A 2026 modeling study estimates roughly 28,000 to 30,000 premature deaths annually in Canada may be attributable to recreational cannabis use, but the figure is built from observational associations and significant modeling assumptions rather than confirmed causal data. Both the mortality estimate and a companion analysis of cannabis farmland reallocation raise important policy questions while carrying uncertainty that far exceeds what the headline numbers suggest.

Canada’s 2018 legalization of recreational cannabis represents one of the most closely watched natural experiments in drug policy history, and the question of whether it generates measurable, quantifiable downstream health costs is one that clinicians, patients, and policymakers alike are tracking closely. As cannabis use rates climb in the post-legalization era and evidence accumulates linking frequent use to cardiovascular, neurocognitive, and other outcomes, any attempt to systematically quantify these harms and frame them against tangible resource trade-offs carries real policy urgency. Whether the specific numbers in this paper hold up is a separate question from whether the questions it asks deserve answers.

Since Canada’s Cannabis Act took effect in October 2018, national cannabis use has risen from approximately 14.8% to roughly 26% of the population, with notable increases in edible consumption, smoking, and co-use with alcohol. Against this backdrop, Jamil and Pearce constructed a policy analysis synthesizing published epidemiological risk estimates for cardiovascular disease, dementia, lung cancer, colon cancer, motor vehicle accidents, trauma, suicide, and opioid poisoning, applying population-attributable fraction calculations to Canada’s post-legalization population. In parallel, they modeled an agricultural counterfactual: what would happen if the more than two million square meters of land currently devoted to recreational cannabis cultivation were instead used to grow nutrient-dense food crops.

The paper’s headline finding estimates roughly 28,000 to 30,000 premature deaths annually attributable to cannabis across the modeled health outcomes, with cardiovascular disease and dementia representing the largest shares. The dementia figure draws on a large Canadian cohort study of over six million individuals, which found substantially elevated dementia risk in those with cannabis-related acute care encounters. The land-use analysis concludes that redirecting cannabis farmland to food production could nourish over 3,600 people annually. Critically, all mortality estimates are derived from observational associations without formal uncertainty quantification, sensitivity analyses, or correction for overlapping populations across health categories. The authors themselves acknowledge that prospective, post-legalization cohort studies with refined exposure measurement are needed to validate these projections.

28,000 Deaths a Year? Parsing the Evidence Behind Canada’s Cannabis Mortality Estimate

A new peer-reviewed paper claims that recreational cannabis use in Canada is linked to approximately 28,000 to 30,000 premature deaths every year. That number is striking enough to demand attention and careful enough scrutiny to resist being taken at face value. On its own terms, this paper does something genuinely useful: it forces a conversation about how we account for the downstream costs of a policy decision that much of the public assumes was cost-free. The cardiovascular risk signals associated with cannabis are real and growing. The dementia data emerging from large Canadian cohorts are concerning and, for most of the public, almost entirely unknown. The pediatric poisoning spike after edible commercialization is measurable and troubling. These are not fabrications. They are legitimate public health findings that clinicians should understand and communicate honestly. Where the paper overreaches is in its aggregation. The headline mortality figure is produced by summing population-attributable fractions drawn from heterogeneous observational studies, each with its own confounders and limitations, without correcting for the simple fact that many of the same individuals appear across multiple risk categories. It is like adding up everyone who drove on a rainy day, everyone who drank coffee, and everyone who worked night shifts, noting that each group has slightly higher rates of certain illnesses, and then claiming you have counted all the deaths caused by those habits without checking whether many of the same people appear in multiple lists. The absence of sensitivity analysis, confidence intervals, or uncertainty propagation means that the final number cannot be evaluated as a precise estimate. It is a rhetorical anchor, not a statistical finding.

I am also concerned about a specific generalization error embedded in the dementia analysis. The elevated dementia risk the paper applies to the broader cannabis-using population is derived from a subgroup that required acute care for cannabis-related events. Using the accident rate of people who have already been hospitalized after car crashes to estimate the accident risk of everyone who drives would overstate the danger for the average driver. The people in the hospital are not representative of everyone on the road. Equally important is what the paper leaves out entirely. There is no accounting for the health benefits cannabis legalization may have produced: reductions in opioid use, harm reduction from regulated supply, therapeutic applications, or the documented costs of the criminalization that legalization was designed to replace. A true sustainability analysis requires both sides of the ledger. Without that balance, the paper functions more as advocacy than as comprehensive policy evaluation.

When a modeling paper converts observational associations into aggregate mortality counts and presents them as policy-relevant facts, the critical question is not whether the model is internally consistent but how much the conclusion depends on the choice of inputs, the direction of unmeasured confounding, and what was left out of the accounting. Here, the answer to all three questions is: substantially. The legitimate core of this paper, that cannabis legalization creates identifiable public health costs and land-use trade-offs that merit rigorous, ongoing policy accounting, is worth amplifying. The specific mortality figure it produces is not. What Canada’s legalization experiment genuinely needs is not a louder alarm built on shaky arithmetic, but a rigorous, prospective, benefit-and-risk accounting that can withstand methodological scrutiny and earn the trust it will need to actually change policy.

This paper sits near the beginning of the research arc for quantifying population-level health externalities of cannabis legalization, a space that currently has very few competitors in the published literature. It provides a structured framework that future investigators can refine, challenge, or build upon. However, for clinicians, the study does not produce usable risk estimates for individual patient counseling. The mortality figures are population-level modeling outputs, not individualized risk data, and the underlying associations come from studies with heterogeneous designs, populations, and levels of adjustment for confounders.

What clinicians can take from this work is the growing body of evidence that cannabis use, particularly heavy or frequent use, is associated with cardiovascular risk, cognitive outcomes in older adults, and pediatric safety concerns from edibles. These associations are supported by the source studies the paper cites, independent of its aggregation methods. From a practical standpoint, clinicians should be asking patients about cannabis use as routinely as they ask about alcohol or tobacco, particularly patients with existing cardiovascular disease, cognitive concerns, or household exposure to children. One actionable recommendation: when patients report cannabis use, document frequency, route of administration, and co-use with other substances, and flag cardiovascular and neurocognitive screening as indicated.

This is a health policy analysis and modeling study, not a primary epidemiological investigation, clinical trial, or systematic review. It occupies a lower tier in the evidence hierarchy because it collects no original data and depends entirely on the quality, representativeness, and correct application of secondary sources. The single most important inference constraint is that all mortality estimates are derived from observational associations, which means unmeasured confounding, reverse causation, and selection bias cannot be ruled out as alternative explanations for the reported risk elevations.

The cardiovascular and neurocognitive risk associations cited in this paper are consistent with a growing body of observational literature, including the large Canadian cohort study by Myran and colleagues (cited as reference 22 in the paper), which found substantially elevated dementia risk in individuals with cannabis-related acute care encounters. The post-legalization increases in cannabis use, edible consumption, and pediatric poisoning are well documented in Canadian surveillance data. What this paper adds is an attempt to aggregate and quantify these disparate signals into a single mortality estimate and to pair that estimate with a land-use opportunity cost analysis. No comparable published analysis has attempted this specific combination, making the paper novel in framing if not in its underlying data. The Fischer et al. baseline epidemiological estimates used for motor vehicle accident and lung cancer mortality attribution are over a decade old, and more recent data may tell a different story.

The most consequential analytic choice in this paper is the decision to sum population-attributable fractions across all health outcome categories without adjusting for overlapping populations. Because the same cannabis users may contribute to cardiovascular, dementia, cancer, and injury categories simultaneously, simple addition almost certainly inflates the aggregate mortality estimate. If the authors had applied competing-risk corrections or even a modest overlap discount, the headline figure could have decreased substantially, potentially by 30 to 50 percent or more, depending on the degree of correlated exposure. A formal sensitivity analysis or Monte Carlo simulation using plausible ranges for each input parameter would have provided readers with an uncertainty envelope rather than a single point estimate, and that envelope would likely have been wide enough to change the policy conversation from “cannabis kills 28,000 Canadians a year” to “cannabis-attributable mortality may plausibly range from several thousand to several tens of thousands.”

The most likely overinterpretation of this paper is treating the 28,000 to 30,000 annual deaths figure as a confirmed count of cannabis-caused fatalities in Canada. It is not. It is a modeled estimate constructed by applying observational associations to population-level data through a chain of assumptions, none of which has been validated against an independent mortality registry. Similarly, the dementia risk elevation should not be generalized to all cannabis users; it was derived from a subgroup who required acute hospital care for cannabis-related events, a population with inherently higher baseline risk. The land-use figure of 3,600 people nourished, while valid in its arithmetic, represents approximately 0.009% of Canada’s population and should not be read as a meaningful solution to food insecurity.

This paper contributes a structured framework for thinking about the public health externalities and land-use opportunity costs of recreational cannabis cultivation in Canada. It does not establish that cannabis causes 28,000 to 30,000 deaths annually, nor does it demonstrate that land reallocation would meaningfully address food insecurity. What it does establish is that the policy questions it raises are worth asking, and that the underlying risk signals, particularly for cardiovascular disease and dementia, deserve ongoing, rigorous investigation using prospective post-legalization data.

Does cannabis really kill 28,000 to 30,000 Canadians every year?

That figure is a modeled estimate, not a confirmed death count. It was produced by applying statistical associations from observational studies to population-level data through a series of assumptions. The actual number of deaths directly caused by cannabis is unknown. The figure should be understood as a rough projection with substantial uncertainty, not as an established fact.

Does cannabis cause dementia?

A large Canadian study found that people who required emergency or acute hospital care for cannabis-related events had a significantly elevated risk of later developing dementia. However, this finding comes from a high-risk subgroup, not the general population of cannabis users, and observational studies cannot confirm that cannabis directly causes dementia. The association is concerning and warrants further research, but a causal link has not been proven.

Should I stop using cannabis based on this study?

This paper alone is not sufficient grounds for changing your cannabis use. However, the broader evidence it draws upon, particularly regarding cardiovascular and cognitive risks, is growing and worth discussing with your physician. If you use cannabis regularly, especially if you have heart disease or cognitive concerns, bring it up at your next appointment so you and your doctor can weigh the risks and benefits specific to your situation.

Could the land used for cannabis really feed thousands of people?

The paper’s calculations are arithmetically sound: the more than two million square meters used for recreational cannabis could, in theory, produce enough calories to nourish over 3,600 people annually. However, this represents less than 0.01% of Canada’s population and does not account for the economic, regulatory, or practical barriers to converting cannabis-growing infrastructure to food production.

Have thoughts on this? Share it: