Policy relevance: The authors point to the recent expiration of the enhanced ACA premium tax credits, which they cite as having left an estimated three million previously privately insured people without coverage, as a reason to understand how dependent hospitals are on private insurance.

Bottom Line: Hospitals are far more exposed to changes in private insurance coverage than their patient mix suggests. When employer-sponsored coverage disappears from a community, hospitals lose their most profitable patients, and the consequences can include closures, mergers, and higher mortality.

Paper 3: New Medical Technology and Clinical Practice: Evidence from AI Stroke Triage

Miaoqing Jia, Weill Cornell Medicine
Lucy Xiaolu Wang, University of Massachusetts Amherst

Research Question

Does adoption of AI stroke-triage software, which flags suspected large-vessel occlusions on imaging and alerts the stroke team, change how often hospitals use mechanical thrombectomy?

Main Findings

Using national multi-payer claims covering 420,294 ischemic stroke admissions at 2,828 facilities from 2017 through 2023, the authors apply a staggered-adoption difference-in-differences design.
Adoption raised the thrombectomy rate by 1.08 percentage points, a 23% increase relative to the 4.73% pre-period baseline. The effect appeared in the month of adoption and persisted through at least two years.
Results held across several modern causal estimators, alternative adoption definitions, and different comparison groups.
The increase was concentrated at facilities with existing but not high baseline thrombectomy use (1.85 and 1.67 percentage points in the lower two tertiles), with no detectable change at facilities that performed no thrombectomies or at those with the highest baseline use.
The effect was largest for managed Medicaid admissions (1.73 percentage points) and smallest for commercially insured admissions.
The increase was not driven by more scanning, and the authors found no change in thrombolysis, transfers, or treatment of hemorrhagic stroke, where thrombectomy is not indicated.
The authors estimate roughly 384 additional thrombectomies per year among the admissions they observe, and about 75 additional patients per year surviving with functional independence, though they describe the disability arithmetic as illustrative.

Key Takeaways

Recognition is the bottleneck: The pattern points to faster identification of already-eligible patients within established stroke programs, not more imaging or substitution between treatments.
Equity opportunity, with a catch: The largest gains were among managed Medicaid patients, but adopting facilities were larger and more experienced, and non-adopters served more Medicaid patients. Population-level benefits depend on AI triage reaching Medicaid-serving hospitals.
Payment matters for diffusion: The code used to identify adoption was tied to a temporary Medicare add-on payment (NTAP) that expired without a permanent replacement, leaving open how reimbursement will shape future adoption.
Limits: The data exclude traditional fee-for-service Medicare, and adoption is inferred from billing, so time-varying confounding cannot be fully excluded.

Bottom Line: Adding an AI notification layer to existing stroke imaging was followed by a meaningful, durable increase in thrombectomy use without a change in scanning practice. Whether those gains reach the patients who need them most will depend on how the technology is paid for and who adopts it.

Paper 4: Projected Population Health Impact of Autonomous AI Primary Care in US Physician-Shortage Counties

Alon Bergman, University of Pennsylvania
Dan Zeltzer, UC Berkeley School of Public Health

Research Question

If an autonomous AI primary care physician were deployed nationally, how many deaths and preventable hospitalizations could it avert in counties with too few primary care physicians?

Main Findings

The authors build a county-level decision-analytic model combining 2025 physician supply data, CDC mortality data, and causal estimates from a systematic review of quasi-experimental studies on primary care supply and outcomes.
A total of 2,303 counties (74.1% of the sample, 141.9 million residents) fall below a 1:1,500 physician-to-population adequacy threshold.
Under a moderate adoption scenario calibrated to observed telehealth use (31% of eligible patients), deployment is projected to avert about 10,700 deaths and 5,400 preventable hospitalizations per year. Across adoption scenarios, projected deaths averted range from roughly 6,600 to 16,200.
The authors describe the moderate-scenario mortality reduction as roughly twice that associated with ACA Medicaid expansion among low-income older adults.
Benefits are concentrated where shortages are deepest: the South accounts for 54% of the projected benefit, and nonmetropolitan counties account for about 35% of benefit despite housing 25% of the treated population.
Estimates remain policy-relevant under more skeptical assumptions. Even at conservative adoption and one-quarter physician-equivalence, the model projects about 1,600 deaths averted annually.
A bottom-up cost estimate of $4–7 billion per year implies roughly $470,000 per death averted.

Key Takeaways

A supply-side policy tool: The magnitudes are large enough that autonomous AI primary care merits evaluation as a national workforce policy instrument.
Access is not automatic: Cost-sharing, broadband, and digital trust barriers could mute benefits in the uninsured, rural, and low-connectivity communities where the shortage, and therefore the projected gain, is concentrated.
Key assumptions: The model assumes the AI performs like an additional physician, and no randomized trials of autonomous AI primary care with long-term outcomes yet exist. The authors treat their estimates as upper-bound projections under local linearity.
Regulatory timing: The authors note that existing FDA frameworks were not designed for adaptive, general-purpose clinical AI, and that the projections quantify the stakes for the licensure and payment frameworks now being developed.

Bottom Line: Even under conservative assumptions, autonomous AI primary care could avert thousands of deaths a year in physician-shortage counties, at an estimated cost of roughly $470,000 per death averted. Realizing that potential will require prospective validation, thoughtful payment and regulatory design, and deliberate attention to digital access.

Paper 5: Arbitration with Regulatory Capture: Evidence from Disputes over Surprise Medical Bills

Benjamin Chartock, Bentley University
Christopher Whaley, Brown University

Research Question

Why do insurers lose so often in No Surprises Act (NSA) arbitration, and do the incentives of the arbitrators themselves help explain it?

Main Findings

The NSA created what is likely the largest final-offer arbitration system in U.S. history, with more than 3 million disputes in its first three years against government projections of about 20,000 per year.
Using public CMS data on 500,000+ cases, the authors find insurers lose 84% of cases, and winning provider offers are about three times the qualified payment amount (QPA) benchmark. For a common emergency visit (CPT 99284), the mean arbitration award is $736 versus $231 for mean in-network rates.
A standard model of arbitration fits the data poorly. The authors rule out differences in risk aversion as an explanation, since matching observed win rates would require implausibly large risk-aversion parameters given the stakes involved.
They cannot rule out regulatory capture: arbitration entities are paid per case, which creates an incentive to rule in ways that induce more cases. Estimated inducement varies across arbitration entities, and many of those with the strongest inducement incentives have financial ties to private equity firms that also back provider groups bringing NSA cases.
Providers more exposed to NSA-related services bring more cases.
Back-of-the-envelope calculations suggest providers could capture an additional $2.7 billion for a single emergency medicine service.

Key Takeaways

Incentives of the referee matter: When arbitrators earn fees per dispute, the system can drift toward outcomes that generate more disputes.
Prices may spill over: A high arbitration outside option can raise negotiated in-network rates and, ultimately, premiums.
Policy options: Use Transparency-in-Coverage data as a benchmark for arbitrators, hold scoring auctions that certify arbitrators for limited terms, and restrict shared corporate ownership between arbitrators and the parties before them.
A caution: The authors emphasize their findings are consistent with, but do not definitively prove, regulatory capture.

Bottom Line: The No Surprises Act protects patients from surprise bills, but its arbitration process has produced lopsided outcomes that favor providers. Fixing arbitrator incentives may be essential to keeping the law’s consumer protections from translating into higher health care prices.

Paper 6: Organizational Form and Regulatory Compliance: Evidence from Medical Loss Ratio Regulation

Lynn M. Hua, Georgia State University

Research Question

Can firms comply with a binding regulation without changing the behavior it was meant to influence, and does organizational form, specifically vertical integration, shape how insurers comply with the ACA’s medical loss ratio (MLR) rule?

Main Findings

The MLR provision requires individual market insurers to spend at least 80% of premium revenue on medical claims and quality improvement, or else pay rebates to enrollees.
Using NAIC regulatory filings from 2001 to 2024 and a triple-differences design that exploits insurers’ pre-regulation distance below the 80% threshold, the paper compares traditional insurers with vertically integrated insurers that own or are affiliated with the providers they pay.
Traditional insurers close about 71% of their pre-regulation compliance gap. Vertically integrated insurers respond about 90% less, and their response is largely independent of how far below the threshold they began.
Several patterns are consistent with transfer pricing rather than efficiency gains or better care coordination. Integrated insurers with larger future gaps reported systematically lower claims before the ACA, the differential response is concentrated in the claims channel rather than in quality improvement or administrative expense, and the gap between integrated and traditional insurers persists and widens over time.

Key Takeaways

Compliance is not the same as behavior change: For traditional insurers, raising the MLR means paying providers more or increasing utilization. For integrated insurers, payments to affiliated providers are internal transactions that can raise reported claims without a corresponding change in care delivered.
Organizational form shapes regulatory effectiveness: The same rule can have different effects depending on how a firm is structured.
Welfare implications: The results add a regulatory arbitrage dimension to the debate over vertical integration in health care.

Bottom Line: Vertically integrated insurers appear able to meet the MLR threshold through internal accounting rather than the cost-and-care adjustments faced by traditional insurers. Regulators designing spending-based rules should account for how ownership structure creates alternative paths to compliance.

Paper 7: Health Insurance Design and the Passthrough of List Prices to Out-of-Pocket Costs

Josh Feng, University of Utah
Luca Maini, Harvard Medical School

Research Question

Why do health plans tie patient cost-sharing to drug list prices that have grown increasingly detached from actual acquisition costs after rebates?

Main Findings

The authors develop a model of health insurance design with adverse selection, showing that plans facing minimum coverage regulation can raise patient cost-sharing by covering drugs with higher list prices, which lowers premiums while circumventing coverage requirements.
Using an instrumental variable approach, they find that higher list prices pass through to higher out-of-pocket costs only in plans whose benefit design is constrained.
Passthrough is near one-to-one in Medicare Part D plans, which must meet minimum benefit-design standards, and in commercial high-deductible health plans, which are constrained by the 100% cost-sharing cap below the deductible.
Passthrough is lowest in more generous commercial PPO and HMO plans that can set cost-sharing freely.
Compared with commercial plans, Medicare Part D plans are also much more likely to cover high-list-price biologics than highly similar, low-list-price biosimilars.

Key Takeaways

Regulation shapes formularies: Benefit-design constraints can push plans toward high-list-price drugs because higher list prices let them shift cost to patients.
Patients bear the consequences: Prior research cited by the authors finds that higher cost-sharing can lead patients to discontinue high-value drugs and may even raise mortality.
Reform implications: Discussions of rebates, list prices, and biosimilar uptake need to account for the benefit-design rules that make list prices matter for patient costs.

Bottom Line: High list prices are not just a manufacturer pricing problem. In regulated benefit designs, they can be a deliberate, rational response by plans, and patients with deductibles or coinsurance end up paying for it.