Published on 18 September 2026
Cannabis legalization in Canada: did use rise more among people with a mental health condition?
The essentials
Six waves of a national survey in Canada, 92,843 respondents aged 16 to 65 between 2018 and 2023, that is, one year before and five years after the legalization of recreational cannabis in October 2018. The regression models include 91,907 respondents. The fear voiced at the time was that opening the market would hit the most psychologically vulnerable people first. That is not what these data show, but not for the reason often given either. Reported use is roughly twice as common among people reporting a mental health condition, and this gap was already there in the wave before legalization. It did not widen afterwards. It narrowed, because the rise was steeper among people reporting no mental health condition. Among people reporting an anxiety disorder or depression, past 12-month use rises from 2019 and stays above the 2018 level in every following year, through 2023: this is not a transient signal. No significant change from 2018 is found for bipolar disorder, post-traumatic stress disorder or schizophrenia. Mental health status is self-reported rather than clinically established, and recruitment relies on an online panel: two limitations that weigh on interpretation.
Context
The question has no direct application in France, where recreational cannabis is not legal. It is useful in a precise way in the consulting room and in public debate, because it brings data to a field dominated by convictions. When a patient, a relative or a colleague argues that legalization would make consumption soar among vulnerable people, there is now something other than an opinion to offer in reply.
Canada provides a workable setting: a clear switch date, a survey system already running before the reform, and five years of observation afterwards. It is still a before-and-after comparison, not an experiment.
The study at a glance
| Population | |
| Residents of Canada aged 16 to 65, recruited through an online panel, with or without a self-reported mental health condition. Five categories analyzed: anxiety disorder, bipolar disorder, depression, post-traumatic stress disorder, and schizophrenia or psychotic disorder grouped together. A total of 92,843 respondents over six annual waves, 91,907 in the regression models after exclusion of non-responses on mental health status, 1,090 people, and on education level, 936 people. | |
| Exposure | |
| The period after the legalization of recreational cannabis, which came into force in October 2018, with annual waves from 2019 to 2023. | |
| Comparison | |
| The 2018 wave, before legalization, taken as the reference. Models are stratified by condition: each diagnostic group is compared with itself from one wave to the next, and a separate group brings together respondents reporting none of the five conditions. | |
| Outcomes | |
| Two outcomes. Past 12-month cannabis use, defined as any use over the period. Daily use, defined as use on five or more days a week at the time of the survey. | |
| Design | |
| National repeat cross-sectional surveys from the International Cannabis Policy Study. Binary logistic regression, adjusted and unadjusted results, pairwise contrasts between years. Adjustment for sex at birth, age group, ethnicity, education level, province and perceived income adequacy, with survey year as an indicator variable. Post-stratification weighting calibrated with a raking algorithm. Analyses in SAS version 9.4. Public funding from the Canadian Institutes of Health Research. |
Results
| Outcome | What the data show |
|---|---|
| Gap in use between groups, main effect | Adjusted odds ratios from 1.55 to 2.41 for past 12-month use, depending on the condition |
| ReadingOver the whole period, past 12-month use is associated with anxiety disorder, adjusted odds ratio 1.96 (95% CI 1.88 to 2.04), bipolar disorder 2.41 (2.17 to 2.68), depression 2.03 (1.94 to 2.12), post-traumatic stress disorder 2.23 (2.09 to 2.39), and schizophrenia or psychotic disorder 1.55 (1.36 to 1.78). For daily use, adjusted values range from 1.56 to 2.38. All of these values come from the supplementary material. This is the most robust finding of the study, and it is descriptive. | |
| Change in the gap between 2018 and 2023 | Gap narrowed, with no dedicated statistical test |
| ReadingThis point matters, and it is where a quick reading goes wrong. The authors do not formally test the change in the gap: there is no interaction term between year and mental health status, only stratified models. On the published prevalences, the ratio between the anxiety disorder group and the group with no condition falls from 2.13 in 2018 to 1.70 in 2023, and for depression from 2.17 to 1.79. In percentage points, past 12-month use rises by 6.5 points among people reporting no condition, against 1.9 points for anxiety disorder and 3.6 points for depression. Saying that the gap stayed stable would be inaccurate. | |
| Self-reported anxiety disorder, past 12-month use | Rise from 2019, sustained through 2023 |
| ReadingAdjusted odds ratio 1.33 in 2019 versus 2018 (95% CI 1.15 to 1.53, p < 0.0001). Prevalence stays above its 2018 level in every following year, including 2023: adjusted odds ratio 1.18 (1.03 to 1.34, p = 0.019). A decline is observed between 2021 and 2022, 0.89 (0.81 to 0.98, p = 0.021). | |
| Self-reported depression, past 12-month use | Same pattern, rise sustained through 2023 |
| ReadingAdjusted odds ratio 1.47 in 2019 versus 2018 (95% CI 1.26 to 1.73, p < 0.0001). In 2023 versus 2018, 1.25 (1.07 to 1.45, p = 0.0040). The 2019 interval does not come close to 1: its lower bound is 1.26. | |
| No self-reported mental health condition, past 12-month use | Steepest rise, sustained across all waves |
| ReadingAdjusted odds ratio 1.52 in 2019 versus 2018 (95% CI 1.37 to 1.68, p < 0.0001), and 1.58 in 2023 versus 2018 (1.43 to 1.75, p < 0.0001). Prevalence goes from 20.8% to 27.3%. It is this group, not the clinical groups, that drives the post-legalization rise. | |
| Bipolar disorder, post-traumatic stress, schizophrenia | No significant change from 2018 |
| ReadingIn the adjusted models, no contrast between 2018 and a later year reaches significance for past 12-month use. Two isolated contrasts between 2022 and 2023 are reported, for bipolar disorder, 1.46 (1.09 to 1.96, p = 0.012), and for schizophrenia, 1.44 (1.00 to 2.08, p = 0.05), the latter interval touching the null value. These are analyses that were performed and proved inconclusive, which is not a demonstration that there is no effect: these strata have the smallest samples in the study, 3,001 for bipolar disorder, 6,522 for post-traumatic stress, 1,882 for schizophrenia, and their confidence intervals are very wide, up to 0.86 to 3.79 for a single contrast. No power calculation is reported. | |
| Daily use | Rise in 2019 only in the group reporting no condition |
| ReadingBetween 2018 and 2019, no change in daily use among people reporting an anxiety disorder, depression, bipolar disorder or post-traumatic stress disorder. Among people reporting no condition, adjusted odds ratio 1.34 (95% CI 1.10 to 1.62), and daily use stays above its 2018 level in every following year. Later rises are described between 2018 and 2021 for anxiety disorder, 1.34 (1.11 to 1.61), and for depression, 1.37 (1.13 to 1.68), followed by a decline in 2022 in both groups. For schizophrenia, the rise between 2018 and 2019 is not significant, 2.31 (0.97 to 5.52, p = 0.06). This outcome says more, clinically, than past 12-month use. | |
A note on wording, because it changes everything. Saying that use rose more among people without a mental health condition, and saying that the gap between the two groups stayed stable, are not equivalent statements, and only one of them matches this publication. It is the first. The authors state explicitly that prevalence increased more among people with no history of a mental health condition, with few changes among the others. The published prevalences point the same way: the ratio between the two groups shrinks from wave to wave. One methodological caveat remains, and it is the reader’s rather than the authors’: this comparison between strata is not tested statistically, and odds ratios from stratified models cannot be compared directly with one another when baseline prevalences differ, 44.4% versus 20.8% in 2018. The direction is clear, its magnitude is not.
Quality control
| Item checked | Verdict |
|---|---|
| Sample size and observation period | Substantial |
| Finding92,843 respondents described, 91,907 in the models, and six annual waves covering one year before and five years after the reform. This duration makes it possible to tell a transient effect from an established one, which few studies in this field allow. | |
| Sample weighting | Post-stratification |
| FindingCalibration to known population totals, age by sex by province, education level, and age by smoking status, then rescaling of the weights to each year’s analytic sample. It corrects observable composition, not the propensity to respond. | |
| Measurement of mental health status | Self-reported |
| FindingNo standardized diagnosis, one self-reported item per condition. Two measures are used, past 12-month experience and lifetime diagnosis, and the question was not asked identically at every wave: in waves 1 and 2, respondents first answered a question on lifetime experience before the question on the past 12 months, whereas waves 3 to 6 included only the latter. The authors themselves point out that bipolar disorder and borderline personality disorder are grouped together, which is a serious limitation in psychiatry. | |
| Recruitment method | Online panel |
| FindingNon-probability sampling through a commercial panel. The most vulnerable people, including those in unstable housing or with a severe, decompensated illness, are structurally under-represented. They are precisely the people the initial concern was about. | |
| Participation across waves | Declining |
| FindingThe cooperation rate falls from 64.2% in 2018 to 53.8% in 2023. A before-and-after comparison assumes that the unobservable composition of respondents does not change over the period, and falling participation makes that less likely. | |
| Social desirability | Changes over time |
| FindingReporting cannabis use does not carry the same weight before and after legalization. Part of the observed change may reflect willingness to report rather than behavior. The authors do not discuss this limitation. | |
| Multiple comparisons | No correction reported |
| FindingTwo outcomes, six groups, six waves, year-to-year contrasts and contrasts against 2018, in adjusted and unadjusted models, plus a sensitivity analysis on lifetime diagnosis. No procedure to control type I error is mentioned. An isolated signal in a single wave should be read with this caveat in mind. | |
| Arithmetic consistency of the published figures | Two discrepancies |
| FindingTwo points do not add up. The sample for the sensitivity analysis is given as 91,794, whereas the stated exclusions, 1,216 and 489 from 93,933, lead to 92,228. And several p values below 0.0001 are paired with confidence intervals that do not allow them, for example an odds ratio of 1.34 with an interval of 1.10 to 1.62. This does not invalidate the conclusions, but it is a reason to cite each value with its interval rather than its p value. | |
| Funding | Public |
| FindingCanadian Institutes of Health Research, Project Grant and Project Bridge Grant, reference PJT-153342. The authors state that the funder played no role in design, data collection, analysis, interpretation, writing or the decision to submit. No industry funding is declared. | |
| Conflicts of interest | One declared |
| FindingThe last author declares paid work as an expert witness on behalf of public health authorities, in response to legal challenges from the tobacco, vaping and cannabis industries. The other authors declare no conflicts of interest. This tie, if anything, cuts against conclusions favorable to industry, which settles nothing: it remains to check that the wording of the conclusions fits the data, and that is precisely where the main reservation of this analysis lies. | |
Critical appraisal
| Domain | Judgment |
|---|---|
| Fit between design and question | Appropriate |
| FindingFor a question about prevalence at the scale of a country, a repeat cross-sectional survey is the right tool. Randomization is impossible here, and a before-and-after comparison is the best available option. It supports no causal inference, neither about the effect of the law nor about the link between cannabis and mental health conditions. | |
| Register of the authors’ statements | Descriptive |
| FindingThe authors keep to a descriptive register and nowhere claim that legalization caused the observed changes. They do not, however, include an explicit warning that this design cannot support causal conclusions. The only causal statement in the text concerns the relationship between cannabis and mental health in general, in the introduction, not their own results. | |
| Faithfulness of the abstract to the results | Wording too smooth |
| FindingThe authors’ conclusion, that “few changes in cannabis use were observed” among Canadians reporting a mental health problem, smooths over a more nuanced result. For anxiety disorder and depression, past 12-month use rises from 2019 and remains significantly above 2018 every year through 2023. That is not nothing, and it deserves to be said before repeating the reassuring conclusion. The accurate, defensible statement is this: the rise was steeper among people reporting no mental health condition. | |
| Validity of the psychiatric exposure measure | Weak |
| FindingSelf-report puts very different clinical situations under a single label. The authors also note that the prevalence of conditions is higher than in the national reference surveys, 25% lifetime diagnoses of anxiety disorder in 2023 versus 15.3%. This imprecision tends to pull the groups closer together, and therefore to attenuate a real gap rather than create one. | |
| Reach of the reassuring conclusion | Limited |
| FindingThe population most at risk is the one the survey captures least well. An absence of signal among members of an online panel says nothing about people no panel recruits. The authors themselves recall that Canadian and US hospital data show, over the same period, an increase in cannabis use disorders and in associations with psychosis. | |
Level of evidence
Confidence is high for the description: cannabis use is roughly twice as common among people reporting a mental health condition, before and after the reform alike. It is moderate for the comparative trajectory of the two groups, because the direction is clear but no test formalizes it. It is low for any causal reading, and low as well for the most severe conditions, which have the smallest samples and the poorest representation.
What is demonstrated, in the sense of what the published data directly establish: cannabis use is markedly more common among people reporting a mental health condition, at every wave, and past 12-month use increased after 2018 in the group reporting no condition as well as in the anxiety disorder and depression groups, with a larger rise in the former. What is suggested: that legalization did not have a differential adverse effect on people with a mental health condition, since the comparison between groups is not tested. What is expert opinion: transposing this finding to another country, with a different market, a different supply and a different prevention policy.
One last point is worth stating explicitly. This study is about the prevalence of use. It says nothing about what cannabis does to those who use it: not about the course of a psychotic disorder, not about treatment response, not about the risk of moving on to problematic use. Finding that the frequency of use did not rise more in one group is in no way an argument that cannabis is harmless.
The colleague test
What an experienced colleague would say after a two-minute summary of this study, between two appointments.
“Useful for the debate, useless for my clinic tomorrow. The twofold gap was already there, and it is people without a condition who caught up, not my patients who pulled away. In anxious and depressed patients, the rise is real and it holds for five years. And it is self-report from an online panel: the patients I worry about most are not in any panel.”
What it means in practice: nothing changes in the care of a patient who uses cannabis. This article serves to answer a health policy question with figures rather than intuition, and to avoid confusing a question about prevalence with a question about individual risk.
What you can do with this
- Have an evidence-based answer ready when legalization comes up in a consultation or a meeting: over five years of Canadian data, the rise in use was steeper among people reporting no mental health condition than among the others.
- Do not claim, however, that nothing changed among people with a mental health condition. Among those reporting an anxiety disorder or depression, past 12-month use stayed above its 2018 level every year through 2023.
- Do not turn this finding into an argument that cannabis is harmless. The study measures how often people use it, not an effect of cannabis on the course of a psychiatric condition.
- Keep in mind that the twofold gap predates any reform. It is a reason to screen systematically for cannabis use in patients under care, whatever the legal framework.
- Remember that the most severely ill patients are the ones this type of survey captures least well. An absence of signal in these subgroups does not carry the same weight as in the others.
Frequently asked questions
Do these results apply to France?
Not directly. Recreational cannabis is not legal there, and the market, supply and prevention policies differ. These data inform a debate, they do not predict what a reform would produce in another country.
So is legalization safe for vulnerable people?
This study cannot say. It observes that reported frequency of use rose less among them than in the rest of the population. It measures neither the clinical consequences of that use nor what happens among people the survey does not reach.
Did use rise among people with anxiety or depression?
Yes, and the point is often lost in summaries. Past 12-month use rises from the first year after legalization, adjusted odds ratio 1.33 (95% CI 1.15 to 1.53) for anxiety disorder and 1.47 (1.26 to 1.73) for depression, and it remains significantly above the 2018 level every year through 2023. What is true is that the rise was even steeper among people reporting no condition.
What about bipolar disorder, post-traumatic stress disorder and schizophrenia?
No significant change from 2018 is found, on either measure of use. This result should be read for what it is: analyses that were performed and proved inconclusive, in the smallest samples of the study, with very wide confidence intervals and no reported power calculation. It is not evidence that there is no effect.
Why does the twofold gap exist?
The study observes it, it does not explain it. Several hypotheses coexist in the literature, from self-medication of symptoms to shared risk factors. A cross-sectional design cannot tell them apart, nor establish the direction of causality.
Does the declared conflict of interest call the results into question?
It does not invalidate them. It is declared, funding is public, and the tie concerns expert work against the tobacco, vaping and cannabis industries, not on their behalf. It warrants attention to how the conclusions are worded, not rejection of the article.
Annotated bibliography
Source study. Rundle S, Iraniparast M, Hammond D. The prevalence of cannabis use pre-versus post-cannabis legalization in Canada by mental health status: findings from national repeat cross-sectional surveys. The Lancet Regional Health – Americas, 2026, volume 55, article number 101373. DOI 10.1016/j.lana.2026.101373. Open access under a Creative Commons license.
Supplementary material of the publication. Supplementary table 1 brings together the main effects of each condition on past 12-month use and on daily use, in 91,907 respondents. It is the source of the odds ratios from 1.55 to 2.41 cited above, and it was consulted. Supplementary table 2, on the sensitivity analysis using lifetime diagnosis and covering 91,794 respondents, was not legible in the document consulted: its values are therefore not reported here, and the authors’ statement that the two measures give highly consistent results could not be verified.
Earlier analysis of the same survey series, cited by the authors. Sorkhou M, Johnstone S, Weinberger AH, et al. Changes in cannabis use patterns in psychiatric populations pre-and post-legalization of recreational cannabis use in Canada: a repeated cross-sectional survey. Cannabis, 2024, volume 7, issue 3, pages 1 to 13. DOI 10.26828/cannabis/2024/000238. Based on the first three waves, it reported an increase in daily use among people with schizophrenia. The authors of the present study attribute this discrepancy to a denominator artifact: the result rested on the relative share of daily use among users, driven by a fall in non-daily use. Reference taken from the bibliography of the source article, not independently verified.