Pablo Zavala · AI Safety Evaluation · Research Engineering

The Policymaker's Bias: When Institutions Misjudge GLP-1 Obesity Drugs

GLP-1 drugs work while public coverage stalls, and the gap reads best as a behavioral-economics failure inside institutions: fiscal-score anchoring, an inherited status-quo default, and loss-averse delay that a multi-attribute objective and two framing interventions can correct.

March 24, 2026 · 7 min

GLP-1 receptor agonists moved obesity treatment from speculative promise to measured clinical effect faster than public policy revised its governing assumptions. Semaglutide produces large mean weight reductions, tirzepatide pushes those reductions further, and semaglutide also lowers major cardiovascular events in adults who carry excess weight alongside established heart disease. Yet Medicare still limits anti-obesity coverage to indications beyond obesity itself, Medicaid coverage stays fragmented, and prescribing trails eligibility by a wide margin. My working manuscript argues that this adoption gap rewards a behavioral reading, because the same cognitive failures that distort a patient's choices also distort the institutions that decide coverage.

The stakes reach far past a pharmacy line item, since the Lancet Commission on the global syndemic described excess body weight as a condition touching billions of people and imposing macroeconomic costs measured in shares of gross domestic product (Swinburn et al., 2019). Public institutions still treat obesity pharmacotherapy as a narrow line item inside a policy world built for episodic treatment and short budget windows, so the manuscript asks whether the state evaluates these drugs with the right objective function at all.

Evidence Moved While Coverage Lagged

The clinical record I cite arrives with sharp numbers. In the STEP 1 trial, adults with overweight or obesity taking semaglutide 2.4 mg lost a mean 14.9% of body weight over 68 weeks (Wilding et al., 2021, NEJM). In SURMOUNT-1, tirzepatide reached mean reductions as high as 22.5% (Jastreboff et al., 2022, NEJM). The stronger policy fact runs cardiovascular: in SELECT, semaglutide cut the composite risk of cardiovascular death, nonfatal myocardial infarction, and nonfatal stroke by 20% among adults with overweight or obesity and established cardiovascular disease who lived free of diabetes (Lincoff et al., 2023, NEJM). One United States cost-effectiveness model placed semaglutide near $23,000 to $27,000 per quality-adjusted life year under its stated assumptions (N. Kim et al., 2022).

Policy moved slower. Under current law, Medicare Part D generally covers anti-obesity drugs only when a patient uses them for an accepted indication beyond obesity, such as diabetes or cardiovascular risk reduction (Cubanski & Williams, 2024, KFF). Medicaid coverage stays fragmented, and KFF counted just 13 state fee-for-service Medicaid programs covering GLP-1s for obesity treatment as of January 2026 (Williams, 2026, KFF). Prescribing lags eligibility sharply, since a national electronic-health-record study found that fewer than 3% of eligible patients had received semaglutide or tirzepatide for obesity, with Hispanic and non-Hispanic Asian patients receiving those therapies less often than White patients (C. Kim et al., 2025, JAMA). These access gaps sit atop an unequal disease burden, because non-Hispanic Black adults showed the highest obesity prevalence in the 2017-2018 CDC estimates (Hales et al., 2020).

Institutions Inherit the Biases of Minds

My central move reframes the whole gap. A shortage of evidence offers a weak explanation, because the trials already exist and the effect sizes already impress. A behavioral-economics account fits the facts better: policymakers anchor on a narrow fiscal score, inherit a status-quo default built for a pre-GLP-1 era, and meet uncertainty with loss-averse caution. The patient's brain and the policy system rhyme, since both bend to salience, inertia, and distorted responses to risk. Once that parallel becomes explicit, the guiding question shifts from whether the drugs work toward whether the state weighs them against the objective it claims to care about.

One Fiscal Number Anchors the Debate

The Congressional Budget Office estimates that authorizing Medicare coverage of anti-obesity medications would raise federal spending across the 2026-2034 window by about $35 billion (Congressional Budget Office, 2024). That figure carries political weight because it arrives first and loud, and in the language of Tversky and Kahneman (1974) it works as an anchor. My deeper claim runs structural rather than rhetorical. A proper evaluation spans many attributes at once, including short-run fiscal cost, mortality reduction, equity, persistence, and downstream savings, so the CBO score amounts to a projection of that fuller problem onto a single dimension. Within that one dimension the estimate can hold up; measured against the full welfare problem, the same estimate stays incomplete. A government that scores only ten-year expenditure answers a narrower question than the one it claims to weigh.

A Multi-Attribute Objective Can Flip the Ranking

I formalize the point, and the formalization sharpens the critique from complaint into result. Write policy value as a weighted sum across attributes, V(a) = Σ wᵢ·vᵢ(xᵢ), then split any pairwise comparison into a fiscal component and the omitted attributes, ΔV(a,b) = ΔF(a,b) + ΔU(a,b). Whenever the fiscal difference favors exclusion yet the omitted health, equity, and durability gains outweigh it, the projected fiscal rule ranks one policy first while the full welfare rule ranks the other first. Therefore a fiscally disciplined score can still encode a welfare-misspecified choice rule. That result gives the anchoring critique its rigor, because the number stays arithmetically correct and normatively incomplete at once, while obesity keeps imposing large direct medical costs that a ten-year expenditure line leaves out (Cawley et al., 2021).

A parallel confusion compounds the error, because modern debate collapses cost-effectiveness and affordability into one question. Cost-effectiveness asks whether a treatment returns enough value per unit of health gain, and a model can still show semaglutide as cost-effective under its assumptions (N. Kim et al., 2022). Affordability asks whether a payer can absorb the budget shock on the relevant timeline. Both statements hold together, and treating them as one more collapse marks yet a further form of misspecification.

Inherited Defaults and Loss-Averse Delay

Two further distortions follow the anchor. Medicare's exclusion reads like a neutral baseline, yet it descends from an earlier regulatory era as an inherited default. Behavioral evidence on defaults runs clear, since automatic enrollment sharply raises take-up while leaving the underlying instrument intact (Madrian & Shea, 2001). Inertia supplies the mechanism, because states must actively choose to cover GLP-1s for obesity and Congress must actively reverse a statutory exclusion. Coverage refusal therefore deserves treatment as an active policy setting rather than a natural resting point.

Loss aversion supplies the third distortion. The evidence base grew rich yet stayed psychologically hard, because the drugs work better than expected, the eligible population runs large, discontinuation erodes gains, and budget exposure stays vivid. Federal experiments such as the BALANCE model show CMS already testing negotiated pricing and pairing medication access with lifestyle support, with a Medicaid launch as early as May 2026 and a $50 monthly payment for eligible Medicare beneficiaries in the bridge demonstration (Centers for Medicare & Medicaid Services, 2026). That richer evidence redistributes uncertainty rather than dissolving it. Under prospect theory, losses loom larger than gains (Kahneman & Tversky, 1979), so a choice with visible fiscal downside and delayed social upside makes caution behaviorally predictable. I capture the pattern with an ambiguity-averse criterion, W(a;P) = min over p in P of Eₚ[V(a,ω)], under which fresh evidence can slow adoption whenever it widens the set of plausible downside states faster than it sharpens the center of the distribution. Therefore more evidence produces more willingness to cover only under conditions the institution rarely enjoys.

A fourth distortion, moralization, keeps exclusion feeling defensible. Obesity discourse still leans on ideas of weak will even as reviews of obesity neuroscience emphasize disrupted reward and control circuitry rather than simple preference failure (Volkow, Wang, & Baler, 2011), while experimental work shows GLP-1 signaling modulates appetite- and reward-related brain areas in humans (van Bloemendaal et al., 2014), and international consensus treats weight stigma as a genuine barrier to care (Rubino et al., 2020). When a payer moralizes a condition, the payer overestimates how much effort a patient should supply unaided, which makes coverage refusal appear more reasonable than the evidence warrants.

Two Interventions, One Behavioral Logic

I propose a twofold intervention that acts on choice architecture rather than on belief. At the institutional level, default coverage with active opt-out would flip which dimension registers as the salient loss for FDA-approved medications that clear a cost-effectiveness screen and pair with behavioral support. Under exclusion, fiscal cost enters first as the immediate loss; under default coverage, foregone health enters as the sharper loss, while the physical policy stays constant and only the reference point moves. The prospect-theory value function, u(x) = xᵃ for gains and −λ(−x)ᵃ for losses with λ > 1, supplies the formal handle, so the reform succeeds by changing which attribute the system codes as the default loss.

The second intervention faces the patient and targets present bias. Gastrointestinal side effects occur commonly during GLP-1 treatment, and stopping the drug brings substantial weight regain along with reversal of cardiometabolic improvement (Wilding et al., 2021, 2022). Standard adherence messaging frames continuation as a gain, while my alternative frames discontinuation as a loss, because stopping puts already-earned cardiovascular and metabolic gains at risk. Bundling reinforces the logic, since the STEP 3 trial showed semaglutide plus intensive behavioral therapy delivering larger average weight loss than behavioral therapy alone (Wadden et al., 2021, JAMA). I read drug access and behavioral support as complements that act through separate channels, so piecemeal provision understates the value of the pair whenever their combined effect on long-run health stays positive. Each arm invites a clean randomized test: a survey experiment with state Medicaid directors for the institutional frame, and a parallel-arm trial among new GLP-1 initiators in safety-net clinics for the patient frame.

A Mirror Image Inside the State

I close with the reflexive point that gives the paper its title. The state now faces a mirror image of the patient's own cognitive problem. Each actor weighs an intertemporal decision under uncertainty, and each overweights the most immediate number, inherits yesterday's default, and guards against visible losses. A patient discounts long-run cardiometabolic gains against near-term discomfort, while an institution discounts long-run health and equity against a salient ten-year score. The remedy in both settings works the same way, because arranging the choice architecture to keep the right quantities in view lets human priorities survive contact with strong evidence. Health, time, dignity, and fair distribution of opportunity are the quantities worth protecting, and the deciding question concerns architecture rather than the mere existence of evidence.

Boundaries

This essay synthesizes my working manuscript, which offers argument and formalization rather than a new trial or dataset of its own; the paper reports zero original empirical measurement.
Every clinical figure comes from the cited randomized trials and observational work (Wilding et al., 2021; Jastreboff et al., 2022; Lincoff et al., 2023; Wadden et al., 2021; C. Kim et al., 2025), reported here as literature I cite rather than as my own findings.
The cost, coverage, and budget figures come from the cited cost-effectiveness model, KFF briefs, the CBO score, and the CMS model page, each carrying its own date and window.
The two policy recommendations, default coverage with active opt-out and loss-framed adherence communication, remain my reasoned proposals, framed for future randomized evaluation rather than presented as demonstrated results.
The formal expressions (multi-attribute value, fiscal-plus-omitted decomposition, ambiguity-averse criterion, prospect-theory value function, and complementarity condition) serve as analytic scaffolding for the argument rather than as estimated quantities.