The associations of medical marijuana policies with opioid‐related health care utilization

Abstract

Objective

To examine the associations between medical marijuana policies and opioid‐related hospitalizations and emergency department visits.

Data Sources

We utilized quarterly rates of hospital discharge data from the Healthcare Cost and Utilization Project's (HCUP) Fast Stats Database from 2005 to 2016 along with state‐level sociodemographic data from US Census Bureau and Bureau of Labor Statistics and opioid‐related state health policy data from publicly available sources for the analysis.

Study Design

Analyses were carried out using a difference‐in‐differences regression approach. We estimate heterogeneous effects of medical marijuana policies such as initial policy, presence of active dispensary, and home cultivation on opioid‐related hospitalizations and emergency department visits related to opioids.

Data Collection/Extraction Methods

Publicly available secondary data were collected, linked, and analyzed. Observations with missing values for explanatory variables were excluded from the analysis.

Principal Findings

Regression results indicate that type of medical marijuana policy has varying effects on opioid‐related hospitalizations and emergency department visits. States that allow home cultivation of medical marijuana experienced significant positive associations with opioid‐related hospitalizations and emergency department visits, while no effect was observed with medical marijuana dispensaries. Moreover, recreational marijuana policies were positively associated with opioid‐related hospitalizations.

Conclusions

The findings indicate that the effects of medical marijuana policies on opioid‐related hospitalizations and emergency department visits vary depending on the type of medical marijuana policy. Our findings indicate that the implementation of home cultivation of marijuana is positively associated with hospitalizations and emergency department visits related to opioids, suggesting that easier access to marijuana among opioid users may result in adverse health conditions that need treatment.

What is Already Known on This Topic

• Medical marijuana policy implementation is associated with reductions in opioid prescriptions, and hospitalizations related to opioid dependence or abuse, although its association with opioid overdose mortality is unclear.

• It is important to investigate not only the impact of medical marijuana policy implementation but also how different types of medical marijuana policies may affect the outcomes of interests when examining the impact of medical marijuana policy.

What This Study Adds

• The effects of medical marijuana policies on opioid‐related hospitalizations and emergency department visits vary depending on the type of medical marijuana policy.

• Increased accessibility of medical marijuana through home cultivation is associated with an increase in the number of hospitalizations and emergency department visits related to opioids.

1. INTRODUCTION

To date, 33 states and the District of Columbia (DC) have passed some form of medical marijuana policy (MMP). However, the Drug Enforcement Administration continues to classify marijuana as a Schedule I drug—defined by the federal government as “drugs with no currently accepted medical use.” While clinical effects of marijuana have yet to be determined, a few studies have examined the association of MMPs with health care utilization including prescription drugs and opioids. ¹ , ² , ³ , ⁴ , ⁵ The literature has used the terms medical marijuana policies and laws as interchangeable. We will use MMPs except for when discussing literature that has used the term medical marijuana laws (MMLs).

Bradford and Bradford ¹ , ² examine the associations of MMLs with prescription drug use among Medicare Part D and Medicaid fee‐for‐service (FFS) enrollees, respectively. They compare states that have legalized vs not legalized medical marijuana and find a reduction in prescription opioid use for which marijuana could be a possible substitute. This finding suggests that patients with such laws might be substituting medical marijuana for prescription opioids. Bradford et al ³ examine the association of MMLs with opioid prescribing in Medicare Part D population and find that MMLs were associated with significant reductions in opioid prescribing. Wen and Hockenberry ⁵ find that both MML and adult‐use marijuana law were associated with reductions in opioid prescription rates in the Medicaid population. Chihuri and Li ⁶ conduct a meta‐analysis and find MMLs are associated with a reduction of opioid prescriptions by 7%.

While recent literature finds a negative association between MMPs and opioid prescriptions, studies examining use of medical or nonmedical marijuana find positive associations between use of marijuana and prescription opioid misuse and/or opioid use disorder (OUD). ⁷ , ⁸ Utilizing a propensity score matching approach Liang et al ⁷ find that use of medical and nonmedical marijuana to be associated with increased risks of prescription opioid misuse. However, Liang and coauthors do not find use of medical marijuana to be associated with prescription OUD, although nonmedical marijuana use was positively associated with prescription OUD. Olfson et al ⁸ find marijuana use to be positively associated with nonmedical prescription opioid use and OUD. They do not differentiate between medical and nonmedical marijuana use.

While increasing opioid prescriptions and opioid overdose deaths have garnered much attention in the United States due to the opioid epidemic, ⁹ , ¹⁰ only a few studies have examined other opioid‐related health care utilizations such as hospitalizations and emergency department (ED) visits. ⁴ , ¹¹ Hospitalizations related to opioid misuse and dependence have increased, with the rate of adult hospitalizations per 100 000 population nearly doubling between 2000 and 2012. ¹² Tedesco et al ¹¹ analyzed national trends in inpatient and ED discharges for opioid abuse, dependence, and poisoning using data from Healthcare Cost and Utilization Project (HCUP) from 1997 to 2014. They found that inpatient and ED discharge rates increased overall across the study period, while a reduction was observed for prescription opioid–related discharges starting in 2010 and a sharp increase was observed in heroin‐related discharges starting in 2008.

A few studies have examined trends in hospitalizations related to opioids and the link between MMPs. Utilizing annual state‐level hospital discharge data from HCUP State Inpatient Database (SID) from 1997 to 2014, Shi (2017) considers the associations between state MMPs and hospitalizations related to both marijuana and opioid pain relievers (OPRs). Shi ⁴ finds that MMP was associated with a 23% and a 13% reductions in hospitalizations related to opioid dependence or abuse, and OPR overdose, respectively.

Pacula and Sevigny ¹³ argued that early studies of MMPs might suffer from a “policy in motion” bias, as prices and potency of marijuana are dependent on the type of MMP. That is, the introduction of more recreational marijuana policies and the entry of dispensaries have a notable effect on price and potency of marijuana, but home cultivation did not elicit a similar response. Pacula et al ¹⁴ underscore the importance of accounting for different type of MMPs such as dispensary or home cultivation when evaluating MMPs since they are not homogenous policies. MMPs also affect the implicit price of marijuana as people may consider it to be less harmful after passage and easier to obtain without using illegal transactions. ¹⁴ , ¹⁵ , ¹⁶ These effects increase the quantity demanded of marijuana. Although Shi (2017) examined the association of MMPs with hospitalizations related to opioid dependence or abuse, it relied on a smaller dataset (included only 27 states) and older data (1997‐2014). Thus, it may not fully capture the effects of MMPs in many states and misses recent implementations of MMPs by some states. Moreover, Shi (2017) was unable to examine the effect of home cultivation of medical marijuana—a major policy provision of medical marijuana legalization on hospitalizations related to opioid dependence or abuse. Both Wen and Hockenberry (2018) and Bradford et al (2018) find that states with active dispensaries and home cultivation of medical marijuana to be associated with reductions in opioid prescriptions in Medicaid and Medicare populations. Thus, it is important to investigate not only the impact of implementation of initial MMP by states but also the heterogeneous effects of MMPs such as the presence of active dispensaries or home cultivation on opioid‐related health care utilization.

Shi (2017) was unable to control for additional state‐level policy variables that may be related to opioid‐related health care utilization such as availability of naloxone, Medicaid expansion, and legalization of recreational marijuana policy. Naloxone is an opioid antagonist that helps reverse an opioid overdose. While some states require a prescription for naloxone, other states made naloxone available via pharmacies without a prescription during the period of this study. The availability of naloxone without a prescription may reduce opioid overdose deaths, but increase overdose hospital admissions and ED visits. Legalization of recreational marijuana increases access to marijuana. Individuals may substitute marijuana for their pain medications, especially opioids. Such substitutions if appropriately done may reduce opioid‐related health care utilization, while inappropriate substitutions may result in adverse health effects that require additional utilization of health care either at inpatient or at ED settings. Medicaid expansion increases access to health care by increasing health insurance coverage for low‐income individuals. These individuals may seek health care both at inpatient and at emergency settings and may get prescribe opioids especially if they are suffering from pain‐related conditions. On the other hand, if these individuals seek health care at primary settings such as doctor visits, care sought at emergency or inpatient settings for opioid‐related conditions may decrease. It is important to control for these additional policy variables since they can affect opioid‐related health care utilization.

Furthermore, Shi (2017) conducted an annual‐level analysis, while this study conducts a quarterly‐level analysis while controlling for additional state policy variables. No study to date has examined the association of MMPs with opioid‐related ED visits. The objectives of this study are to examine the associations between MMPs and opioid‐related hospitalizations and ED visits using state‐level quarterly data from 2005 to 2016. Moreover, we examine heterogeneous effects of MMPs such as presence of active dispensary or home cultivation on opioid‐related hospitalizations and ED visits.

2. METHODS

2.1. Data

The study utilizes state‐level hospital discharge data from HCUP Fast Stats Database. ¹⁷ This database currently provides quarterly rates of opioid‐related hospitalizations and ED visits for years 2005 to 2016 across 47 states and 35 states (including DC), respectively. Not all states report opioid‐related health care utilization data to HCUP since it is a voluntary partnership between the federal government and statewide data organizations. Thus, 47 states reported opioid‐related hospitalizations and 35 states reported opioid‐related ED visits to the database during 2005 to 2016. State‐level data on hospitalizations are drawn from HCUP‐SID, while ED visits are drawn from HCUP State Emergency Department Databases (SEDD). ¹⁸ , ¹⁹ Both SID and SEDD are limited to patients treated in community hospitals, where community hospitals are defined as short‐term, non‐Federal, general, and other hospitals, excluding hospital units of other institutions (eg, prisons). The SID includes more than 95% of all US hospital discharges. ¹⁸

State‐level annual sociodemographic data on population size, median household income, and percent uninsured were gathered from the US Census Bureau, while state annual unemployment rates were gathered from the US Bureau of Labor Statistics for the study period. ²⁰ , ²¹ State‐level annual per capita ethanol consumption data and beer excise tax rates were gathered from the National Institute on Alcohol Abuse and Alcoholism ²² and Tax Policy Center of the Urban Institute & Brookings Institute where the primary source is the Federation of Tax Administrators, ²³ respectively. Since state sociodemographic characteristics are annual data, we used the linear interpolation command in STATA to estimate quarterly rates of sociodemographic characteristics at the state level. We used annual values for the first quarter and estimated values for the other three quarters using linear interpolation.

Additionally, we gathered state‐level policy variables that are likely to affect opioid‐related health care utilization. These include MMP implementation, recreational marijuana legalization, presence of medical marijuana dispensary, home cultivation of medical marijuana, presence of Prescription Drug Monitoring Program (PDMP), mandatory PDMP access by providers, pill mill legislation, Good Samaritan laws, naloxone availability, and Medicaid expansion. The adoption dates of these policies were primarily gathered from Prescription Drug Abuse Policy System website, which is maintained by the RAND Corporation ²⁴ and recent literature. ³ , ¹⁴ , ²⁵ , ²⁶ , ²⁷ The opioid‐related hospitalization data sample included 2060 state‐quarter observations from 47 states while opioid‐related ED visit data sample included 1412 state‐quarter observations from 35 states between 2005 and 2016. The average number of observations per state was around 40‐43 depending on the analysis. The unit of analysis was state‐quarter.

2.2. Measures

The two key dependent variables of interest were the quarterly rates of opioid‐related hospitalizations and ED visits per 100 000 population. HCUP Fast Stats Database has calculated quarterly rates of opioid‐related hospitalizations and ED visits by dividing the total number of opioid‐related inpatient stays or ED visits in each quarter by the US resident population and multiplying that by 100 000. ¹⁷ Opioid‐related hospitalizations and ED visits include opioid abuse, dependence, and poisoning, and are identified by the database using ranges of ICD‐10‐CM and ICD‐9‐CM codes provided in Appendix S1: Table S1. Identification of opioid‐related hospitalizations and ED visits is based on all‐listed diagnoses and includes events associated with prescription opioids or illicit opioids such as heroin.

One of the key independent variables of interests is an indicator variable representing initial MMP implementation in a given state. This indicator variable takes the value of one for all quarters in the state after the implementation date of initial MMP and zero otherwise. Out of the 47 states (including DC) included in the analysis, 13 states (Arizona, Connecticut, District of Columbia, Illinois, Maryland, Massachusetts, Michigan, Minnesota, New Jersey, New Mexico, New York, Pennsylvania, and Rhode Island) implemented MMP between 2005 and 2016; 10 states (Alaska, California, Colorado, Hawaii, Maine, Montana, Nevada, Oregon, Vermont, and Washington) had already implemented MMP by 2005; and 24 states (Arkansas, Florida, Georgia, Indiana, Iowa, Kansas, Kentucky, Louisiana, Mississippi, Missouri, Nebraska, North Carolina, North Dakota, Ohio, Oklahoma, South Carolina, South Dakota, Tennessee, Texas, Utah, Virginia, West Virginia, Wisconsin, and Wyoming) had not implemented MMP by the end of 2016.

Some states extend initial MMP by allowing active dispensaries of medical marijuana, or home cultivation of a specified amount of marijuana ²⁵ , ²⁸ or both. To date, 24 states and DC have active dispensaries (at least one dispensary open in the state), while 15 states allow home cultivation of marijuana for medical purposes. Therefore, understanding whether the heterogeneity of these MMPs is associated with opioid‐related hospitalizations and ED visits is imperative. Since these policies of active dispensary and home cultivation of medical marijuana are implemented at different time points in different states, we have included two additional indicator variables to capture the implementation of a dispensary and home cultivation of medical marijuana in each state. These indicator variables, serving as additional key independent variables of interests, take the value of one for all quarters in the state after the implementation date of each policy (active dispensary present or home cultivation allowed) and zero otherwise. Implementation dates of MMPs are presented in Appendix S1: Table S2.

In order to account for confounding factors that could affect the outcome variables of interests, other time‐varying state‐level control variables were included in the analysis. These included demographic characteristics: population size (log transformed); socioeconomic factors: median household income (in constant 2017 US dollars and log transformed) and unemployment rate (percent unemployed); access to health care measure: percent uninsured; and access to alcohol measures: ethanol consumption per capita and beer taxes. We use log‐transformed values of median income and population in order to account for skewness in these data and for better fit with the outcome variables in the study which are linear. Additional state policy variables included in the analysis are indicator variables for (a) recreation marijuana implementation; (b) presence of PDMP; (c) mandatory access of PDMP by providers; (d) presence of pill mill legislation (pain management clinic regulation); (e) availability of naloxone without a prescription; (f) Good Samaritan law; and (g) Medicaid expansion. Dates used for these state policies are provided in Appendix S1: Table S3.

2.3. Analysis

To find the associations of MMP implementation on opioid‐related hospitalizations with ED visits, two sets of separate difference‐in‐differences regression analyses were conducted. ²⁹ , ³⁰ We estimate the following equation using ordinary least‐squares regression.

$$Y_{\mathit{it}}=\beta {\mathit{MMP}}_{\mathit{it}}+{\theta }_1{\mathit{Dispensary}}_{\mathit{it}}+{\theta }_2{\mathit{HomeCultiv}}_{\mathit{it}}+X_{\mathit{it}}\gamma +{\alpha }_i+{\epsilon }_t+e_{\mathit{it}},$$

where $Y_{\mathit{it}}$ is the dependent variable capturing either the rate of opioid‐related hospitalizations or ED visits per 100 000 people in state i during quarter t; ${\mathit{MMP}}_{\mathit{it}}$ is an indicator variable associated with the date of an effective initial MMP; ${\mathit{Dispensary}}_{\mathit{it}}$ is an indicator variable holding the value of one when there is an active medical marijuana dispensary; and ${\mathit{HomeCultiv}}_{\mathit{it}}$ is an indicator variable that takes the value of one if the state allows home cultivation of medical marijuana. The presence of a dispensary or home cultivation can occur after the implementation of the initial MMP or at the same time depending on the state. The initial MMP implementation provides legal protection for physicians to recommend marijuana for a medical purpose for certain patient indication (eg, nonspecified pain), while it can also provide legal protection for individuals with a physician's recommendation to possess marijuana. MMP implementation is associated with a decrease in the perceived harm of marijuana by adolescents and young adults as well as an increase in general marijuana use. ¹⁵ , ¹⁶ , ³¹ , ³²

State policies such as recreation marijuana implementation, PDMP, pill mill legislation, naloxone availability, Good Samaritan laws, and the Medicaid expansion along with state sociodemographic characteristics are captured in matrix $X_{\mathit{it}}$. Both state (${\alpha }_i$) and quarter (${\epsilon }_t$) fixed effects were included in all specifications to account for state‐specific unobserved time‐invariant factors and for time (quarter) specific unobserved factors that may be common to all states. Robust Huber‐White standard errors clustered at the state level were estimated to capture arbitrary within‐state heteroscedasticity. Following Allison, ³³ we tested for multicollinearity between median household income, unemployment rate, and percent uninsured using variance inflation factor, and no evidence of multicollinearity was detected. In Appendix S1, we conduct an event study to provide suggested evidence that the parallel trends assumption of the difference‐in‐differences model holds. We estimate treatment effects four years before and after implementation. We do find some anticipatory treatment effects within one year of implementation that may correspond with the passage of an MMP, but we find no statistically significant effects prior to this point for either hospitalizations or ED visits. All statistical analyses were conducted in Stata version 14.2 (StataCorp LP). Several robustness measures are included in Appendices S1 and S2.

3. RESULTS

The average opioid‐related hospitalization and ED rates during 2005‐2016 were higher in states that implemented MMP compared with states that did not implement MMP (Table 1). Interestingly, states that have implemented MMP in the study sample have not implemented pill mill legislations (Table 1). Moreover, higher proportion of states with MMP have expanded Medicaid coverage under the Affordable Care Act (Table 1).

Table 1.

Summary statistics of variables by states that have implemented medical marijuana policy and states that have not implemented medical marijuana policy

Variable	MMP = 0			MMP = 1
	Obs.	Mean	Std. Dev.	Obs.	Mean	Std. Dev.
Opioid‐related hospitalization per 100,000 population	1072	158.33	78.26	988	245.46	95.43
Opioid‐related ED visits per 100 000 population	70	109.55	71.15	652	185.90	98.11
Medical marijuana policy implementation	1072	0.00	0.00	988	0.69	0.46
Medical marijuana dispensary open	1072	0.00	0.00	988	0.34	0.47
Medical marijuana home cultivation	1072	0.00	0.00	988	0.58	0.49
Pill mill legislation	1072	0.14	0.34	988	0.00	0.00
Recreation marijuana legalization	1072	0.00	0.00	988	0.05	0.23
Naloxone availability	1072	0.15	0.36	988	0.23	0.42
Prescription drug monitoring program (PDMP)	1072	0.66	0.48	988	0.59	0.49
Mandatory access of PDMP	1072	0.08	0.28	988	0.07	0.26
Medicaid expansion	1072	0.07	0.25	988	0.25	0.43
Good Samaritan law	1072	0.14	0.34	988	0.32	0.47
Unemployment rate	1072	5.92	2.00	988	6.58	2.17
Percent uninsured	1072	13.51	4.09	988	11.41	4.62
Log(Population)	1072	15.28	0.91	988	15.29	1.06
Log(Median Income) in 2017US$	1072	10.88	0.13	988	11.03	0.13
Ethanol consumption per capita (in Gallons)	1072	2.20	0.41	988	2.53	0.35
Beer taxes	1072	0.31	0.25	988	0.22	0.21

MMP = 1 refers medical marijuana policy implementation states, while MMP = 0 refers to states that did not implement medical marijuana policy. Obs. refers to number of observations, and Std. Dev. refers to standard deviation.

Tables 2 and 3 present the results of the primary regression analyses that examined the associations of MMPs with opioid‐related hospitalizations and ED visits, respectively. In Tables 4 and 5, we present results of an extension of our primary analyses as robustness measures, where we explicitly model states with MMP only separately from states with MMP + dispensary only (no home cultivation), MMP + home cultivation only (no dispensary), and MMP + dispensary +home cultivation. We present the results of our primary analyses along with a comparison of the results of the robustness checks below.

Table 2.

Regression results of the associations of medical marijuana policies with opioid‐related hospitalizations per 100 000 population

Variables	(1)	(2)	(3)
	Opioid inpatient rate	Opioid inpatient rate	Opioid inpatient rate
MMP implementation	−28.62 (14.20)*	−30.09 (14.13)*	−30.66 (14.43)*
Medical marijuana dispensary open	14.27 (7.846)	15.35 (7.832)	15.11 (7.655)
Medical marijuana home cultivation	53.87 (18.47)**	54.09 (18.52)**	53.60 (18.39)**
Pill mill law	27.65 (12.61)*	22.83 (11.67)	19.88 (12.40)
Recreational marijuana	41.16 (20.37)*	40.24 (20.12)	41.53 (19.31)*
Naloxone access	22.00 (7.061)**	20.70 (6.990)**	20.06 (6.942)**
PDMP effective	−7.974 (5.632)	−7.017 (5.715)
Medicaid expansion	10.85 (9.252)	14.08 (9.233)	11.43 (8.981)
Good samaritan laws	−3.902 (8.111)	−5.115 (7.772)	−5.450 (7.784)
Unemployment rate	3.649 (1.367)*	3.471 (1.372)*	3.144 (1.312)*
Percent uninsured	−1.470 (1.296)	−1.733 (1.243)	−1.653 (1.225)
LN‐population	−116.5 (163.9)	−125.8 (162.2)	−108.3 (159.5)
LN‐median income (in 2017US$)	32.81 (35.34)	23.78 (33.52)	24.86 (32.60)
Ethanol consumption per capita	40.98 (37.47)	31.64 (34.05)	30.47 (33.83)
Beer taxes		61.71 (11.71)**	55.52 (11.75)**
PDMP mandatory			12.63 (10.08)
Constant	1451 (2608)	1702 (2551)	1426 (2500)
Observations	2060	2060	2060
R 2	0.801	0.808	0.809
Number of states	47	47	47
STATE FE	YES	YES	YES
Quarter FE	YES	YES	YES

Abbreviations: MMP, medical marijuana policy; PDMP, prescription drug monitoring program; robust standard errors in parentheses.

^** P < 0.01

^* P < 0.05.

Table 3.

Regression results of the associations of medical marijuana policies with opioid‐related emergency department visits per 100 000 population

Variables	(1)	(2)	(3)
	Opioid emergency rate	Opioid emergency rate	Opioid emergency rate
MMP implementation	1.888 (18.16)	1.042 (17.90)	2.054 (17.93)
Medical marijuana dispensary open	7.731 (12.91)	7.720 (12.86)	6.929 (12.66)
Medical marijuana home cultivation	82.86 (32.22)*	82.76 (32.28)*	77.18 (28.28)**
Pill mill law	22.06 (12.71)	19.36 (13.33)	11.68 (14.19)
Recreational marijuana	0.564 (10.00)	0.747 (9.936)	1.954 (9.977)
Naloxone access	17.55 (8.977)	16.62 (9.165)	15.77 (9.145)
PDMP effective	−6.216 (7.302)	−5.585 (7.415)
Medicaid expansion	27.32 (13.42)*	29.27 (13.77)*	20.92 (13.70)
Good Samaritan laws	8.296 (9.656)	8.688 (9.754)	7.355 (10.08)
Unemployment rate	−2.891 (2.161)	−3.091 (2.153)	−3.448 (2.112)
Percent uninsured	−0.428 (1.785)	−0.415 (1.794)	−0.853 (1.545)
LN‐population	−678.7 (199.9)**	−687.7 (201.9)**	−663.6 (207.9)**
LN‐median income (in 2017US$)	−49.27 (36.53)	−53.03 (37.04)	−51.58 (40.14)
Ethanol consumption per capita	−61.83 (46.91)	−69.40 (46.61)	−76.23 (48.17)
beer taxes		23.11 (12.40)	6.044 (15.43)
PDMP mandatory			27.74 (13.68)
Constant	11 109 (3106)**	11 300 (3137)**	10 945 (3273)**
Observations	1412	1412	1412
R 2	0.832	0.833	0.838
Number of states	35	35	35
STATE FE	YES	YES	YES
Quarter FE	YES	YES	YES

Abbreviations: MMP, medical marijuana policy; PDMP, prescription drug monitoring program; robust standard errors in parentheses.

^** P < 0.01

^* P < 0.05

Table 4.

Regression results of the associations of medical marijuana policies with opioid‐related hospitalizations per 100 000 population, using mutually exclusive four dummy variables

Variables	(1)	(2)	(3)
	Opioid inpatient rate	Opioid inpatient rate	Opioid inpatient rate
MMP implementation only	−29.73 (14.31)*	−30.24 (14.87)*	−29.59 (14.55)*
MMP + Dispensary only	−15.17 (15.60)	−16.06 (15.55)	−15.13 (16.00)
MMP + Home cultivation only	23.84 (13.89)	22.75 (13.37)	22.41 (13.20)
MMP + Dispensary+Home cultivation	39.46 (16.06)*	38.19 (16.21)*	38.93 (15.50)*
Pill mill law	22.85 (11.68)	19.90 (12.43)	22.75 (11.65)
Recreational marijuana	40.19 (20.04)	41.47 (19.23)*	40.11 (19.41)*
Naloxone access	20.72 (6.969)**	20.08 (6.922)**	20.90 (7.133)**
PDMP effective	−7.012 (5.729)		−7.176 (5.759)
Medicaid expansion	14.07 (9.242)	11.42 (8.985)	14.26 (9.068)
Good Samaritan laws	−5.114 (7.770)	−5.449 (7.782)	−4.466 (7.630)
Unemployment rate	3.469 (1.372)*	3.141 (1.309)*	2.814 (1.526)
Percent uninsured	−1.726 (1.267)	−1.646 (1.246)	−1.649 (1.259)
LN‐population	−125.8 (162.2)	−108.3 (159.5)	−156.2 (153.2)
LN‐median income (in 2017US$)	23.73 (33.57)	24.81 (32.63)	31.18 (33.01)
Ethanol consumption per capita	31.52 (34.06)	30.34 (33.86)
Beer taxes	61.75 (11.68)**	55.56 (11.78)**	63.96 (11.28)**
PDMP mandatory		12.63 (10.10)
Constant	1703 (2551)	1427 (2501)	2158 (2411)
Observations	2060	2060	2060
R 2	0.808	0.809	0.807
Number of states	47	47	47
STATE FE	YES	YES	YES
Quarter FE	YES	YES	YES

MMP implementation only takes the value of 1 only during MMP implementation then reverts back to 0 when dispensary or home cultivation becomes available.

Abbreviations: MMP, medical marijuana policy; PDMP, prescription drug monitoring program; robust standard errors in parentheses.

^** P < 0.01

^* P < 0.05.

Table 5.

Regression results of the associations of medical marijuana policies with opioid‐related emergency department Visits per 100 000 population, using mutually exclusive four dummy variables

Variables	(1)	(2)	(3)
	Opioid emergency rate	Opioid emergency rate	Opioid emergency rate
MMP implementation only	13.74 (18.59)	14.93 (19.11)	14.36 (19.20)
MMP + Dispensary only	−4.131 (19.97)	−3.948 (18.14)	−3.347 (20.51)
MMP + Home cultivation only	76.91 (28.25)*	72.31 (24.63)**	83.71 (28.68)**
MMP + Dispensary+Home cultivation	98.26 (32.92)**	93.01 (28.38)**	102.8 (32.02)**
Pill mill law	20.48 (13.20)	12.83 (14.17)	21.71 (13.82)
Recreational marijuana	−2.659 (10.90)	−1.453 (11.13)	−4.493 (11.00)
Naloxone access	18.71 (9.351)	17.91 (9.411)	19.07 (9.619)
PDMP effective	−5.218 (7.041)		−3.891 (7.370)
Medicaid expansion	28.45 (13.12)*	20.06 (12.87)	25.79 (13.09)
Good Samaritan laws	7.618 (9.774)	6.291 (10.08)	5.009 (10.17)
Unemployment rate	−3.039 (2.185)	−3.387 (2.160)	−1.959 (2.147)
Percent uninsured	−0.300 (1.744)	−0.736 (1.486)	−0.773 (1.654)
LN‐population	−677.8 (194.9)**	−653.7 (200.3)**	−612.5 (195.3)**
LN‐Median Income (in 2017US$)	−57.64 (34.40)	−55.87 (36.84)	−79.84 (33.18)*
Ethanol consumption per capita	−74.54 (46.49)	−81.61 (47.66)
Beer taxes	23.55 (12.11)	6.369 (15.32)	13.83 (12.63)
PDMP mandatory		27.74 (13.52)*
Constant	11 211 (3011)**	10 852 (3139)**	10 289 (3004)**
Observations	1412	1412	1412
R 2	0.837	0.842	0.833
Number of states	35	35	35
STATE FE	YES	YES	YES
Quarter FE	YES	YES	YES

MMP implementation only takes the value of 1 only during MMP implementation then reverts back to 0 when dispensary or home cultivation becomes available.

Abbreviations: MMP, medical marijuana policy; PDMP, prescription drug monitoring program; robust standard errors in parentheses.

^** P < 0.01

^* P < 0.05.

The results from the full specifications (columns 2 & 3 in Table 2) indicate that initial MMP implementation is associated with a reduction of about 30 opioid‐related hospitalizations per 100 000 population per quarter (P < 0.05). This result is consistent with the results in Table 4 that compares MMP implementation only states (no home cultivation or dispensary) with no MMP states and is statistically significant at 5% level. We show this result is robust to additional specifications and subsamples in Appendix S1: Table S6. We report the results of a Goodman‐Bacon Decomposition in Appendix S2. ³⁴ , ³⁵

Many states choose to extend the initial MMP by allowing marijuana dispensaries, home cultivation, or both. The effect of marijuana dispensaries and home cultivation in primary regression analyses (Tables 2 and 3) should be interpreted as a deviation from having the initial MMP alone. States that allow home cultivation of medical marijuana experiences 54 additional opioid‐related hospitalizations per 100 000 per quarter (P < 0.01) than states with no home cultivation but only the initial MMP or a net effect of 24 additional opioid‐related hospitalizations per 100 000 per quarter compared with states that have no MMP, which is qualitatively consistent with the results in Table 4 that compares states with MMP and home cultivation (MMP + Home cultivation) with states that have no MMP. This net effect translates to about 9.8% additional opioid‐related hospitalizations from the average of 245 hospitalizations per quarter. Results in Table 4 also show that states with MMP, dispensary, and home cultivation (MMP + dispensary +home cultivation) are positively and significantly associated with opioid‐related hospitalizations compared with states with no MMP (P < 0.05). However, we do not find any discernible effect of medical marijuana dispensaries on opioid‐related hospitalizations after conditioning on the MMP.

Table 3 presents the results of the regression analyses that examined the associations of MMPs with opioid‐related ED visits. The results from the full specification (Column 3, Table 3) show that states that allow home cultivation of medical marijuana are associated with about 77 additional opioid‐related ED visits compared with states without home cultivation (P < 0.01) but with initial MMP. This result is consistent with the results in Table 5 that compares states with MMP and home cultivation (MMP + home cultivation) with no MMP states, and is statistically significant at 5% level. Moreover, results in Table 5 show that states with MMP, dispensary, and home cultivation (MMP + dispensary +home cultivation) are positively and significantly associated with opioid‐related ED visits compared with states with no MMP (P < 0.01). The availability of marijuana dispensaries or the initial MMP is not significantly associated with opioid‐related ED visits.

In Appendix S1: Table S7, we disaggregate home cultivation policies into “unsupervised” and “permit required.” We find home cultivations policies requiring a permit do not adversely affect opioid‐related hospitalization, but unsupervised home cultivation policies are associated with increases in hospitalizations. We disaggregate dispensaries into “legal” vs “open” where the former has a registered business permit, but do not find any statistically significant differences. As a further robustness measure, we restrict the initial MMP with only those that allow medical marijuana for “nonspecified pain.” We find the bulk of the decrease in hospitalizations is associated with this type of MMP. States allowing medical marijuana for nonspecified pain as a valid reason experience 22‐31 fewer hospitalizations per 100 000 people per quarter.

States that have implemented recreational marijuana policies experience about 41 more opioid‐related hospitalizations per 100 000 per quarter compared with states that have not implemented recreational marijuana policy (Tables 2 and 4; P < 0.05). Note that no state has implemented recreational marijuana policy without implementing MMP first. Additionally, states that make naloxone available without a prescription experience about 20 additional opioid‐related hospitalizations compared with states requiring a prescription for naloxone (Tables 2 and 4; P < 0.01). This result is potentially driven by the ability of naloxone to revive individuals who have overdosed, so they can receive medical attention. Both unemployment rate (P < 0.05) and beer taxes (P < 0.01) were positively associated with opioid‐related hospitalizations.

4. DISCUSSION

The findings indicate that the effects of MMPs on opioid‐related hospitalizations and ED visits vary depending on the type of MMP. States that allow home cultivation of marijuana seem to be experiencing higher rates of opioid‐related hospitalizations compared with states without home cultivation but with initial MMP (Table 2). This result contrasts with those of Shi ⁴ that indicated negative association of MMP with hospitalizations related to opioid dependence or abuse as well as opioid overdose although her analysis did not include home cultivation. Our result of medical marijuana dispensaries having no effect on opioid‐related hospitalizations is consistent with the findings of Shi ⁴ where medical marijuana dispensaries have no impact on opioid‐related hospitalizations.

We expand the analysis of Shi (2017) by estimating the heterogeneous effects of MMP on opioid‐related hospitalizations and ED visits. Our findings indicate that states with home cultivation of medical marijuana in addition to having the initial MMP experience higher rates of opioid‐related hospitalizations and ED visits compared with states with only the initial MMP (Tables 2 and 3). Our results show that states with MMP, dispensary, and home cultivation are significantly and positively associated with opioid‐related hospitalizations and ED visits (Tables 4 and 5). Together, these results indicate that while medical marijuana may help reduce opioid overdoses ³⁶ due to possible substitution of marijuana for opioids, ³ home cultivation of marijuana may provide an easier avenue for accessing marijuana not only for the patient but also for anyone connected with the patient (ie, family members, friends, etc) increasing misuse of marijuana (in the place of opioids or in combination with opioids) among opioid users that results in adverse health effects needing medical attention either at hospital inpatient or ED settings. One could argue the MMPs do not have an immediate effect. We present results of the lagged MMP variables (Appendix S1 Tables S8 and S9). These results show that home cultivation of medical marijuana remains significantly and positively associated with hospitalization and ED visit rates even after lagged for three quarters. Although the negative association of initial MMP with hospitalization rate remains consistent with additional lags, it is no longer significant at accepted levels.

A recent study by Liang et al ⁷ finds use of medical marijuana to be associated with increased risks of prescription opioid misuse, while Olfson et al ⁸ find marijuana use to be positively associated with nonmedical prescription opioid use and OUD, which may increase opioid‐related health care utilization among opioid users. Thus, our results seem to indicate that home cultivation may not be an effective policy tool in reducing opioid‐related hospitalization and ED visits.

Recreational marijuana legalization was associated with about 16%‐17% increase in opioid‐related hospitalizations, while it has no significant association with opioid‐related ED visits. This indicates that use of recreational marijuana by opioid users may result in adverse health effects such as increased risks of prescription opioid misuse and OUD that requires hospitalizations. ⁷ , ⁸

We also find that availability of naloxone without a prescription is significantly associated with increased opioid‐related hospitalizations. We find a similar effect on opioid‐related ED visits though it was not significant at 5% level. While naloxone may help reduce mortality by reversing opioid overdoses, individuals that experience overdoses may seek health care at inpatient or ED settings. Thus, it is not surprising that availability of naloxone is associated with increased opioid‐related hospitalizations.

These findings contribute to the literature on MMPs and opioid‐related health care utilization by providing additional evidence on implementation of MMP along with different types of MMP impact on hospitalizations and ED visits related to opioids. Although initial MMP implementation allows physicians to write a recommendation for medical marijuana for a condition that the state law has approved to be treated by medical marijuana and provides legal protection for patients to possess marijuana, it is up to the patient to obtain medical marijuana via a dispensary, or home cultivation. ³⁷ , ³⁸ Thus, it is important to investigate not only the impact of initial MMP implementation but also how different types of MMPs may affect the outcomes of interests when examining the impact of MMPs. ³

While some studies have supported MMPs as a potential harm reduction mechanism toward opioid epidemic policy debate, ¹ , ² , ²⁷ , ³⁹ , ⁴⁰ it remains unclear whether marijuana liberalization may be beneficial to the society as a whole especially since recent findings have indicated positive associations of marijuana use and MMPs with opioid misuse, OUD, and opioid overdose mortality. ⁷ , ⁸ , ⁴¹ The findings of this study underscore the importance of weighing potential benefits and harms of different types of MMPs in the policy debate on medical marijuana as a policy alternative in addressing harms related to the opioid epidemic.

This study has a few limitations. First, this study focuses only on the period of 2005‐2016 and uses state‐level aggregated data. If earlier year data or discharge‐level data were available, results may differ from the current study. Second, this study focuses on opioid‐related hospitalizations and ED visits that include opioid‐related disorders, opioid dependence, abuse, and overdose, and is unable to separately identify the reasoning behind hospitalizations or ED visits due to the aggregate nature of the data used in the analysis. Third, this study includes state‐level aggregated data on opioid‐related hospitalizations and ED visits from only 47 states and 35 states, respectively. If data on all 51 states were available for both outcome variables of interests, the findings may differ from the current study. Fourth, this study is unable to identify the source of drugs used such as prescription or illicit (ie, heroin) opioids that resulted in hospitalizations or ED visits due to the aggregate nature of the data used in the analysis. Finally, since no state has adopted recreational marijuana policy without first implementing MMP in the state, the estimation of recreational marijuana policy is likely to provide only a partial estimate of that policy.

Despite these limitations, the findings of this study provide an important contribution to the policy debate on medical marijuana legalization. Our findings show that the effects of MMP on opioid‐related hospitalizations and ED visits vary depending on the type of MMP. The findings indicate that increased access to marijuana via home cultivation seems to be positively associated with opioid‐related hospitalizations and ED visits, suggesting that easier access to marijuana among opioid users may result in adverse health conditions that need treatment. While it remains unclear whether marijuana liberalization may be beneficial to the public health in the fight for opioid epidemic, the results of this study taken together with recent findings in the literature help support the argument that potential benefits and adverse health outcomes associated with different types of MMPs should be taken into consideration when discussing marijuana as a policy alternative in addressing the opioid epidemic.

Supporting information

Supplementary Material

Appendix S1

Jayawardhana J, Fernandez JM. The associations of medical marijuana policies with opioid‐related health care utilization. Health Serv Res. 2021;56:299–309. 10.1111/1475-6773.13632