These two models were able to isolate the impacts of recreational marijuana legalization from the impacts of medical marijuana legalization on opioid prescriptions. Comparing eight states and DC to all the remaining 42 states violated the difference-in-difference assumption, because states without medical marijuana legalization had different trends in opioid prescriptions compared to states with medical marijuana legalization . To test the assumption of parallel trends in treatment and comparison states in the absence of policy change, we used repeated ANOVA to compare time trends in opioid outcomes between treatment and comparison states in years 2010, 2011, and 2012 when none of these states adopted recreational marijuana legalization. Linear regressions were used, controlling for time-varying state covariates, state indicators, year and quarter indicators, and state-specific linear time trends. State indicators accounted for time-invariant state-level unobserved heterogeneities such as social norms about opioid use. Year and quarter indicators accounted for time-specific heterogeneities common to all the states at the same time, such as CDC Guideline for Prescribing Opioids for Chronic Pain published in 2016 . State-specific linear time trends accounted for statelevel time-variant trends in outcomes. Standard errors in the regression were clustered at the state level. To test the robustness of results, we conducted a series of sensitivity tests. 1) Because California, Maine, Massachusetts, and Nevada had limited number of post-legalization observations, the policy impacts in these states may not be statistically discernable. In addition, the most recent quarters of Medicaid State Drug Utilization data may contain errors and are often subject to future revision. We excluded the observations after 3rd quarter of 2016 when these states implemented recreational marijuana legalization to focus the analyses on states legalizing recreational marijuana before 2016. 2) We moved Hydrocodone-combination drugs to Schedule III opioids to test results sensitivity to recent drug reclassifications. 3) It is suggested that adding state-specific time trends may attenuate estimates of policy impact if the policy impact acts upon the trend itself . We removed state-specific time trends in regressions and expected that the associations would be more discernable. 4) Following previous research ,ebb and flow tray we also performed falsification tests on 4 drug classes, including blood-thinning agents, phosphorous-stimulating agents, antivirals, and antibiotics, as there is no scientific evidence suggesting the associations between marijuana and the underlying conditions that these drugs treat.

These drugs were assumed to be not associated with recreational marijuana legalization but associated with common unmeasured confounding factors such as those affecting general prescribing, healthcare utilization, and healthcare resources at the state level. Table 2 reports the descriptive statistics of pooled data by recreational marijuana legalization status. Compared to six states with medical marijuana legalization but without recreational marijuana legalization, eight states and DC that legalized recreational marijuana in the study period had slightly and insignificantly higher rates of Schedule II and III opioid prescriptions. Supplemental Table S22 reports ANOVA tests for time trend comparisons, suggesting that the time trends prior to legalization did not significantly differ between the states legalizing recreational marijuana in 2012 and the six comparison states, between the states legalizing recreational marijuana in 2015 and the six comparison states, or between the states legalizing recreational marijuana in 2016/7 and the six comparison states. Figure 1 shows the unadjusted time trends in number of Schedule III opioid prescriptions by legalization status. States that legalized recreational marijuana in 2015 and 2016/7 saw a reduction in number of Schedule III opioid prescriptions after the legalization took effect, whereas states that legalized recreational marijuana in 2012 saw a slight increase. Trends in number of Schedule II opioid prescriptions did not appear to differ by legalization status . Based on difference-in-difference regressions, Figure 2 reports predicted percentage changes in number of opioid prescriptions associated with recreational marijuana legalization . In Model A that compared among eight states and DC with recreational marijuana legalization, recreational marijuana legalization was not associated with number of prescriptions, total doses, or spending of Schedule II opioids.In Model B that compared eight states and DC to six states with medical marijuana legalization, recreational marijuana legalization was not associated with any Schedule II or Schedule III opioid outcome. Using eight-year quarterly data on prescription opioids received by Medicaid enrollees in the US, the study added to the still limited literature about the impacts of recreational marijuana legalization on opioid use. It enhanced internal validity by adding comparison states and controlling for multiple confounders that were absent in previous research , such as presence of prescription drug monitoring program, Medicaid expansion.

It also enhanced generalizability by investigating all states legalizing recreational marijuana in the US. We found no evidence to support the concern that recreational marijuana legalization increased opioid prescriptions received by Medicaid enrollees. Instead, there was some evidence in some model specifications that the legalization might be associated with reduction in Schedule III opioids in states that implemented legalization in 2015 . It appeared that, if the hypotheses about marijuana’s substitution effect and gateway effect on opioid use are both valid, the gateway effect of marijuana did not outweigh its substitution effect. Another possibility is that the hypothesis about marijuana’s gateway effect lacks support. Unfortunately, we were not able to directly assess these mechanisms in this study. It is not clear why two comparisons yielded slightly different results. Both models have advantages and limitations. The treatment and comparison states in the first model comparing among eight states and DC were more comparable, as they all had adopted recreational marijuana legalization at some time points. On the other hand, the second model comparing eight states and DC to six states with medical marijuana legalization had a larger sample size to detect statistical significance. We therefore chose to report findings in both comparisons. Irrespective of their slight differences, the core findings from the two comparisons were consistent that recreational marijuana legalization did not increase prescription opioids received and most coefficients for the outcome variables were non-significant. In accordance with our previous study on medical marijuana legalization and prescription opioids received by Medicaid enrollees , the association between recreational marijuana legalization and reduction in prescription opioids seemed to be only evident in some models for Schedule III opioids but not for Schedule II opioids. Because this line of research only emerged recently, the explanation for the differential associations remains unknown. As discussed in our previous study , we hypothesized that such differences may be partly attributable to the differences in clinical practice and drug efficacy between the two drug classes. According to Controlled Schedule Schedules classified by US Drug Enforcement Administration, Schedule II opioids have greater potential for opioid misuse and overdose than Schedule III opioids . In clinical practice, Schedule II opioids must be refilled with monthly prescriptions whereas Schedule III opioids are fillable within six months without new prescriptions . Receiving regular monitoring and evaluations from physicians, patients prescribed with Schedule II may be less likely to switch to other drugs. Regarding drug efficacy, Schedule III opioids are often used to treat mild to moderate pain symptoms,rolling greenhouse benches for which marijuana is suggested to be also effective . But the evidence for marijuana’s efficacy to treat severe pain symptoms is still limited. Patients prescribed with Schedule II opioids might be less likely to receive recommendation from physicians to switch to marijuana. These hypotheses need future research on individual observations to provide empirical support. It is also worth noting that despite large effect size detected for Schedule III opioids in terms of percentage point reduction , the absolute level of opioid prescribing rates was low for this drug class . The impact of the legalization converted to absolute levels was modest. This study has limitations primarily related to data availability. First, we evaluated the implementation of legalization instead of commercialization . Because several states did not open retail markets during our study period, our results may be biased toward the null. Second, despite the size of state-level observations is larger in this study than previous research, our study sample is still small and some statistically non-significant associations may simply reflect the lack of statistical power. Particularly, observations in post legalization period were limited for states implementing legalization in 2016/7. Third, we were not able to explore why states implementing legalization at different time points may demonstrate differential changes in opioid prescriptions. Fourth, we grouped states based on their law implementation dates. However, the states implementing the legalization on the same dates may have opened their retail markets on different dates . We were not able to identify the degree of marijuana commercialization in each state or evaluate the independent impacts of commercialization because of limited sample size. Further, similar to other state-level investigations of aggregate data, we were not able to explore causal mechanisms of the findings at individual level.

Particularly, the hypotheses about substitution and gateway effects of marijuana cannot be directly tested. Additionally, the outcomes analyzed in this study represented opioid prescribing but not patients’ legitimate use or misuse of prescription opioids. Finally, the findings may not be generalizable to opioids dispensed in non-outpatient settings or to the general population. The findings represented a limited number of states in the US but may not be generalizable to other states in the US or to population in other countries. Alcohol and marijuana use are common in adolescence. In 2003, 31% of 12th graders reported getting drunk in the past month, 21% of 12th graders revealed using marijuana in the past month, and 6% of 12th graders disclosed daily marijuana use . Further, 40% of high school students who used marijuana in the past year met criteria for marijuana abuse or dependence . Moreover, 58% of adolescent drinkers also report marijuana use , and alcohol and marijuana use disorders are highly comorbid . Despite the prevalence of heavy alcohol and marijuana use in teenagers, it is unclear how such protracted use may affect brain functioning during youth, particularly as adolescent neuromaturation continues. Neuropsychological studies of teens with alcohol use disorders have reported decrements in language skills, problem solving, verbal and non-verbal retention, working memory, and visuospatial performance . In addition, we previously examined functional magnetic resonance imaging brain response during a spatial working memory task among teens with AUD and demographically similar non-abusing controls . Groups performed comparably on the task, but AUD teens demonstrated less brain response than controls in the midline precuneus/posterior cingulate, and more activation in bilateral posterior parietal cortex, suggesting subtle alcohol-related neural reorganization and compensation. These neuropsychological and imaging findings suggest that heavy alcohol use during youth adversely affects frontal and parietal circuitry, but the additional impact of marijuana use is less well understood. Neuropsychological assessments of substance use disordered teens have described marijuana use related deficits in learning and memory and attention . A longitudinal study of marijuana dependent adolescents demonstrated further short term memory decrements that persisted after 6 weeks of monitored abstinence . In addition, compared to individuals with adult-onset cannabis use disorder and non-abusing controls, adolescent-onset cannabis use disordered adults showed attenuated electrophysiological response during selective attention , as well as smaller frontal and parietal volumes and increased cerebral blood flow . These studies indicate that heavy marijuana use during youth may adversely affect cognition and brain functioning, particularly short-term memory and attention, and raise questions about the integrity of frontal and parietal brain regions in adolescents with marijuana use disorders. In order to understand the neural correlates of concomitant heavy marijuana and alcohol use during youth, we assessed blood oxygen level dependent fMRI response among short term abstinent teens with comorbid marijuana and alcohol use disorders compared to AUD-only and non-abusing control teens reported in a previous study . We measured BOLD response during an SWM task that typically activates bilateral prefrontal and posterior parietal networks among adults and youths . Based on our earlier findings among AUD and control adolescents, we predicted that MAUD teens would show greater fMRI response than controls in regions sub-serving SWM, including prefrontal and bilateral posterior parietal cortices. We hypothesized further that MAUD teens would show more prefrontal and parietal activation than AUD youths, since we predicted that concurrent heavy marijuana and alcohol use would influence functioning more than protracted alcohol use alone.Flyers were distributed at local high schools to recruit adolescents, as described previously .