Jonathan Zhang

Examining GLP-1s usage using data on 1.4M diabetic or obese veterans finds that PCP adoption drives large metabolic improvements matching trials—but no short-run cuts to emergency department use or medical spending, from Bock, Moshfegh, and @jzhangecon.bsky.social www.nber.org/papers/w34667

Actually no, the studies show that there are no significant cost reductions. So it's not relative to current cost.

Two new preprints on the efx of GLP-1s just dropped.

They use different samples, data, & designs.

But they come to the same conclusion on one point:

GLP-1s do not generate meaningful short- to medium-run reductions in non-GLP-1 medical spending, despite clear health benefits.

Another great GLP-1 NBER WP came out today! Coady Wing, Sih-Ting Cai, Daniel Sacks, and Kosali Simon analyzed commercial data. Using different data sources, different identification strategies, yet the results align quite closely: no cost-savings

www.nber.org/papers/w34678

Bottom line: GLP-1s deliver real health benefits in the real world—but they are unlikely to yield large cost savings in the short run, at least for an older, high-comorbidity population.

As access expands, who gets treated matters as much as whether treatment works.

10/10

What about the big policy question-cost savings?

We don't see statistically significant reductions in emergency department visits or total (non-GLP-1) spending. We can rule out cost savings larger than ~7% over four years with a 30-week regimen.

9/10

But here’s the twist: these metabolic benefits fade over time…

By 2024, treatment effects on weight are half as large, and A1C improvements largely disappear. Evidence suggests selection on treatment: early compliers were sicker, while later compliers were healthier.

8/10

ToT estimates of 30-week regimen in the initial take-up period, lines up closely with estimates from SUSTAIN RCTs. RCT means (blue line) fall in our confidence intervals.

We also replicate: fewer heart attack/stroke among cardiovascular patients, reduced moderate drinking.

7/10

GLP-1 adoption leads to large and clinically meaningful improvements:

⬇️A1C
⬇️Body weight

In other words, in a sample of 1.4M patients, clinical trial effects show up in routine care. Effect sizes next...

6/10

Using an intent-to-treat panel framework, we control for baseline differences across patients, PCPs, and regional trends.

First stage is strong: Patients whose 2018 PCP became always-adopter were 20pp more likely to use GLP-1s by 2024 and accumulated 400 more days supply.

5/10

Our research design mimics how many patients receive GLP-1s: through their PCPs. When Ozempic entered the VA formulary in late 2020, some PCPs adopted quickly, others didn’t. Patients had prior existing PCP relationships.

➡️Quasi-experimental variation in GLP-1 exposure.

4/10

Why VA?

The VA's rich electronic health records include biomarkers (A1C, weight). There is no insurance churn, no attrition. Veterans are one of the largest diabetic populations in the US, and hundreds of thousands have received GLP-1s at low cost.

3/10

What do weight loss and glycemic control look like for the average patient? Do alcohol/substance use outcomes really improve? What about medical spending?

We bring quasi-experimental evidence to these questions by studying 1.4M diabetic/obese patients in US Dept of Veterans Affairs (VA)

2/10

Weighing the Impacts of GLP-1s: Quasi-Experimental Evidence From Provider Adoption Founded in 1920, the NBER is a private, non-profit, non-partisan organization dedicated to conducting economic research and to disseminating research findings among academics, public policy makers, an...

🚨Do GLP-1s “pay for themselves”? We study real-world impacts at scale. An estimated 1 in 8 Americans have tried GLP-1s. But beyond clinical trials, we know surprisingly little about their real-world effects.

Paper: nber.org/papers/w34667 (w/
@jasminmoshfegh.bsky.social
& Sam Bock)

🧵1/10

We examine how bias in a widely used medical device against individuals with darker skin tones propagates into disparities in subsequent medical treatment