2.8 Million Americans Are Not Vomiting From Weed

The New York Times just repeated a debunked statistic to justify a policy reversal. Here's what the consumption data actually show.

On February 9, 2026, The New York Times editorial board published a remarkable about-face. Having championed marijuana legalization in a six-part series in 2014, the paper now declared it was "time to acknowledge reality" and admit that America has "a marijuana problem." Among the evidence marshaled for this reversal: "Each year, nearly 2.8 million people in the United States suffer from cannabinoid hyperemesis syndrome, which causes severe vomiting and stomach pain."

That number — 2.8 million — has been ricocheting through policy debates, JAMA patient pages, and anti-legalization advocacy for years. It originates from a single 2018 study by Joseph Habboushe and colleagues at Bellevue Hospital in New York City. It is, as I demonstrate in a new paper currently under submission to medRxiv, almost certainly wrong by two orders of magnitude.

The real number is probably closer to 17,700.

Where the 2.8 Million Comes From

In 2018, Habboushe et al. surveyed 155 patients at a single urban public hospital emergency department who reported smoking cannabis at least 20 days per month. Using a symptom-based questionnaire — do you experience nausea, vomiting, or abdominal pain? does hot water help? — they determined that 32.9% met criteria for cannabinoid hyperemesis syndrome, or CHS.

The researchers then extrapolated this to the national daily-use population and arrived at approximately 2.75 million affected Americans. That figure was published in the paper itself, picked up by The New York Times, cited in a 2024 JAMA Patient Page, and amplified by advocacy organizations like Smart Approaches to Marijuana (SAM). A 2026 nationally representative survey (NFACS) produced a somewhat lower but still enormous estimate: 17.8% prevalence, translating to over 7 million affected adults.

These are staggering numbers. If true, they would mean that roughly one in three to one in five daily cannabis users is experiencing a condition so severe that clinical consensus identifies complete cessation of cannabis as the only definitive treatment. The Rome IV criteria, the Simonetto criteria, and standard clinical guidance all agree: if you have CHS, you have to stop using cannabis. There is no pharmaceutical alternative. The vomiting stops when the cannabis stops.

This creates a testable prediction. If tens of millions of daily cannabis users include 2.8 to 7 million people experiencing intractable vomiting that resolves only with abstinence, then aggregate cannabis consumption data should show a massive, ongoing hemorrhage of heavy users from the market. These aren't casual weekend consumers drifting away — they're the heaviest users in the distribution, consuming 4 grams per day according to the largest clinical survey of CHS patients (Russo et al. 2022). They account for a wildly disproportionate share of total market volume. Their departure would be unmistakable.

It doesn't show up. Anywhere. In any market. Over any time horizon.

319 Weeks of Silence

Florida operates the largest medical cannabis program in the United States, with over 932,000 active patients. It is medical-only — there is no recreational market to absorb departing patients. Maintaining an active card requires annual re-application and approximately $332 in fees and physician visits. There is no auto-renewal. The default state is unenrolled. When patients stop buying, they simply disappear from the registry.

Every gram of flower and every milligram of THC dispensed in Florida is tracked through a seed-to-sale system and reported weekly by the Office of Medical Marijuana Use (OMMU). I compiled 319 consecutive weekly reports covering January 2020 through February 2026 — over six years of continuous, gram-level dispensation data.

Here is what the data show: per-patient flower consumption tripled from 0.21 grams per day in early 2020 to 0.64 grams per day in late 2025. That's roughly equivalent to 1.0 grams per day when factoring additional, non-flower products. Active patient counts rose from approximately 400,000 to 932,000. Total dispensed volume grew in almost every quarter. There is no attrition signal. No plateau. No structural contraction. No cohort of heavy users vanishing from the registry.

Under Habboushe's 32.9% prevalence estimate, Florida's registry would contain approximately 307,000 CHS patients. At the cessation rate documented by Russo et al. (87.7%), roughly 269,000 patients consuming 4 grams per day would exit the market. Their collective departure would remove approximately 393 metric tons of flower-equivalent annually from a system tracking every gram — a volume contraction of roughly 20%. This signal is absent from 25 consecutive quarters of data.

At the NFACS estimate of 17.8%, the implied annual volume loss is still nearly 11% — impossible to miss in gram-level tracking data. At 5% prevalence, the quarterly loss would match the largest quarter-over-quarter decline ever observed in the Florida data (-5.3% in Q2 2022, which was followed by an immediate rebound driven by normal seasonal and purchasing variability). Even at 1%, you'd see a persistent drag. Even at 0.5%, with Florida's patient growth now decelerated to roughly 3.3% annually, more than a quarter of all new patient volume would be consumed simply backfilling CHS exits — leaving no room for the continued rise in per-patient consumption the data actually show.

The math doesn't work. It doesn't come close to working.

The Destigmatization Problem

Before examining where the prevalence estimates went wrong, it's worth addressing the headline statistic that anchors the entire NYT editorial: the claim that daily cannabis use has "surged" from under 1 million in 1992 to 18 million today.

This is real NSDUH data, and the trend line is genuine. But what does it actually measure? The National Survey on Drug Use and Health is a self-reported household survey. It asks people whether they use an illegal (or until recently, quasi-legal) substance. The response rate to that question is not independent of the social and legal environment in which it's asked.

In 1992, cannabis was illegal everywhere. Possession could result in jail time. The Reagan-Bush drug war had been running for over a decade. Answering "yes, I smoke marijuana daily" on a government-affiliated survey carried real perceived risk. By 2022, cannabis was legal for adult use in 21 states and medically legal in 37. The social and legal calculus of admitting use had fundamentally changed.

We have a near-perfect natural experiment for what destigmatization does to self-reported survey data, and it has nothing to do with drugs.

Gallup has tracked LGBTQ+ identification in America since 2012. That year, 3.5% of U.S. adults identified as lesbian, gay, bisexual, or transgender. By 2024, that number had reached 9.3% — nearly tripling in twelve years. Among Generation Z adults, more than one in five now identifies as LGBTQ+.

Did America's LGBTQ+ population actually triple in a decade? Of course not. What changed was the willingness to disclose. As legal protections expanded, social acceptance grew, and the perceived cost of identification dropped, people who had always been LGBTQ+ began saying so on surveys. Gallup itself notes that "the pronounced generational differences raise questions about whether higher LGBT identification in younger than older Americans reflects a true shift in sexual orientation, or if it merely reflects a greater willingness of younger people to identify as LGBT."

The parallel to cannabis self-reporting is direct. In 1992, fewer than a million Americans told a government survey they used cannabis daily. In 2022, 17.7 million did. Some of that increase reflects genuine growth in use — legalization expanded access, reduced prices, and introduced new products and consumption methods. But some of it — potentially a large fraction — reflects the same destigmatization effect that tripled LGBTQ+ self-identification. People who were already using daily simply became willing to say so.

This distinction matters enormously for interpreting CHS prevalence. If the denominator of "daily users" was artificially suppressed in earlier surveys by stigma-driven underreporting, then the apparent surge in use is partly an artifact of measurement — and the population at risk for CHS hasn't grown as dramatically as the raw numbers suggest. More importantly, if you're building a prevalence estimate on top of a denominator that's itself shifting for reasons unrelated to actual behavior change, the resulting extrapolation inherits all of that instability.

The NYT editorial treats the 1-million-to-18-million trend as an uncomplicated story of escalating use. It isn't. It's a story of escalating disclosure, with genuine use growth layered on top. Disentangling the two is hard. But citing the raw numbers as proof that legalization "led to much more use" — without acknowledging that the survey instrument itself responds to the policy environment — is analytically incomplete in a way that matters for every downstream claim built on those numbers, including CHS prevalence.

Where the Surveys Go Wrong

The Habboushe methodology has several well-documented problems that I detail at length in the paper, but the core issue is simpler than any of them: the estimate fails a basic reality check against consumption data.

Survey-based CHS prevalence estimates suffer from the same fundamental weakness as all self-reported symptom surveys applied to common complaints. Nausea, vomiting, and abdominal pain are among the most frequently reported symptoms in the general population. When you ask daily cannabis users whether they've experienced these symptoms and whether hot water provides relief, you're capturing a broad swath of gastrointestinal distress that may have nothing to do with cannabis.

The Russo et al. (2022) CHS patient survey — the largest clinical characterization of confirmed CHS patients — illustrates this directly. Among their 205 confirmed CHS patients, 48.8% had also been diagnosed with cyclic vomiting syndrome (a condition with identical symptoms and no cannabis etiology), 32.5% had migraine, and 28.3% had irritable bowel syndrome. Among women who had been pregnant, 57.9% reported a history of hyperemesis gravidarum — a rate 20 to 40 times the general population prevalence.

This isn't a portrait of a common drug reaction. It's a portrait of a genetically susceptible subpopulation prone to vomiting disorders broadly. Russo et al. reached this conclusion themselves, characterizing CHS as "a manifestation of gene–environment interaction in a rare genetic disease unmasked by a toxic reaction to excessive THC exposure." A rare genetic disease. Not one in three daily users. Not one in five. Rare.

The hot shower criterion — often treated as the clinical hallmark that distinguishes CHS from ordinary vomiting — turns out to be even less specific than it appears. A 2023 study by Venkatesan et al. surveyed 346 cyclic vomiting syndrome patients who had never used cannabis. Forty-eight percent reported that hot baths or showers relieved their symptoms. A "diagnostic" criterion with a 100% positive rate in CHS patients and a 48% positive rate in non-cannabis vomiting patients is not diagnosing anything cannabis-specific. It's measuring how many sick people like hot showers.

What the Real Numbers Look Like

If CHS is genetic — as Russo et al. concluded — then the susceptibility rate applies to the general population, not specifically to cannabis users. A reasonable prevalence estimate of approximately 0.1% among the general adult population yields about 258,000 Americans who carry the genetic vulnerability. Only those who actually consume cannabis daily at sufficient intensity would manifest symptoms.

At 17.7 million daily or near-daily cannabis users (Caulkins 2025), 0.1% gives approximately 17,700 active CHS cases nationally. This is consistent with population-level ED data: Shalaby (2025), analyzing the Nationwide Emergency Department Sample covering 188 million ED visits, found CHS-coded encounters represent roughly 0.05% of all visits. A separate analysis of the Epic Cosmos database identified 134,059 CHS encounters across 248 million ED visits over nine years — again, a tiny fraction.

Seventeen thousand cases is not trivial. CHS is a real condition, and for the people who have it, it is genuinely miserable — cyclic vomiting, dehydration, sometimes hospitalization. It deserves clinical attention and continued research. Russo's work on genetic susceptibility is valuable precisely because it points toward diagnostic tools (his company, CReDO Science, has a patent pending on a DNA test) that could identify susceptible individuals before they develop symptoms.

But 17,700 is not 2.8 million. It's not 7 million. It's two orders of magnitude smaller. And the difference matters because the larger numbers are being used to justify specific policy interventions — federal excise taxes, potency caps, restrictions on product types — that affect tens of millions of consumers based on a prevalence estimate that cannot survive contact with the very consumption data it purports to describe.

The Antabuse Trap

The logic of high-prevalence CHS contains a built-in contradiction that its proponents cannot escape.

Approximately 560 million people worldwide carry the ALDH2*2 genetic variant, which causes acute nausea, vomiting, and flushing during alcohol consumption. The behavioral response across this population is unambiguous and well-documented: carriers drink less or abstain entirely. This effect is so well-established that an entire pharmaceutical category was built on it. Disulfiram (Antabuse), FDA-approved since 1951, works by artificially inducing the same reaction, and it has been prescribed to millions of patients over seven decades on the explicit premise that acute illness during substance consumption produces behavioral change.

CHS, as described in the clinical literature, claims to produce this same aversive response — severe, recurrent vomiting triggered by cannabis consumption — except it supposedly occurs in 17 to 33% of daily users. Yet where ALDH2 deficiency produces a measurable, population-level reduction in alcohol consumption across hundreds of millions of carriers, CHS at claimed prevalence rates produces no detectable reduction in cannabis consumption across any observed market.

The forced choice is simple: either CHS causes people to stop (in which case, where's the consumption decline?), or it doesn't cause them to stop (in which case, why is it severe enough to drive policy?). There is no version of this story where one in five daily users is experiencing intractable vomiting and the market just... doesn't notice.

What This Means for the Policy Debate

The New York Times editorial board is not wrong that cannabis policy deserves serious reassessment. The paper's broader points — about inadequate regulation, inconsistent product testing, the failure of social equity programs, and the need for better research — are well-taken. The rollout of legal cannabis in states like New York has been a genuine mess. There are legitimate public health concerns about potency, adolescent access, and impaired driving that warrant evidence-based policy responses.

But evidence-based means exactly that. When you tell 18 million daily cannabis users that nearly 3 million of them are suffering from a vomiting syndrome — and you build federal tax proposals and regulatory frameworks on that claim — you need the number to be right. The consumption data say it isn't right. It isn't close to right.

The danger of inflated prevalence estimates isn't just statistical imprecision. It's that they provide intellectual ammunition for policies that would otherwise have to stand on their own merits. A federal excise tax on cannabis can be debated on fiscal, public health, and equity grounds without pretending that one in three daily users is vomiting uncontrollably. Potency regulations can be evaluated on neuroscience and adolescent development research without citing a debunked ED survey from a single New York hospital. Product safety standards can be justified by contamination data from unlicensed markets without inflating a rare genetic condition into an epidemic.

The prevalence claims also cause real harm to the people they're supposed to help. When emergency physicians are told that a third of their daily-cannabis-using patients have CHS, every unexplained vomiting episode in a cannabis user gets attributed to the drug — regardless of actual etiology. Patients with cyclic vomiting syndrome, gastroparesis, or functional GI disorders who happen to use cannabis are told to stop using and sent home. The over-diagnosis crowds out differential diagnosis, which is the opposite of good medicine.

CHS is real. It is rare. It is almost certainly genetic. And 2.8 million is not a number — it's a policy weapon dressed up as epidemiology.

The full preprint, including the complete 319-week Florida OMMU dataset, sensitivity analyses at multiple prevalence and consumption tiers, and the cumulative attrition model, will be available on medRxiv. The underlying data are archived on Harvard Dataverse or https://knowthefactsmmj.com/about/weekly-updates/ for independent replication.