State Cannabis Legalization and Cannabis Use Disorder in the US Veterans Health Administration, 2005 to 2019

This observational study assesses the role of medical cannabis law and recreational cannabis law enactment in diagnosed cannabis use disorder prevalence among Veterans Health Administration patients by using yearly electronic medical record data from 2005 to 2019.

C annabis is a widely used psychoactive substance. 1 While many individuals can use cannabis without harm, 20% to 33% 2 of individuals who use cannabis develop cannabis use disorder (CUD), 3 which is characterized by problematic use, clinically significant distress or impairment, symptoms including tolerance, withdrawal, and neglect of other activities, 4 and psychosocial and health-related problems. [5][6][7][8] Adult CUD rates have increased in the US general population, 9,10 inpatients, 11 and veterans, 12,13 amidst declining perceptions of cannabis risk, 14,15 increasing cannabis potency, 16 and legalization. As of November 17, 2022, 37 US states had enacted medical cannabis laws (MCLs) and 21 states and Washington, DC, had enacted recreational cannabis laws (RCLs). MCLs and RCLs could increase rates of CUD by decreasing perceptions of cannabis risk and increasing availability and commercialization. 1,17,18 MCL and RCL have played a significant role in the increased prevalence of adult cannabis use, 19-21 but few studies examined the role of MCL or RCL in the national increases in adult CUD rates. In analyses of 3 US surveys conducted between 1991 and 2013, the prevalence of adult CUD increased more after MCL enactment than overall contemporaneous prevalence increases. 22 In 2008 to 2016 yearly national survey data, 17 among adults 26 years and older, the risk of adult CUD increased more after RCL enactment than before such enactment and more than overall national contemporaneous increases in prevalence. Neither CUD study used data after 2016, estimated the amount of overall change associated with the laws, or focused on a large health care population whose characteristics could increase vulnerability to effects of the changing laws.
The Veterans Health Administration (VHA) is the largest integrated health care system in the US, 23 now providing care to over 6 million patients each year. 24 The VHA patient population is predominantly male, has low income, 25,26 and is characterized by high rates of psychiatric disorders 27 and painful medical conditions incurred during military service. 26,28-30 Compared with the general population and with other veterans, these characteristics increase the risk for CUD, 31,32 making the VHA patient population a large, important group for studying the associations of MCLs and RCLs with outcomes. Additionally, many VHA patients are 65 years or older. 33 Understanding CUD in older adults is important to inform screening and service planning given the disproportionately increasing rates of cannabis use in older adults. 34,35 Therefore, leveraging the comprehensive VHA electronic health record, we investigated the role of MCLs and RCLs enactment in the national increases in rates of CUD diagnoses from 2005 to 2019, overall and by age group.

Statistical Analysis
Initial analyses of diagnosed CUD prevalence (hereafter, CUD prevalence) across 2005 to 2019 were grouped by state law status in 2019: (1) no CL, (2) MCL only, and (3) MCL/RCL. Adjusted prevalence estimates across each year in each of the 3 groups were obtained from a linear binomial regression model controlling for age, sex, race and ethnicity, and time-varying state covariates.
To estimate the role of MCL and RCL enactment in the national increases in CUD prevalence using all yearly information from 2005 to 2019, the staggered adoption difference-indifference (DiD) model 40 was used. This DiD model uses each state that enacts a law as its own control, comparing aggregated postlaw years to prelaw years and controlling for historical trends over time with data from all other states that had not enacted the respective law in contemporaneous years. A time-varying indicator was constructed for each state-year indicating no CL, MCL only, or MCL/RCL for that year. The DiD estimates for MCL only and MCL/RCL associations were obtained from fitting a linear binomial regression model with fixed effects for state, categorical year, time-varying law status, individual-level covariates, and time-varying state-level covariates. Resulting DiD estimates include the effect size of a state moving from no CLs to MCLs only and from MCLs only to MCLs/RCLs. (All states with RCLs previously had MCLs). The 17 no-CL states and 3 MCL-only states that did not change their laws between 2005 and 2019 contributed to DiD estimates by providing information for the contemporaneous trend estimates of the respective type of state. Note that the 2015 ICD-9-CM to ICD-10-CM change resulted in a slight downward shift in CUD prevalence in 2015 across the entire VHA system and all states. 13 DiD estimates take this shift into account by using states that had not yet passed MCLs or RCLs, which experienced the ICD-9-CM to ICD-10-CM change at the same time, as contemporaneous secular controls. While 95% CIs are provided for DiD estimates, given the large numbers and resulting precise estimates, interpretation focuses on magnitude rather than P values. To illustrate the magnitude of the DiD estimates compared with the overall increases in CUD prevalence (ie, the amount of change that could be attributed to the laws), the DiD estimates were divided by the absolute changes between 2005 and 2019 in the states with the respective laws by 2019. To explore whether law outcomes differed by earlier or later enactment, we obtained state-specific DiD estimates with interaction terms between state and timevarying no-CL/MCL-only/RCL status. All procedures were then redone within age groups (18-34, 35-64, 65-75 years), adjusting for within-group continuous age.
Sensitivity analyses used similar methods. We examined legalized dispensaries by replacing state/year no-CL/MCLonly/RCL variables with the year medical or recreational dispensaries were first operational. 36 We examined lagged MCL-only/RCL states effect sizes by replacing MCL-only/RCL state/year variables with 1-year postenactment dates. Analysis took place between February and December 2022.

DiD Estimates of the Role of MCLs and RCLs in the National Increases in CUD Rates
The DiD estimate of the CUD prevalence increase due to MCL enactment was 0.05% (95% CI, 0.05%-0.06%;  However, no state law association reached 1% absolute increase in CUD prevalence, and no patterning was evident by earlier or later enactment.

State-Specific DiD Estimates
eFigure 2 in Supplement 1 shows state-specific DiD results by age. In patients aged 18 to 34 years, 12 states showed an increase associated with law enactment, while in those aged 35 to 64 years, 16 states showed an increase associated with law enactment. In patients aged 65 to 75 years, CUD prevalence increase was associated with MCLs or RCLs occurred in 19 of the 30 states and Washington, DC. However, few increases that could be attributable to state-specific MCL-only or RCL enactment were greater than 1% in magnitude, and no patterning was evident by earlier or later enactment.

Discussion
We examined the association of state medical and recreational cannabis legalization (MCLs and RCLs) to diagnosed CUD prevalence in VHA patients between 2005 and 2019, a period of increasing CUD prevalence in the US adult population and in VHA patients. 13 DiD models that controlled for contemporaneous trends before and after MCL or RCL enactment in states with and without the respective laws suggested that CUD prevalence in VHA patients increased after MCL enactment by 0.05% more than would have occurred in the absence of MCLs and by an additional 0.12% more after RCL enactment than would have occurred in the absence of RCLs. These absolute increases represent the estimated effect sizes on CUD prevalence specifically associated with the enactment of the laws and imply that MCL and RCL accounted for 4.7% and 9.8%, respectively, of the 2005 to 2019 change in CUD prevalence. Considering states individually, a majority of MCL-enacting and RCL-enacting states had increases in CUD attributable with law enactments, although no state-specific increase attributable to law enactment reached 1%. By age, neither MCLs nor RCLs had a significant association with increases in CUD prevalence among those aged 18 to 34 years. While both MCL and RCL enactment were associated with CUD increases among patients aged 35 to 64 years, the largest effect size of RCLs was found in patients aged 65 to 75 years, in whom 8.1% of the increase in CUD prevalence could be attributed to MCL enactment, and 18.1% of the increase could be attributed to RCL enactment. Thus, in this national patient population, state cannabis legalization was followed by increases in CUD prevalence, but the increases attributable to the changing laws were relatively small compared with the overall increases in CUD prevalence. Two US general population studies showed greater increases in adult CUD prevalence after states enacted cannabis laws relative to no-CL states. 17,22 Between 1991 and 2013, 22 CUD prevalence increased from 1.48% to 3.10% after MCL enactment, a 0.70% greater increase than contemporaneous increases in no-CL states. Between 2008 and 2016, 17 CUD increased from 0.90% to 1.23% after RCL enactment among adults 26 years or older, a 0.33% greater percentage point increase than contemporaneous increases in no-CL states. Compared with the national survey results, VHA results are smaller (ie, 0.05% MCL; 0.12% MCL/RCL, summing to 0.17%, a global estimate of effect sizes associated with change from no CL to RCL) but similar to the general population studies in showing a significant although modest role of MCL and RCL in the national increases in CUD prevalence. The lack of RCL results among young adults in the general population study, 17 together with our null results in patients aged 18 to 34 years, suggests that MCLs and RCLs operate differently among younger and older individuals. Possible explanations for these age differences include that younger individuals may be less concerned about the legal status of cannabis than older individu-als, who may be more law-abiding or less risk-taking, or that younger individuals have readier access to illicit cannabis, making legalization less relevant to whether they use cannabis and subsequently, in a vulnerable subset, develop CUD.
In sensitivity analyses, 1-year lags did not meaningfully change results, nor did replacing RCL enactment year with the year recreational dispensaries became operational. However, replacing MCL enactment year with the year medical dispensaries became operational eliminated a result for MCL-only states. This suggests that MCL enactment may work through influences on perceived safety rather than through greater distribution and availability via medical dispensaries.
Factors that potentially minimized our estimates of the role of MCL and RCL in the national increases in CUD prevalence include a general diffusion of positive attitudes toward cannabis use, decreased harm perception, and increased use across the entire US adult population as more and more states legalized medical and recreational cannabis use. Perceived risk has decreased, 14,15 and despite inconclusive evidence, a majority of adults now see cannabis as beneficial to treat or prevent health problems. 41 The multibillion dollar cannabis industry, 42 seeking further expansion, must increase demand by generating new customers and/or by generating greater use among existing ones. 43 Websites of medical cannabis companies often imply product safety and efficacy, potentially leading policy makers and the public into believing unconfirmed claims. 44,45 Cannabis companies carefully design social media promotional profiles to attract customers 45 via content focusing on cannabis normalization, 46 and state regulations of cannabis advertising are often violated. 45,47,48 These industry activities could contribute to changing attitudes, increased cannabis use, and CUD among some users, both within and across state boundaries.
Additional possible mechanisms of the overall increases in CUD prevalence that should be examined in future studies include increasing rates of CUD risk factors, eg, pain 49,50 and psychiatric disorders, 51 and the increasing tetrahydrocannabinol (THC) potency of cannabis, 16 which increases addiction potential. 52,53 Highly potent cannabis products are increasingly popular in both medical and recreational cannabis markets. [54][55][56][57] Limitations Study limitations are noted. First, VHA patients are not representative of all veterans 25,30 or all adults. Second, the ICD CUD diagnoses were made by clinicians, not structured research assessments. VHA clinicians are most likely to diagnose severe disorders 58,59 and may miss the mild cases commonly found in general population surveys using structured assessment instruments. 9,31 While the actual number of missed VHA cases is unknown and may have varied over time, the overall 2019 VHA CUD prevalence (1.9%) 60 was higher than in National Survey on Drug Use and Health adults (1.7%), 61 despite the younger mean age of adult National Survey on Drug Use and Health participants. Thus, VHA findings serve as a useful counterpart to general population findings, providing information on what are likely to be severe cases in a clinical population in a national health care system that has many risk factors for CUD. Further, research diagnostic interviews with 3 to 4 million patients yearly are not feasible, so the VHA data provide a unique opportunity to examine the role of MCL and RCL in the national increases in CUD prevalence. Third, cannabis law provisions are heterogeneous. 62- 64 We examined whether states permitted dispensaries, but other differences (eg, possession limits, price, taxation) should be addressed in future studies. Fourth, state law effects may be delayed. We analyzed 1-year lags to include recently enacted RCLs; longer lags should be analyzed later. Fifth, demographic characteristics could influence vulnerability to MCLs or RCLs; studies should investigate these as modifiers of law effects. Sixth, a cannabis use measure was not available, so CUD within users and specific use patterns could not be examined. Seventh, while we controlled for many time-varying state-level confounders, others (eg, state opioid policies) could have affected results and should be addressed in future studies. Eighth, the DiD methodology estimates law effects in the states that enacted them and does not account for spillover effects into other states. If patients in no-CL states (the contemporaneous secular controls) are influenced by CLs in other states, estimated CLs effects will be biased toward the null. For example, border crossing to buy cannabis by patients in no-CL states living near a border with an MCL-only or RCL state 65 could have elevated CUD rates in some no-CL states, potentially mitigating the ability to find stronger MCL or RCL effects at the state level. Ninth, no study can be unequivocal about the causal nature of observed effects. However, our prepost DiD analysis that controlled for contemporaneous trends and many other state-level factors that might have changed over time provides stronger support for estimated effects of changing cannabis laws than studies without such design and analytic rigor. [66][67][68][69] These limitations are offset by notable strengths and novel aspects, including a large sample size, information about MCL/RCL associations in an important older age group about whom little is known, estimates of statespecific MCL and RCL results by the order (year) in which they were enacted, and a focus not only on significance but also on the magnitude of effect sizes.

Conclusions
In this study, cannabis did not have the same overdose/ mortality risk profile as opioids or stimulants. However, CUD, a diagnosable disorder with many associated problems, 5-8, 70 is prevalent among cannabis users (approximately 20%-33%), which is more than commonly assumed. 2 The US national increase in CUD diagnoses regardless of state laws underscores a growing need in the VHA and elsewhere to screen for cannabis use and offer evidence-based treatments for CUD. 71 Additionally, while VHA patients aged 65 to 75 years had the lowest CUD prevalence in this study, estimated RCL results were strongest in this group, suggesting a need to attend to potential CUD in older veterans.
To conclude, alcohol, tobacco, and prescription opioids have undergone major shifts in public acceptance or rejection across decades and generations. 72,73 Public health efforts regarding these substances have long competed with commercial interests. With cannabis increasingly legalized, similar competing public health and commercial interests are now emerging. To inform future health and policy efforts, researchers must monitor harms related to increasing CUD, identify whether subgroups show particular risk due to changing cannabis laws, and ensure that this knowledge is clearly communicated to policy makers, clinicians, and the public.