A cohort study following more than 460,000 adolescents found that past-year cannabis use at ages 13 to 17 was associated with a significantly elevated risk of being diagnosed with a psychotic, bipolar, depressive, or anxiety disorder by age 26. This is one of the largest studies of its kind, and it lands squarely on the question our own Risk and Age content treats as the one requiring the most care.
What the study design can and cannot tell you. This is an observational cohort, not a randomised trial, nobody was assigned to use cannabis or not. That means it can show a strong, consistent association across a very large sample, which this does, but it cannot by itself prove that cannabis use caused the later diagnoses. Shared risk factors, family psychiatric history, other substance use, socioeconomic stress, could plausibly contribute to both the cannabis use and the later diagnosis. The researchers are not claiming otherwise; good observational work says what it found and is explicit about what it cannot rule out. We are repeating that caveat here because a lot of secondary coverage drops it.
Why the timing detail matters more than the topline finding. The consistent pattern across this kind of research is that the risk signal is strongest for use that begins in adolescence specifically, not for cannabis use in general. That lines up with basic neuroscience: the prefrontal cortex continues developing into the mid-twenties, and the endocannabinoid system is unusually active during that developmental window. It is the reason public health guidance, CDC included, frames the policy response around youth-access prevention specifically, age verification, retailer licensing, honest age-specific education, rather than general abstinence messaging aimed at all ages equally.
Where this sits in the wider evidence. This finding does not stand alone. It is consistent with a broader literature on adolescent-onset cannabis use and psychosis risk that we summarise on The Plant page, in the harms section. What this study adds is scale: 460,000 people is a sample size that makes sampling noise an unlikely explanation for the pattern, whatever else is driving it.
We treat this as exactly the kind of finding our Risk and Age framing exists for: real evidence about a developmental window, stated as risk, not as a reason to write dosing advice for anyone.