In this nationwide, register-based cohort study, we found evidence of a stronger association between CUD and schizophrenia for males than for females, consistent with the results from a small clinical sample indicating that females experiencing CUD are at lower risk of developing psychosis than males (Arranz et al., 2015). Further, the stronger PARFs of CUD in schizophrenia for males than females consistently increased from 1972 to 2021. Under the assumption of causality, in 2021, approximately 15% of recent cases of schizophrenia among males would have been prevented in the absence of CUD, in contrast to 4% among females. For younger males, the proportion of preventable CUD-associated cases may be as high as 25% or even 30%. This increase in PARF is related to both increasing associations, likely caused by more potent cannabis, and increasing the prevalence of CUD with time.

The aHR for CUD on risk of schizophrenia were slightly higher for males than females, which might be misinterpreted to suggest an overall absence of strong sex-specific effects of cannabis, and instead indicate that the lower PARF for females is due to the fact that fewer females than males have CUD. Importantly, when subdividing the sample into specific age groups, a strong interaction effect among age, sex, and CUD on schizophrenia became evident. For 16–20-year-olds, the adjusted IRR for the association between CUD and schizophrenia was nearly twice as high for males than females; for 21–25-year-olds, the IRR was approximately 50% higher for males than females, whereas for those aged >26 years of age the IRRs was similar for males and females. Assuming causality, this suggests that young males compared to females of the same age might be more susceptible to the psychotogenic effects of cannabis on schizophrenia. Further research is needed to examine potential differences in THC concentration of exposures and frequency of cannabis consumption between young males and young females (Khan et al., 2013).

Previous studies have indicated that partial genetic confounding factors likely exist on the association between CUD and schizophrenia, i.e. genes shared between these conditions may account for some but not all of the association (Gillespie & Kendler, 2020). However, genetic confounding factors would be unlikely to explain the steeper increases in the PARFs of CUD on schizophrenia that we identified for males than females, as changes in the genetic risk-profile of an entire population would have to occur over generations. Both preclinical and clinical studies have provided evidence of significant differences between the sexes in response to the acute and long-term effects of cannabis (Cooper & Craft, 2018). We were able to adjust for alcohol and other specific drug use disorders in our study, but not for tobacco use or tobacco use disorder, which has also been linked to psychosis in some (Gurillo, Jauhar, Murray, & MacCabe, 2015) but not all studies (Fergusson, Hall, Boden, & Horwood, 2015). Thus, future research is needed to investigate the mechanisms underlying the higher vulnerability of young males to the effects of cannabis on schizophrenia than that of young females.

This study adds to the evidence suggesting a relationship between intense use of cannabis and risk of developing schizophrenia (Marconi et al., 2016; Urits et al., 2020; Volkow et al., 2016). At the individual level, this increased risk occurs in both sexes, but especially appears higher in young males. At the population level, this translates to CUD being a major modifiable risk factor for schizophrenia, particularly among males. Notably, an increasing proportion of cases with schizophrenia may be avertible by preventing CUD, and this increase is likely linked to the increase in THC concentration in cannabis as has been observed in confiscated samples in Denmark (Freeman et al., 2019; Thomsen et al., 2019). This apparent of schizophrenia conferred by CUD, in combination with observations that cannabis use among youth is associated with impaired cognition and reduced academic achievement (Lorenzetti, Hoch, & Hall, 2020), highlights the need to prevent cannabis use among youth and young adults. Interestingly, it has previously been shown that the association between cannabis and schizophrenia may be bidirectional (Ferdinand et al., 2005; Petersen, Toftdahl, Nordentoft, & Hjorthøj, 2019), and further investigation of the reverse association, schizophrenia being a risk factor for future cannabis use, by sex and over time warrants further study.

Our study has several strengths. The use of nationwide registers largely removes the risk of selection bias, since consent was not required for study sample participation. Furthermore, this data source reduces information bias to a certain degree, as a large range of information is available in the registers. Moreover, the registers are free from missing data in the traditional sense. Our results are likely highly generalizable to populations exposed to the same types of cannabis as are available on the Danish market. Finally, the national register-based nature of the longitudinal data over 5 decades allowed us to study nearly six million people, providing highly robust risk estimates.

This study also has certain limitations. The register-based nature of the study means that we only have information on both diagnosed CUD and diagnosed schizophrenia. This bias, however, is likely to be toward the null hypothesis, thus indicating that our results may be conservative, and PARFs of CUD on schizophrenia may be underestimated. Moreover, unmeasured and residual confounding factors likely exist, as the registers do not provide information on potentially important items such as frequency and amount of cannabis used, age of first use, or the THC content of cannabis products. Furthermore, we did not have access to genetic information, but as mentioned above, genetic confounding factors are unlikely to account for the observed differences. Finally, although the observational nature of this study does not directly allow for causal inference and we cannot be certain of the proportion of exposed individuals who might have developed schizophrenia even in the absence of CUD, it is unlikely that all of the associations between CUD and schizophrenia would be explained by confounding factors [e.g. tobacco use disorder (Fergusson et al., 2015; Gurillo et al., 2015)].

A further limitation is that the contents of the registers change over time. For instance, prior to 1995, outpatient psychiatric contacts were not included in the registers. Based on the results of our joinpoint regression analyses, there was no significant joinpoint identified in 1995 or years after 1995. It is thus unlikely to influence the increase in PARFs which only occurred later. We adjusted our models for other psychiatric disorders, which was the driver of the dramatic reduction in the magnitude of the HR compared to the unadjusted models. This is likely a case of over-adjustment, as some of these other psychiatric disorders might well be intermediate diagnoses between CUD and schizophrenia, and thus act as mediators rather than confounders. Consequently, our estimated aHR are likely highly conservative. Finally, we decided not to adjust for socioeconomic status, as this is more likely to be a product of CUD, other alcohol or substance use disorders, or schizophrenia, and thus not a potential confounder. By contrast, one of the salient factors is having a family history of schizophrenia. We controlled for the parental history of schizophrenia in our analyses. Moreover, we adjusted for age, sex, AUD, other substance use disorders, other psychiatric disorders, and parental history of CUD, AUD, and other psychiatric disorders, which are highly associated with schizophrenia.

In conclusion, this study finds strong evidence of an association between CUD and schizophrenia among both males and females, and the magnitude of this association appears to be consistently larger among males than females, especially among those aged 16–25. Importantly, 15% of cases of schizophrenia in males may be preventable if CUD was avoided. Although CUD is not responsible for most schizophrenia cases in Denmark, it appears to contribute to a non-negligible and steadily increasing proportion over the past five decades. In young males (21–30 years, possibly up to 40), the proportion may even be as high as 25–30%. There are global increases in legalization of nonmedical use of cannabis, increases in THC content of cannabis and in total THC doses consumed (Caulkins, Pardo, & Kilmer, 2020), increases in the prevalence of cannabis use and CUD, and decrease in public perception of harm from cannabis use (Chiu, Hall, Chan, Hides, & Leung, 2022). Alongside the increasing evidence that CUD is a modifiable risk factor for schizophrenia, our findings underscore the importance of evidence-based strategies to regulate cannabis use and to effectively prevent, screen for, and treat CUD as well as schizophrenia.