The Tobacco Exclusion That’s Distorting Every Psychosis-Prevention Trial You’re Running

She was 21, Spanish-monolingual, referred from a community mental health clinic in Anaheim with a structured interview confirming attenuated psychotic symptoms and a SOPS score that put her squarely in the clinical high-risk window. She also smoked half a pack a day — had since she was 16 — and told us through our interpreter that she’d tried to quit twice and couldn’t. We were screening for a psychosis-prevention trial at the time. The protocol’s exclusion criteria read: “current tobacco or nicotine product use within 30 days of baseline.” She was gone before we finished explaining the rest of the visit. That clause, sitting quietly in a single line of a 47-page protocol, did what it was designed to do.

A preprint published last week on medRxiv from the North American Prodrome Longitudinal Study (NAPLS2) offers a neurobiological reason why that exclusion is also scientifically indefensible. Using resting-state fMRI data from 211 CHR-P participants and 132 healthy controls — mean age 19 across both groups — the investigators examined connectivity from the left lateral parietal node of the default mode network to the rest of the brain. Two clusters emerged: the right hippocampus (peak voxel MNI +30, -12, -27) and the left parahippocampus, where higher LLP-cluster connectivity was significantly associated with more frequent tobacco use. Among current tobacco users, a tobacco-by-CHR diagnosis interaction on LLP–right hippocampus connectivity reached significance (estimate=0.306, SE=0.149, t=2.051, p=0.045), driven entirely by the CHR-P group. Heavier smoking was associated with hippocampal hyperconnectivity in clinical high-risk participants but not in healthy controls. The relationship between tobacco use and default mode network connectivity, it turns out, looks different in brains already tracking toward psychosis.

The Exclusion Clause Nobody Audits

That finding has a direct operational implication for every psychosis-prevention trial currently recruiting: if tobacco use is neurobiologically distinct in CHR-P populations — linked to the same DMN-hippocampal circuitry that antipsychotic and psychosocial interventions are trying to normalize — then excluding tobacco users does not produce a cleaner sample. It produces a biologically unrepresentative one, and it produces it selectively, because tobacco use clusters where psychotic risk clusters.

To be fair to the instinct behind the clause: excluding tobacco users does buy a measure of internal validity — fewer confounds on a connectivity or cognitive endpoint, and a tighter signal. The honest question is what that tightness is worth if it describes a sample the real CHR-P population does not resemble. This is the familiar tradeoff between internal validity and generalizability, and in psychosis prevention it tilts hard toward generalizability, because the patients removed are not a random slice — they are the highest-smoking, highest-risk, most underserved fraction of the target group. And the modern version of this clause rarely stops at cigarettes: a “tobacco or nicotine product” criterion now sweeps in vaping, e-cigarettes, and pouches, which — given how common vaping has become among adults under 25 — can quietly widen the excluded group well beyond traditional smokers, in exactly the young CHR-P population these trials most need to reach.

The prevalence numbers make this impossible to ignore. In NAPLS2 itself, CHR subjects were three times more likely to be light smokers than unaffected subjects (OR 3.0, 95% CI 1.9–5.0) and 4.8 times more likely to be heavy smokers (OR 4.8, 95% CI 1.7–13.7). A systematic meta-analysis published in the Journal of Clinical Psychiatry found that 58.9% of patients with first-episode psychosis were tobacco users at the time they first presented for treatment. And a 2023 analysis of 29,045 community-dwelling adults found that adults with a lifetime diagnosis of psychosis had an adjusted past-month tobacco use prevalence of 41.3%, compared to 27.7% in adults without psychosis (adjusted RR 1.49, 95% CI 1.36–1.63). One precision caveat for trial designers: the 58.9% and 41.3% figures describe first-episode and established-psychosis cohorts, not CHR-P samples specifically; the CHR-P-specific estimates are the NAPLS2 odds ratios above and a pooled three-to-five-times smoking rate. That distinction matters when setting eligibility for a clinical-high-risk sample, even though all of these populations point the same way. These are not marginal smokers drifting into your sample. They are the population you claim to be studying.

When a CHR-P trial excludes current tobacco users, it is removing somewhere between a third and a half of its target population before the first baseline visit. The patients who remain are systematically less symptomatic, less substance-burdened, and — as the NAPLS2 imaging data now demonstrates — operating with a measurably different DMN-hippocampal architecture than the patients who were turned away. Every efficacy signal you generate in that cleaned sample is conditional on a patient profile that does not dominate the real CHR-P population. And the patients being removed are disproportionately young, lower-income, and from communities — Hispanic, Black, safety-net-system-dependent — where smoking prevalence is higher and cessation support is structurally unavailable.

Where the Signal Lives, and Who Gets Excluded to Find It

A companion preprint from earlier this year on DMN connectivity in psychosis reported a significant diagnosis-by-DMN interaction for lifetime tobacco use (p=0.0281, coefficient=0.457, OR 1.579, 95% CI 1.063–2.411), with higher DMN connectivity associated with increased tobacco likelihood specifically in individuals with psychosis. Taken together, these two datasets point in the same direction: tobacco use and DMN-hippocampal dysconnectivity are entangled in CHR-P and psychotic populations in ways that cannot be controlled away by exclusion. None of this establishes that tobacco use causes the dysconnectivity — the data are cross-sectional, and smoking may be a marker of underlying risk as much as a contributor to it — but the association is enough to carry the design point. They can only be studied — or they can be hidden. Most trials are currently choosing to hide them.

The mechanism behind routine tobacco exclusion is rarely malicious. Protocol teams inherit the clause from prior iterations, or borrow it from addiction-trial templates where nicotine metabolism genuinely confounds the pharmacokinetic primary endpoint. In a psychosis-prevention trial with a functional MRI or cognitive primary endpoint, that rationale does not transfer — but the exclusion clause does, because no one on the protocol team is required to justify it against the demographic profile of CHR-P patients at community psychiatric sites. The FDA’s 2024 Diversity Action Plan guidance, issued under FDORA, requires sponsors to address enrollment barriers for underrepresented populations and to explain eligibility criteria that may disproportionately exclude them. An eligibility criterion that removes 40–59% of the target population, clustered by race and socioeconomic status, is exactly the kind of design choice that framework is meant to bring into the open. The point is not that the FDA singles out tobacco exclusions for disapproval — it does not — but that a Diversity Action Plan obliges the sponsor to justify such criteria against the populations they screen out, rather than leaving them unexamined. At our site, it never comes up in the sponsor’s DAP submission. The criterion is treated as a scientific given rather than an enrollment policy.

Safety-net recruitment compounds the problem geometrically. Research conducted at an urban safety-net hospital — drawing on 28 semi-structured interviews with investigators, coordinators, and research nurses between October and December 2022 — identified safety-net systems as requiring additional infrastructure investment to support trial participation, with language access, transportation, and patient trust among the named structural barriers. CHR-P patients who present at community mental health clinics in Anaheim, Compton, or East Los Angeles are already navigating those barriers. Adding a tobacco-cessation prerequisite — which is the practical implication of a “no current nicotine use” criterion for a patient who smokes — means asking a 20-year-old with attenuated psychosis to complete a cessation program before she qualifies for a trial studying whether we can prevent her psychosis. The sequencing is backwards.

What Stratification Instead of Exclusion Actually Looks Like

The fix the NAPLS2 data points toward is stratification, not exclusion. The investigators themselves tracked tobacco use frequency as a continuous variable and used it to map a neurobiology — which means tobacco-using CHR-P patients are analytically tractable, not noise. A trial designed with tobacco-use status as a stratification variable at randomization, and LLP-hippocampal connectivity as either a secondary endpoint or a biomarker-enrichment criterion, would generate exactly the kind of subgroup data the field is missing: does a given psychosocial or pharmacological intervention close the DMN-hippocampal gap in heavy-smoking CHR-P patients, and does that closure predict conversion outcomes? That is a scientifically important question. No trial has been designed to answer it, because the patients who would answer it are sitting in a screen-failure log.

The hippocampal connectivity literature does support the biological plausibility of using these circuits as trial endpoints. A longitudinal resting-state fMRI study enrolled 34 unmedicated patients with schizophrenia or schizoaffective disorder, scanned them off medication, treated them with risperidone for six weeks, and rescanned — finding measurable change in hippocampal functional connectivity with treatment. If antipsychotic intervention moves these circuits in established schizophrenia, a prevention trial with a tobacco-stratified imaging arm in CHR-P is not a speculative design. It is the logical next step.

The operational requirement is adding tobacco-use frequency as a mandatory stratification variable in CHR-P trial randomization schemas — not as a post-hoc covariate buried in a sensitivity analysis, but as a prospectively powered stratum. That means protocol teams need to stop treating tobacco exclusions as legacy boilerplate and start treating them as eligibility decisions that require the same scientific justification as any other criterion. Before protocol lock on any psychosis-prevention or early-intervention trial, the eligibility-criteria audit should include a written response to a single question: given that CHR-P patients smoke at 3–5 times the rate of healthy controls, what is the scientific rationale for excluding them, and what enrollment percentage does that exclusion remove at community psychiatric sites serving Hispanic and Black young adults?

The next data point to watch is the NAPLS2 longitudinal conversion analysis — whether tobacco-associated LLP-hippocampal hyperconnectivity at baseline predicts transition to full psychosis over the 24-month follow-up window. If it does, that connectivity signature becomes a candidate biomarker for enriched enrollment in prevention trials, and the case for stratifying rather than excluding tobacco-using CHR-P patients moves from scientifically reasonable to scientifically necessary.

References

Mumbi Kairu

+ posts

Mumbi Kairu is the Director of Clinical Research at the Pamoja Clinical Institute in Anaheim, California.

Moe Alsumidaie

Chief Editor