Gambling Diagnoses Rose Over 60% in Betting States

The finding

On June 26, 2026, Epic Research published an analysis of gambling disorder diagnoses recorded in electronic health records across the United States. The authors are Kersten Bartelt, RN; Caleb Cox; Joe Deckert, PhD; Brad Fox, MD; and Sam Sahakian. Epic Research is the research arm of Epic Systems, the electronic health record company, and it draws on Cosmos, a de-identified dataset built from records contributed by thousands of hospitals and clinics.

The sample is enormous: more than 197 million U.S. adults aged 18 and older with at least one qualifying healthcare encounter between January 1, 2018 and March 31, 2026.

Two groups of states started at essentially the same place in early 2018 and ended more than twice as far apart eight years later. That's the story, and it's why NBC News and outlets across the country covered it the day it landed.

The age pattern matters too. Adults 30 to 49 carried the highest overall rate throughout, reaching 5.8 per 100,000 by Q1 2026. But the sharpest proportional increase was among adults 18 to 29, whose rate more than doubled. Rates in men were consistently higher than in women across the study period. That tracks with what we've written about why age of first bet predicts so much and about the early large payout that sets the hook.

What Epic actually counted

This part matters more than the headline, so it's worth being precise.

Epic measured the share of "active patients" in each quarter who had an encounter carrying a gambling-related diagnosis code. Two codes qualified: ICD-10-CM F63.0, pathological gambling, which is the code that maps to DSM-5 gambling disorder, and Z72.6, "gambling and betting," which is a lifestyle-problem code rather than a psychiatric diagnosis. Including Z72.6 casts a slightly wider net than a strict count of the clinical disorder. If you want the diagnostic criteria themselves, we've laid them out in the DSM-5 criteria for gambling disorder.

The comparison groups came from residence. Epic classified patients as living in one of the 39 states or the District of Columbia that legalized sports betting during the study window, or in one of the 11 that never did: Alabama, Alaska, California, Georgia, Hawaii, Idaho, Minnesota, Oklahoma, South Carolina, Texas and Utah.

That list is worth sitting with. The "control group" is not a random slice of America. It's California and Texas plus nine mostly smaller states, held together by nothing except the absence of a betting law.

A rise in diagnoses is not automatically a rise in disorder

Here's the distinction that most coverage skipped, and it's the reason this article exists.

A diagnosis is not a case of illness. A diagnosis is a case of illness that a clinician found, recognized, and wrote down in a chart in a format a billing system accepted. Anything that changes the finding, the recognizing, or the writing down will move the count without moving the underlying reality by a single person.

And a lot has changed. States that legalized sports betting also tended to fund problem-gambling services, run awareness campaigns, and expand treatment capacity. More funded treatment means more clinicians positioned to make the diagnosis. More awareness means more people naming the problem out loud in an exam room. Both push recorded diagnoses up for reasons that have nothing to do with how many people are actually sick.

Epic states this plainly in its own limitations. The authors write that F63.0 "may be more recognized and coded in recent years than in 2018 due to DSM-5 diffusion, Mental Health Parity and Addiction Equity Act enforcement, and growing clinician awareness," and that "a real change in prevalence cannot be cleanly distinguished from a coding or recognition change." They also state that the Cosmos rate "represents a floor on diagnosed gambling and is not a population prevalence estimate."

The American Gaming Association made exactly this argument in response to the study, telling NBC News that increased screening and greater awareness of available resources "can often lead to more diagnoses." That's a fair point on its face, and any honest reading of this data has to hold it.

How thin is the top of the funnel? Set Epic's number against the best available prevalence estimate. The National Council on Problem Gambling estimates that 2.5 million U.S. adults, about 1%, meet criteria for a severe gambling problem in a given year, with another 5 to 8 million showing mild or moderate problems. One percent works out to roughly 1,000 people per 100,000. Epic recorded 4.8 per 100,000 patients in a quarter. The timeframes aren't directly comparable, and quarterly counts will always run below annual prevalence, but the gap is orders of magnitude, not percentage points.

That's the honest frame. Epic didn't measure how many Americans have gambling disorder. It measured how many showed up in a chart. Those are different numbers separated by a system that, as our reporting on Arizona's screening gap documented, asks patients about gambling only 7.6% of the time.

So why is the 29% decline the interesting half?

Because awareness is a national phenomenon, and the divergence isn't.

DSM-5 diffusion, parity enforcement, media coverage of gambling addiction, and the general rise in clinician familiarity didn't stop at the Texas or California border. If better recognition were the whole story, diagnoses should have climbed in the non-legalizing states too, just more slowly. Instead they fell by nearly a third.

Epic even notes that prediction markets are accessible nationwide regardless of state betting law, which would push the non-legalizing states up, not down. The decline runs against that too.

This is the strongest thing the study has going for it, and it's why the finding shouldn't be dismissed as an artifact. A pure measurement effect predicts parallel lines. These lines cross.

But it isn't proof, and the design can't make it proof. Here's what's still unresolved:

• The groups aren't comparable in the ways that matter. Eleven states, dominated by two very large ones, differ from the other 39 in politics, insurance markets, health system structure, and behavioral-health funding. Any of those could move diagnosis rates on its own.
• The timing isn't clean. Epic says so directly: legalization was staggered across states "with no single timestamp marking impact onset," and the study "cannot attribute trends to specific legal events." There's no before-and-after event study here, just two trend lines.
• The network composition isn't addressed. Epic's published limitations cover coding drift, COVID-era utilization, telehealth expansion, 42 CFR Part 2 reporting variability, and cross-state moves. They do not address whether the mix of health systems contributing to Cosmos shifted over eight years in ways that differ between the two groups. That's a real open question, and we're flagging it as unanswered rather than answered badly.
• It hasn't been peer reviewed. Epic Research publishes its own analyses. This one is not stated as peer reviewed, and it isn't in a journal.

The financial data points the same way

Health records aren't the only place this shows up. Economists Brett Hollenbeck, Poet Larsen and Davide Proserpio studied consumer credit records covering millions of U.S. adults, exploiting the staggered rollout of legalization across states. Their paper, "The Financial Consequences of Legalized Sports Gambling," is published in Management Science and is paywalled; a free working-paper version is posted by UCLA Anderson.

According to UCLA Anderson Review's summary of the published paper, personal bankruptcy filings rose 25% to 30% in the years following legalization in states allowing online betting, debt sent to collections rose about 8%, and credit card and auto loan delinquencies rose by similar amounts. Average credit scores declined. (The exact credit-score estimate is stated differently across versions of the paper, so we're not putting a single number on it.) The harms concentrated where betting was available on a phone; retail-only betting showed little effect.

That's a different dataset, a different research team, a different outcome measure, and a stronger causal design. It moves in the same direction. Our piece on what the bankruptcy filings show about women bettors covers who's landing in those courtrooms.

Participation keeps climbing, too. The Siena Research Institute and St. Bonaventure University's Jandoli School surveyed 3,084 U.S. adults from February 16 to 27, 2026 and found 27% have an active online sports betting account, up from 22% in 2025 and 19% in 2024. Among account holders, 60% reported chasing bets, up from 52% a year earlier, and 42% said they'd wagered more than they intended, up from 37%.

Chasing is not a personality flaw. It's the defining behavior of the disorder, and we've written about why loss chasing feels rational from the inside.

What this means if you're the person in the numbers

Here's the practical translation of everything above.

Being undiagnosed doesn't mean being fine. The entire lesson of the diagnosis-versus-prevalence gap is that the healthcare system is catching a sliver of this. Nobody asked you at your last physical. That's a fact about the system, not a fact about you. The national research gap is real, and the treatment supply shortfall behind it is worse.

You don't need a chart code to start. F63.0 is a billing artifact. Gambling disorder is defined by behavior: preoccupation, escalating stakes to get the same charge, repeated failed attempts to cut back, restlessness when you stop, going back the next day to get even, lying about it, and risking what matters to keep going. If several of those are true, that's the information you needed. A clinician confirming it later doesn't change what today calls for.

Access is the mechanism. Both of these research streams point at the same thing, and it isn't willpower. It's a product that's always open, always in your pocket, and engineered around intermittent reinforcement, the reward schedule that produces the most persistent behavior of any known. The lever you actually control is access. Not moderation targets. Not "only on Sundays." Cutting the account off is the intervention.

Concrete next steps, in order:

1. Block the apps and the payment rails on every device you own, including the tablet you forgot about.
2. Self-exclude where your state offers it, which puts a barrier between you and the account that you can't undo at 1 a.m.
3. Tell one person. Secrecy is the load-bearing wall of this thing.
4. Get real-time support for the gap between sessions. Join Cope Compass and build a plan, and download the app so something is with you at the moment a bet gets decided, not four days later at an appointment.

Nothing resets. A stretch of days without a bet is still yours, even if there's a bad one in the middle.

Common questions

Does this study prove sports betting causes gambling disorder?

No. It's an observational analysis of diagnosis rates, not a causal study. It shows that diagnosed rates diverged sharply between states that legalized and states that didn't. Epic's own limitations note that a real change in prevalence can't be cleanly separated from a change in coding or recognition, and that the analysis can't attribute trends to specific legal events.

Why did diagnoses fall in states that didn't legalize?

Nobody knows for certain, and that's the most interesting unanswered question in the data. It argues against a pure nationwide awareness effect, because rising awareness should have lifted both groups. It could also reflect differences in health system composition, funding, or reporting between the two sets of states.

How many people with gambling disorder actually get diagnosed?

A small minority. Researcher Rachel Volberg told NBC News that typically no more than 3 to 10% of people who would be diagnosed with disordered gambling actually access treatment services. The National Council on Problem Gambling estimates about 2.5 million U.S. adults meet criteria for a severe gambling problem in a given year, far more than the number appearing in medical records.

What is F63.0?

F63.0 is the ICD-10-CM code for pathological gambling, the code used to record what the DSM-5 calls gambling disorder. Epic's analysis also counted Z72.6, "gambling and betting," which is a lifestyle-problem code rather than a psychiatric diagnosis.

Was the study peer reviewed?

No. Epic Research publishes its own analyses of the Cosmos dataset. This one was not published in a peer-reviewed journal and is not stated as peer reviewed.

Sources

• Epic Research: Gambling Disorder Diagnoses Have Risen More Than 60% in States That Legalized Sports Betting (June 26, 2026): primary source for all diagnosis rates, methods, state classifications, and limitations
• NBC News: Gambling disorder diagnoses spike in states that legalized sports betting (June 26, 2026): Volberg quote, American Gaming Association response, context on federal prevalence data
• National Council on Problem Gambling: FAQ: prevalence estimates for severe and mild-to-moderate gambling problems
• Hollenbeck, Larsen and Proserpio, "The Financial Consequences of Legalized Sports Gambling" (working paper, April 2025): free version of the paper published in Management Science, which is paywalled
• UCLA Anderson Review: As States Permitted Online Sports Gambling, Citizens' Personal Financial Health Suffered (February 4, 2026): bankruptcy, collections and delinquency effect sizes
• Siena Research Institute and St. Bonaventure University: More Than a Quarter of Americans, 27%, Have an Active Online Sports Betting Account (April 13, 2026): account ownership trend and chasing figures