This paper examined 734 individuals already classified as clinically high risk for psychosis and evaluated whether baseline cannabis use, tobacco use, or cannabis-tobacco co-use predicted later conversion to psychosis over approximately two years.

The most attention-grabbing finding involved a subgroup reporting heavy cannabis use combined with light tobacco use, which showed higher observed conversion risk than people reporting no cannabis or tobacco exposure.

But several related analyses weakened after adjustment, some findings became statistically borderline, and the study relied on broad self-reported exposure categories without cannabinoid composition, potency, route, or cumulative-dose detail.

The most clinically responsible interpretation is that heavier cannabis exposure may identify increased vulnerability within already high-risk psychiatric populations, not that the study proved cannabis exposure independently caused psychosis.

The investigators used data from the North American Prodrome Longitudinal Study-2 (NAPLS2), a large multicenter cohort following individuals identified as clinically high risk for psychosis.

Participants completed baseline substance-use assessments and longitudinal psychiatric follow-up over approximately two years. The study evaluated whether baseline cannabis use, tobacco use, or cannabis-tobacco co-use patterns were associated with later conversion to psychosis.

Exposure was measured using broad self-reported frequency categories over the prior 30 days. Participants were grouped into cannabis-only, tobacco-only, cannabis-plus-tobacco co-use, other-substance-use, and no-substance-use categories, with additional light-versus-heavy exposure subgroup analyses.

This was not a randomized trial. Researchers observed naturally occurring substance-use patterns within an already vulnerable psychiatric population.

Higher cannabis-use frequency showed an association with higher conversion risk in one age- and sex-adjusted Cox model.

But when cannabis use was modeled while controlling for tobacco exposure, the association weakened and no longer reached conventional statistical significance.

Categorical cannabis-tobacco co-use also failed to reach conventional statistical significance in one analysis.

The subgroup producing the strongest signal involved heavy cannabis use combined with light tobacco use, which showed elevated conversion risk relative to no cannabis/no tobacco exposure.

One nuance that deserves more attention is that the strongest signal did not emerge in the heavy cannabis/heavy tobacco subgroup. Instead, it appeared in the heavy cannabis/light tobacco subgroup. That complicates any simplistic interpretation that “more tobacco plus more cannabis equals more psychosis risk,” and raises the possibility of subgroup instability, interaction effects, vulnerability clustering, or chance findings that require replication.

Another important finding received much less public attention: the different clinical-high-risk substance-use groups did not significantly differ in baseline psychosis symptom severity.

This is a meaningful prospective cohort study published in a high-impact journal using a relatively large multicenter sample. The longitudinal design is stronger than purely cross-sectional association studies because it followed participants over time before conversion outcomes occurred.

But the study still remains observational, which means it cannot fully separate cannabis exposure from shared vulnerability factors such as trauma, psychiatric severity, genetics, social adversity, impulsivity, developmental instability, or other substance use.

The exposure definition was also remarkably broad. “Cannabis use” in this paper included highly heterogeneous real-world exposures compressed into self-reported frequency categories without THC potency, CBD content, route, product type, cumulative dose, or simultaneous-versus-asynchronous tobacco-use detail.

The study also relied on baseline past-30-day self-reported exposure categories collected between 2009 and 2013. That predates much of the modern retail cannabis market, including widespread concentrates, contemporary vaping patterns, and many high-potency commercial products.

Another important limitation is that subgroup findings are statistically vulnerable to instability. The strongest signal emerged in a narrower intensity-defined subgroup rather than consistently across all co-use models.

The population was already highly vulnerable. These participants were not representative of the general population. They were already classified as clinically high risk for psychosis before substance-use analyses occurred.

The exposure definition was crude. The study could not distinguish THC potency, cannabinoid composition, route, cumulative dose, developmental timing, or simultaneous versus asynchronous co-use.

The broader co-use finding was weaker than many headlines will imply. The categorical cannabis-tobacco co-use model itself did not reach conventional statistical significance. The strongest signal emerged only after dividing participants into narrower intensity-defined subgroups.

Residual confounding remains substantial. Shared vulnerability factors such as trauma exposure, family history, psychiatric burden, developmental instability, and social adversity could plausibly explain part of the observed associations.

E-values do not eliminate the main uncertainty problems here. E-values address one dimension of unmeasured confounding, but they do not protect against subgroup instability, exposure misclassification, sparse-event amplification, or analytic multiplicity.

The findings come from an older cannabis-exposure era. Data collection occurred before many contemporary retail cannabis products and modern vaping behaviors became widespread.

This paper does not prove that cannabis causes psychosis.

It does not establish that all cannabis products carry equal psychiatric risk.

It does not determine whether cannabis use functioned as a causal exposure, a coping mechanism, a marker of psychiatric vulnerability, or some combination of those possibilities.

It does not establish that most cannabis users face major psychosis risk.

It does not provide product-specific guidance regarding THC potency, CBD ratios, inhalation methods, or safer-use thresholds.

It does not isolate cannabis exposure from trauma, genetics, social adversity, psychiatric severity, or polysubstance use.

And it does not justify broad public narratives claiming cannabis exposure alone explains psychotic illness across populations.

The broader cannabis-psychosis literature has repeatedly identified concerning association signals, especially around heavier cannabis exposure, earlier onset, relapse risk, and vulnerable psychiatric populations.

But one reason the literature remains debated is that observational studies continue struggling to disentangle cannabis exposure from genetics, developmental instability, trauma burden, social adversity, psychiatric severity, and other overlapping vulnerability structures.

This paper fits within that broader pattern. It reinforces that concern around heavy THC exposure in vulnerable populations is reasonable, while also reinforcing how difficult it remains to identify a clean independent causal effect.

The study is therefore best understood as an incremental and hypothesis-generating contribution rather than a definitive settlement of the cannabis-psychosis debate.

What catches my attention most here is not simply the association signal itself, but how easily a paper like this can become flattened into a simplistic cultural headline. The study identified a meaningful risk signal inside a very vulnerable psychiatric population. That deserves serious attention. But it is still very different from proving cannabis broadly causes psychosis.

The part I would be careful with is the exposure definition. “Cannabis use” in this paper includes highly heterogeneous real-world exposures compressed into broad self-reported frequency categories. Clinically, THC potency, CBD composition, developmental timing, route of administration, and reason for use may matter enormously.

One of the most important humility points in this paper is that the participants were already considered clinically high risk before cannabis exposure entered the analysis. That makes this a vulnerability-framing study more than a population-wide causation study. In practice, I think that distinction gets lost constantly once papers like this become headlines.

At the same time, I would not dismiss the findings entirely. In real clinical care, I do think heavy THC exposure deserves thoughtful caution in people with prodromal symptoms, paranoia, strong family history, severe dissociation, or unstable psychiatric trajectories. But this paper should be read as an important observational signal, not a deterministic causal map.