Development of the crude (A) (adjusted for age) and adjusted (B) (adjusted for alcohol use disorder, other psychiatric disorders, parental schizophrenia, parental other psychiatric disorder, parental alcohol or substance use disorder, parental educational attainment, sex, and age) hazard ratios (HRs) between cannabis use disorder and schizophrenia in Denmark and of the incidence of schizophrenia (C) and cannabis use disorder (D) in the same period. Shaded areas indicate 95% CIs.
Shaded areas indicate 95% CIs.
The lower bound of the shaded area is the E-value for the lower bound of the 95% CI for the hazard ratio used to estimate the E-value.
eTable. All Parameter Estimates From the Regression Models for 1972, 2000, and 2016
Customize your JAMA Network experience by selecting one or more topics from the list below.
Hjorthøj C, Posselt CM, Nordentoft M. Development Over Time of the Population-Attributable Risk Fraction for Cannabis Use Disorder in Schizophrenia in Denmark. JAMA Psychiatry. 2021;78(9):1013–1019. doi:10.1001/jamapsychiatry.2021.1471
Has the population-attributable risk fraction for cannabis use disorder in schizophrenia increased over time, as would be expected with increasing use and potency of cannabis?
In this Danish nationwide, register-based cohort study, the population-attributable risk fraction for cannabis use disorder in schizophrenia increased from approximately 2% in the period to 1995 to approximately 6% to 8% since 2010.
These findings may indicate that cannabis use disorders are associated with an increase in the proportion of cases of schizophrenia.
Cannabis use and potency of cannabis have increased during the past 2 decades. If the association between cannabis use and schizophrenia is causal, this should be reflected in an increase in the proportion of cases of schizophrenia being attributable to cannabis, the population-attributable risk fraction (PARF).
To determine whether the PARF for cannabis use disorder in schizophrenia has increased over time.
Design, Setting, and Participants
This nationwide, register-based historical prospective cohort study included all people in Denmark born before December 31, 2000, who were alive and 16 years or older at some point from January 1, 1972, to December 31, 2016. Data analysis was performed from August 2020 to April 2021.
Diagnosis of cannabis use disorder.
Main Outcomes and Measures
Diagnosis of schizophrenia, with estimated PARF of cannabis use disorder in schizophrenia from 1972 to 2016.
A total of 7 186 834 individuals were included in the analysis, including 3 595 910 women (50.0%) and 3 590 924 men (50.0%). The adjusted hazard ratio for schizophrenia fluctuated at approximately 4 (with 95% CIs ranging from approximately 3 to 6) throughout most of the study period when people diagnosed with cannabis use disorder were compared with those without cannabis use disorder. The PARF of cannabis use disorder in schizophrenia also fluctuated, but with clear evidence of an increase from 1995 (when the PARF was relatively stable around 2.0%, with a 95% CI of approximately 0.3% to either side) until reaching some stability around 6.0% to 8.0% (with a 95% CI of approximately 0.5% to either side) since 2010.
Conclusions and Relevance
The results from these longitudinal analyses show the proportion of cases of schizophrenia associated with cannabis use disorder has increased 3- to 4-fold during the past 2 decades, which is expected given previously described increases in the use and potency of cannabis. This finding has important ramifications regarding legalization and control of use of cannabis.
Several studies and meta-analyses1,2 have indicated that the risk of schizophrenia and related psychoses is increased for people who use cannabis. The association appears to be particularly driven by heavy and frequent cannabis use.2,3 This has led to a persisting hypothesis that cannabis may be a component cause of schizophrenia, at least in some people. This hypothesis is strengthened by several factors, such as strong attempts to control for reverse causation (eg, people self-medicating prodromal symptoms of schizophrenia). Similarly, cannabis-induced psychosis has been found to be associated with later onset of schizophrenia.4-6 Furthermore, experimental evidence strongly suggests that cannabis, even in relatively low doses, causes psychoticlike symptoms in healthy individuals.7-9
Despite this, some debate persists regarding whether the association between cannabis use and schizophrenia is truly causal. Epidemiological studies, although able to provide strong evidence, can never completely exclude the possibility of residual confounding. One argument against the causal hypothesis has been that, with increasing use and potency of cannabis in the population, one should have observed an increase in the incidence of schizophrenia, which has been postulated to be absent.10-12 A long-standing dogma states that the incidence of schizophrenia has remained stable over time, but Kühl et al13 recently showed that the incidence of schizophrenia has steadily increased in Denmark from 2000 to 2012, especially in younger groups. This study did not, however, make any claims that this should have been due to increasing use or potency of cannabis.13
Denmark has the benefit of having unselected nationwide Danish registers that allow for large-scale studies using the entire population. Using these registers, Nielson et al14 showed that cannabis use disorder, an indicator of regular use of cannabis, is associated with schizophrenia. Both cannabis use and cannabis use disorder have indeed been increasing over time in many countries, such as Denmark.12 The same is true for the potency of cannabis products available on the market.10,11 If all other risk factors for schizophrenia remained stable in the same period, this should then lead to an increase in the population-attributable risk fraction (PARF) of cannabis use disorder in schizophrenia. The PARF estimates the proportion of cases with schizophrenia that would have been prevented if nobody had cannabis use disorder. We thus aimed to investigate whether the PARF for cannabis use disorder in schizophrenia has increased over time in the Danish population.
We used the nationwide Danish registers, full linkage of which is made possible through the personal identification number issued to all people born in Denmark or obtaining legal residence in Denmark since 1968.15 We included all people born before December 31, 2000, who were alive and 16 years or older at some point from January 1, 1972, to December 31, 2016, ensuring that all people would be able to turn at least 16 years of age during follow-up. Purely register-based studies do not require ethics approval under Danish law. This study was approved by the Danish Data Protection Agency, and analyses were conducted on servers at Statistics Denmark using encrypted personal identification numbers, making identification of single individuals impossible. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.
Information on both cannabis use disorder and schizophrenia was obtained from the Psychiatric Central Research Register, in which all inpatient treatment at psychiatric departments has been registered since 1969, as well as outpatient treatment at psychiatric departments since 1995.16 In the case of cannabis use disorder, this was supplemented with data from the National Patient Register, which contains data on inpatient treatment at general (nonpsychiatric) departments since 1977 and outpatient treatment since 1995.17 Cannabis use disorder was defined as International Classification of Diseases, Eighth Revision (ICD-8) diagnostic code 304.5 and International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) diagnostic code F12.X. Schizophrenia was defined as ICD-8 diagnostic code 295.X (except 295.7) and ICD-10 diagnostic code F20.X.
The following potential confounders were included in the analyses: alcohol use disorder (ICD-8 codes 291, 303, and 571.0; ICD-10 codes F10.x, E24.4, E52, G31.2, G62.1, G72.1, K29.2, K70, K86.0, O35.4, Y57.3, Z50.2, Z71.4, and Z72.1) and other substance use disorder (ICD-8 codes 304.x except 304.5; ICD-10 codes F1X.X except F10.X, F12.X, and F17.X), both as time-varying covariates identified in the Psychiatric Central Research Register and the National Patient Register; sex; and other psychiatric diagnoses except those pertaining to substance use disorder or schizophrenia (ICD-8 codes 290-315; ICD-10 codes FX, except those previously used) as time-varying covariates identified in the Psychiatric Central Research Register. We also included parental psychiatric history and parental history of alcohol, cannabis, and substance use disorder, using the same diagnostic codes as described previously. Analyses were further adjusted for sex, age at the beginning of each year, whether a person was born outside of Denmark, and the parents’ highest level of education. In cases of missing data, we included a unique value in the analyses indicating missingness.
The data analysis was performed from August 2020 to April 2021. The PARF is calculated by the following formula: PARF = pd([RR − 1]/RR), where pd refers to the proportion of exposed participants among cases (ie, the proportion of those who develop schizophrenia who were diagnosed with cannabis use disorder), and RR is the rate ratio, which in this study was estimated using the hazard ratio following a Cox proportional hazards regression model with time-varying covariates. We applied this formula for the PARF because it is internally valid when confounding exists and adjusted rate ratios must be used.18 We used the Greenland method for estimating 95% CIs of the PARF.19
We estimated the PARF for each year from 1972 to 2016 by using delayed entry into the Cox proportional hazards regression model on January 1 of a given year, and censoring individuals on December 31 of the same year if they had not either developed incident schizophrenia or been censored owing to death or emigration. People were entered into the analyses no earlier than their 16th birthday. For each year under investigation, we considered a person positive for cannabis use disorder only if at least 1 diagnosis thereof had been given no more than 3 years before the start of that year. The proportion of individuals with diagnosed cannabis use disorder in a given year was evaluated using this time-varying covariate. Age of the individual was used as the underlying timescale, ensuring that analyses were adjusted for age.
We also calculated the E-value for the estimate and the lower limit of the 95% CI each year in the same period. The E-value is a measure of the strength of unmeasured confounders on both cannabis use disorder and schizophrenia that would need to exist to completely explain the observed association.20
We conducted sensitivity analyses in which we censored individuals as they turned 50 years of age, to better reflect the population most at risk of both cannabis use disorder and incident schizophrenia. We conducted further sensitivity analyses, expanding the time frame for positive cannabis use disorder to 7 years and 1 sensitivity analysis without a time limit of exposure. In a final set of sensitivity analyses, we did not adjust for other psychiatric disorders, which may be a case of overadjustment, for instance because other psychiatric disorders might rather be considered mediators. All analyses were conducted in Stata/MP, version 16.1 (StataCorp LLC).
We included 7 186 834 individuals in at least 1 of the yearly PARF analyses (3 595 910 women [50.0%] and 3 590 924 men [50.0%]). Additional characteristics of the study population are provided in the Table.
The association between cannabis use disorder and schizophrenia over time is depicted in Figure 1. Figure 1A and B show the hazard ratios, which, unadjusted, decreased from more than 100 to reaching some stability at approximately 50 or even 30. The adjusted hazard ratio fluctuated over time, but with no clear trend, and generally centered around an adjusted hazard ratio of approximately 4 (with 95% CIs ranging from approximately 3 to 6). The association remained statistically significant throughout the period, as indicated by the 95% CIs. The 95% CIs tended to narrow after 1995. This substantial decrease in the hazard ratios is likely a case of overadjustment, because most of the reduction occurred when adjusting for other psychiatric disorders. This would indicate that the fully adjusted hazard ratios may actually be artificially low. However, we opted to use these because conservative results, biased toward the null hypothesis, are preferable owing to our desire to avoid false-positive findings. Figure 1C and D show the development over the period of incident schizophrenia and of recently diagnosed (within 3 years) cannabis use disorder. The latter increased almost linearly throughout the period, whereas the incidence of schizophrenia showed a generally increasing tendency, albeit with some fluctuations. Parameter estimates for all variables are shown in the eTable in the Supplement for the years 1972, 2000, and 2016.
The development of the PARF over time is depicted in Figure 2. Despite some fluctuation, a general increase over time is evident, starting first around 1995 (when the PARF was relatively stable at approximately 2.0% with a 95% CI of approximately 0.3% to either side) until reaching some stability around 6.0% to 8.0% (with a 95% CI of approximately 0.5% to either side) since 2010. Figure 3 shows the E-value over time, indicating that an unmeasured confounder would need to be associated with both cannabis use disorder and schizophrenia at a relative risk of well above 5, and sometimes even above 10, beyond the other variables already included in the adjusted analyses.
Figure 4 shows results from the 4 sensitivity analyses. The results were largely the same, with the same apparent increase being evident, albeit with slightly higher PARFs.
An increase in the PARF over time for cannabis use disorder in schizophrenia was clearly observed. The PARF increased from approximately 4.0% around the turn of the millennium to 8.0% since 2010, albeit with some fluctuations, especially in 2006, which appears to be an unexplained anomaly, because ignoring 2006 would yield a regular linear trend. A further increase was observed beginning already in the mid-1990s, until which point it had been relatively stable at around 2.0%. This was consistent with our hypothesis. Because the PARF was estimated under the assumption of potential causality, maintaining this assumption and increasing potency and frequency of use of cannabis use in recent years should yield a corresponding increase in PARF.10-12 Ours is, to our knowledge, the first study to demonstrate such an increase. Although not actual proof of a causal link, our results from longitudinal analyses of the associations over time between cannabis use disorder and schizophrenia provide evidence that some of this association may be causal.21 This evidence might also help explain the general increase in the incidence of schizophrenia that has been observed in recent years.13 In total, our study thus provides some support for a potential causal inference that the long-observed association between cannabis and schizophrenia is likely partially causal in nature.1,2
The PARF is an estimate of the proportion of cases of schizophrenia that would have been prevented if no individuals had been exposed (in this case to cannabis use disorder), under the assumption that the association between cannabis and schizophrenia may be causal. In light of this assumption, the PARF cannot confirm whether an association is truly causal. However, an increase in the PARF co-occurring alongside an increase in either the proportion using cannabis or in the potency of cannabis would be expected if the association was indeed causal. Previous data suggest an increase in the use and problematic use of cannabis in Denmark.12 Moreover, cannabis on the Danish market is noted to be among the most potent variants in Europe and has increased from approximately 13% Δ9-tetrahydrocannabinol in 2006 to nearly 30% in 2016.10 The fact that the hazard ratio did not increase over time was unexpected, because this would have been expected if the association between cannabis use disorder and schizophrenia was primarily driven by high-potency cannabis, as has previously been suggested.2 This could either indicate that potency of cannabis in Denmark has long been above the threshold for psychotogenic effects, or that cannabis use disorder is itself an indicator of severe exposure. Furthermore, we estimated the E-value over time, which is a measure of how strong an unmeasured confounding effect would be required to conclude that cannabis use disorder was not causally associated with schizophrenia. With some fluctuations, a requirement for such an unmeasured confounder would be association with both cannabis use disorder at a relative risk of above 5, often approximately 10, and with schizophrenia at the same magnitude of relative risk, after adjustment for all the covariates included in the present study (ie, alcohol or other substance use disorders, other psychiatric disorders, parental schizophrenia, parental other psychiatric disorders, parental alcohol or substance use disorder, parental educational attainment, sex, and age). Although some unmeasured confounding may exist, it is unlikely to have such strong associations as required by the E-value to explain the associations we observed in the present study. Use of tobacco has been suggested as such a potential risk factor for schizophrenia22-24; first, some evidence actually suggests that cannabis use is the confounder of this association and not the other way around23-25; and second, tobacco use is unlikely to yield relative risks from 5 to 10 with schizophrenia, although it may do so for cannabis use disorder.22 Another often-mentioned risk factor for schizophrenia is urbanicity24; we did not include urbanicity in our study because (1) urbanicity may be considered a mediator, in which people at risk of using cannabis and other substances gravitate toward urban centers, and (2) the variable was not available for the full study period. We did not have information on childhood trauma, but the associations between childhood trauma and either cannabis or schizophrenia are generally lower than our estimated E-value, and previous studies have indicated at least partial independence between the 2 risk factors.26-28 Other potential risk factors for schizophrenia, such as season of birth, vitamin D, and left-handedness, are likely not strongly associated with cannabis use disorder.24,29-31 Genetic liability has been suggested as a partial confounder (ie, that the association between cannabis use and schizophrenia is partly causal and partly confounded by genes; eg, by risk genes for schizophrenia increasing the risk of cannabis use).32,33
To our knowledge, ours is the first study to investigate the contribution of time to the PARF of cannabis in schizophrenia. However, the multisite EU-GEI (European Network of National Schizophrenia Networks Studying Gene-Environment Interactions) study recently found that sites with high rates of use of high-potency cannabis or high rates of daily use of cannabis were associated with the incidence of schizophrenia.34 In that study, researchers found that the theoretical removal of high-potency cannabis would prevent 12% of cases of schizophrenia, and with PARFs as high as 30% in London and 50% in Amsterdam.
Although not in itself proof of causality, our study provides evidence of the theory of cannabis being a component cause of schizophrenia. Furthermore, our study challenges the often-cited argument against causality that an expected increase in cases of schizophrenia attributable to cannabis use has not been observed. Our study is particularly important with the increasing legalization of cannabis for both medicinal and recreational uses seeming to lead to an increase in the perception of cannabis as relatively harmless and possibly in the uptake of cannabis use, especially among youth.35,36 Although psychosis is not the only outcome of interest in terms of cannabis use, our study clearly indicates that cannabis should not be considered harmless. Cannabis on the Danish market is among the strongest in the European market, which may compound the problem, because previous studies2,10,11 have shown that the cannabis-psychosis link may be most strongly relevant for high-potency cannabis. An important implication of our study may thus be that both use of cannabis and the potency of cannabis available should be reduced to achieve primary prevention of schizophrenia.
Our study holds both important strengths and limitations. Regarding strengths, we were able to use the unselected nationwide Danish registers to estimate annual associations between cannabis use disorder and schizophrenia, without the risk of selection biases related to, for instance, nonparticipation or dropouts. Furthermore, we were able to adjust for a range of potential confounders that may have changed their effect with the association between cannabis use disorder and schizophrenia over time. It is a further strength that we quantified the risk of unmeasured confounding through the use of the E-value. Finally, we conducted sensitivity analyses, revealing that our results were robust to choices regarding the analytical procedure.
A few limitations also warrant mention, however. Use of register-based information allowed us to assess only cannabis use disorder and not cannabis use per se. However, it has previously been suggested that the association between cannabis and schizophrenia is driven by high-potency cannabis and problematic use of cannabis.2 Consequently, this may not have a dramatic effect on our results. If anything, it would mean that some cases of schizophrenia were erroneously classified as unexposed to cannabis, making our results slightly conservative. If the probability of being erroneously classified as not having a cannabis use disorder has decreased over time, this might explain a small part of the increase in the PARF. A similar limitation exists in that we were only able to assess cannabis use disorder in patients who had contact with the treatment system. Again, this is likely to have yielded slightly conservative results, and without a clear effect on the development of the PARF. A third and related limitation is that we do not have information on the types or potency of cannabis use by the study population on an individual level.
In this Danish, register-based cohort study, an increase in the proportion of cases of schizophrenia that may be attributable to cannabis use disorder to a relatively stable level of 8.0% since 2010 was observed. This increase is in concordance with what would be expected, given observed increases in use and potency of cannabis. These results from longitudinal analyses of the associations over time between cannabis use disorder and schizophrenia lend further support to the hypothesis that cannabis may be involved in the etiology of schizophrenia.
Accepted for Publication: April 15, 2021.
Published Online: July 21, 2021. doi:10.1001/jamapsychiatry.2021.1471
Corresponding Author: Carsten Hjorthøj, PhD, Copenhagen Research Center for Mental Health–CORE, Mental Health Center Copenhagen, Gentofte Hospitalsvej 15, Fourth Floor, DK-2900 Hellerup, Denmark (firstname.lastname@example.org).
Author Contributions: Dr Hjorthøj had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Concept and design: Hjorthøj, Posselt.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Hjorthøj, Posselt.
Critical revision of the manuscript for important intellectual content: Nordentoft.
Statistical analysis: Hjorthøj, Posselt.
Obtained funding: Hjorthøj, Nordentoft.
Administrative, technical, or material support: Posselt.
Supervision: Posselt, Nordentoft.
Conflict of Interest Disclosures: None reported.
Funding/Support: This study was supported by a grant from Lundbeckfonden.
Role of the Funder/Sponsor: The sponsor had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.