The impact of medical and recreational marijuana laws on opioid prescribing in employer-sponsored health insurance

Abstract

Using data from Truven Health MarketScan Commercial Claims and Encounters Database between 2009 and 2015, we studied the effects of medical and recreational marijuana laws on opioid prescribing in employer-sponsored health insurance. We used a differences-in-differences (DD) approach and found that the implementation of medical marijuana laws (MMLs) and recreational marijuana laws (RMLs) reduced morphine milligram equivalents per enrollee by 7% and 13%, respectively. The reduction associated with MMLs was predominately in people aged 55–64, whereas the reduction associated with RMLs was largely in people aged 35–44 and aged 45–54. Our findings suggest that both MMLs and RMLs have the potential to reduce opioid prescribing in the privately insured population, especially for the middle-aged population.

1 |. INTRODUCTION

The opioid epidemic in the United States has reached a crisis level. The economic cost of the crisis was estimated at $504 billion in 2015, or 2.8% of the total gross domestic product (GDP; The Council of Economic Advisers, 2017). Excessive prescribing of opioids for pain management is viewed as a major driver of the ongoing opioid epidemic in the United States (Manchikanti et al., 2012; Rudd, Aleshire, Zibbell, & Matthew Gladden, 2016; Volkow, 2014).

Prescription opioids are primarily used for adult pain management in the United States (e.g., Brummett et al., 2017). However, the scientific evidence to back the effectiveness of prescription opioids in chronic noncancer pain is limited (Lee, Silverman, Hansen, Patel, & Manchikanti, 2011; Manchikanti et al., 2012), and there is evidence of serious consequences of opioid use, including opioid addiction, opioid-induced hyperalgesia and opioid-related deaths (Chu, Clark, & Angst, 2006; Lee et al., 2011; Rudd et al., 2016; Yin, Mufson, Wang, & Shi, 1999).

Marijuana may provide an alternative for pain management at a relatively low risk of addiction and virtually no risk of overdose (Abrams, Couey, Shade, Kelly, & Benowitz, 2011; Hill, 2015; Lynch & Campbell, 2011; National Academies of Sciences, Engineering, and Medicine, 2017; Reiman, Welty, & Solomon, 2017; Whiting et al., 2015). While marijuana is still a Schedule I drug at the federal level,¹ 36 states and the District of Columbia have legalized medical marijuana use as of December 2020 (Berke & Gould, 2019; ProCon.org, 2020a). Fifteen states with medical marijuana laws (MMLs) and the District of Columbia further legalized adult recreational marijuana use (Berke & Gould, 2019; ProCon. org, 2020a).

Although MMLs and recreational marijuana laws (RMLs) were not originally adopted to reduce opioid prescribing and harms associated with opioid use, empirical studies have found that state MMLs and RMLs are associated with reductions in opioid prescribing for Medicare Part D enrollees (Bradford, Bradford, Abraham, & Bagwell Adams, 2018), for Medicaid enrollees (Shi et al., 2019; Wen & Hockenberry, 2018), and for the privately insured (McMichael, Van Horn, & Viscusi, 2020), as well as reductions in opioid-related overdose deaths (Bachhuber, Saloner, Cunningham, & Barry, 2014; Livingston, Barnett, Delcher, & Wagenaar, 2017; Powell, Pacula, & Jacobson, 2017), treatment rates related to opioid addiction and overdose (Powell et al., 2017; Shi, 2017), and opioid-related fatal accidents (Kim et al., 2016). Those empirical findings suggest that MMLs and RMLs may affect opioid use and downstream adverse consequences.

We contribute to the growing literature by examining how state MMLs and RMLs may affect opioid prescribing in the working age population (age from 18 to 64) with employer-sponsored health insurance and by exploring the heterogeneous effects of the marijuana laws in different age groups. Nearly four in ten people addicted to opioids are covered by private health insurance, yet this population has not been studied extensively in research on marijuana policy and opioid prescribing (Cox, Rae, & Sawyer, 2018). Our study population is of particular importance in the current opioid crisis, which is largely a crisis in the working age population. The opioid overdose death rates of working age population (age from 18 to 64) are substantially higher than those of other age groups and have increased significantly since 1999 (Supplementary Figure A1). Particularly, the United States opioid crisis is also a midlife crisis. Drug overdose deaths, especially opioid-related drug overdose deaths, disproportionally contributed to the mortality rate of individuals aged 45 to 54 (Supplementary Figure A1). Opioid overdose has shortened the life expectancy of middle-aged people in the United States (Case & Deaton, 2015). Our study looks into different age groups among individuals with employer-sponsored health insurance, thus better capturing the at-risk population than the previously studied Medicare and Medicaid populations (Bradford et al., 2018; Wen & Hockenberry, 2018).

2 |. LITERATURE

The effects of MMLs and RMLs on opioid prescribing depend on whether marijuana is a substitute for or complement to prescription opioids as both laws de facto lower the price of qualified marijuana use through removing legal penalties and increasing marijuana supply.

State MMLs authorize both adults and minors to use marijuana to treat qualified conditions. Conditions qualified for medical marijuana use vary from state to state, but generally include severe or chronic pain, as well as other conditions, such as cancer, glaucoma, acquired immunodeficiency syndrome (AIDS) (or human immunodeficiency virus (HIV) positive), and Hepatitis C. The majority of the medical marijuana states require patient registration, meaning that patients generally need to provide a physician written certification, in exchange for legal protection. Some states provide affirmative defense for medical use of marijuana even if self-claimed medical marijuana patients fail to register with states (PDAPS, 2017; ProCon.org, 2020a).

State RMLs (i.e., adult-use marijuana laws) expanded the reach of marijuana legalization by allowing almost all adults aged 21 and above to use marijuana (ProCon.org, 2020b). Therefore, the laws provide additional legal protection for people who were not qualified as medical marijuana patients to use marijuana for various purposes, including self-medicating.

Empirical studies have found that state MMLs and RMLs are associated with decreases in opioid use. Bradford et al. (2018) find that an MML with either operational dispensaries or home cultivation provisions is associated with reductions in daily doses of opioid prescriptions in state Medicare Part D populations between 2010 and 2015. Wen and Hockenberry (2018) find that MMLs and RMLs are associated with reductions in opioid prescribing rates and spending in Medicaid enrollees between 2011 and 2016. Shi et al. (2019) find that RMLs are associated with reductions in the number of prescriptions, morphine milligram equivalents (MME), and spending specific to Schedule III opioids in Medicaid enrollees between 2010 and 2017. McMichael et al.’s (2020) find that MMLs and RMLs are associated with reductions in opioid prescriptions covered both by public insurance and by private insurance between 2011 and 2018.

MMLs and RMLs are also found to reduce harms related to opioid use, such as opioid-related overdose deaths (Bachhuber et al., 2014; Livingston et al., 2017; Powell et al., 2017), opioid-positive fatalities (Kim et al., 2016), and opioid-related hospitalization (Shi, 2017).

3 |. METHODS

3.1 |. Data and sample

We used the Truven Health MarketScan Commercial Claims and Encounters Database between 2009 and 2015, which captures medical claims and encounters of a national and state representative sample of active employees and their dependents, early retirees, and Consolidated Omnibus Budget Reconciliation Act (COBRA) enrollees.² Our data were aggregated at the state/month level. We also limited our sample to those aged 18–64 with at least one-year continuous enrollment (enrollment span greater than or equal to 365 days). Please see Supplementary Table A1 for the number of enrollees in each state and each year.

3.2 |. Variable measurement

Our main outcome is monthly MME^3,4 per enrollee. MME is the most commonly used and best available way to standardize prescription opioids according to the formulation, strength, and dosage.

We identified opioid prescriptions based on Medispan Generic Product Identifiers.⁵ We tracked each prescription back to seven days prior to its fill date to link the prescription to at least one diagnosis. We excluded prescriptions prescribed for the patients in hospice or palliative care (the International Statistical Classification of Diseases and Related Health Problems (ICD)-9: V66.7 and ICD-10: Z51.5), with cancer diagnosis (ICD-9: 140-239.90 and ICD-10: C00-C97),⁶ and with missing diagnoses (no diagnosis recorded in the seven days prior to the prescription fill date). We also excluded buprenorphine prescriptions that are commonly prescribed for medication-assisted treatment of opioid addiction (e.g., Wen, Schackman, Aden, & Bao, 2017).

We further studied MME per enrollee separately by age groups (i.e., aged 18 to 24, 25 to 34, 35 to 44, 45 to 54, and 55 to 64) to investigate the potential policy heterogeneity. MME per enrollee in each age group was calculated as the total MME prescribed in an age group divided by the total number of enrollees in that age group.

We also looked into the sources of changes in MME (i.e., extensive margin vs. intensive margin). Considering that prescription opioids are primarily used in pain management, while off-label drug use is also a common practice in the United States, we estimated the effects of MMLs and RMLs in pain patients and nonpain patients separately. We identified pain patients based on ICD-9 or ICD-10 codes and included all diagnoses likely to be associated with chronic or acute pain conditions (Centers for Disease Control and Prevention, 2013; Ilgen et al., 2013; Mack, Zhang, Paulozzi, & Jones, 2015; Narayana et al., 2015). To explore the intensive margin (i.e., MME prescribed to each patient) and extensive margin (the number of patients prescribed opioids) of the changes in MME per enrollee, we studied the following four additional outcomes: MME per pain patient prescribed opioids, MME per nonpain patient prescribed opioids, the number of pain patients prescribed opioids per 1000 enrollees, and the number of nonpain patients prescribed opioids per 1000 enrollees.

The key independent variables are the implementation of an MML and the implementation of an RML in a given state during a given month. We defined an MML in a state to be in effect if the state provides legal protection for patients who possess or use marijuana for medical purposes based on their physicians’ recommendations complying with the law and for physicians who recommend medical marijuana to their patients complying with the law. We defined an RML in a state to be in effect if the state provides legal protection for adults who possess or use marijuana for nonmedical purposes complying with the law.⁷ Please see Tables 1 and 2 for detailed policy summaries of MMLs and RMLs.

TABLE 1.

State medical marijuana law (MML) effective dates, as of December 10, 2020

State	Effective date	Data source
Alaska	1999/03	PDAPS.org
Arizona	2011/04	PDAPS.org
Arkansas	2016/11	PDAPS.org
California	1996/11	PDAPS.org
Colorado	2001/06	ProCon.org
Connecticut	2012/06a	PDAPS.org
Delaware	2011/07	PDAPS.org
Florida	2017/01	PDAPS.org
Hawaii	2000/06	PDAPS.org
Illinois	2014/01	PDAPS.org
Louisiana	2016/08	ProCon.org
Maine	1999/12	PDAPS.org
Maryland	2014/06	ProCon.org
Massachusetts	2013/01	PDAPS.org
Michigan	2008/12	PDAPS.org
Minnesota	2014/06a	PDAPS.org
Mississippi	2020/11	ProCon.org
Missouri	2018/12	ProCon.org
Montana	2004/11	PDAPS.org
Nevada	2001/10	PDAPS.org
New Hampshire	2013/07	PDAPS.org
New Jersey	2010/10	PDAPS.org
New Mexico	2007/07	PDAPS.org
New York	2014/07	PDAPS.org
North Dakota	2016/12 to 2017/01b 2017/04	PDAPS.org
Ohio	2016/09	PDAPS.org
Oklahoma	2018/06	ProCon.org
Oregon	1998/12	PDAPS.org
Pennsylvania	2016/05	PDAPS.org
Rhode Island	2006/01	PDAPS.org
South Dakota	2020/11	ProCon.org
Utah	2018/12	ProCon.org
Vermont	2004/07	PDAPS.org
Virginia	2020/10	NBC12.com
Washington	1998/12	PDAPS.org
West Virginia	2017/04	ProCon.org

Sources: Prescription Drug Abuse Policy System (PDAPS, 2017) and ProCon.org. (2020a).

Notes: We first relied on Prescription Drug Abuse Policy System (PDAPS) to collect the MML effective dates. Then, we read legal documents through the ProCon.org to verify the dates on PDAPS. When there was inconsistency in an effective date, we relied on the legal documents to adopt the most possible date. Since PDAPS only updated MML effective dates through February 2017, we collected MML effective dates after February 2017 from ProCon.org.

^aConnecticut and Minnesota had MML in effect at the end of the months, so the effective month is the next month.

^bNorth Dakota had MML effective for less than a month, and then put MML on hold.

TABLE 2.

State recreational marijuana law effective dates, as of December 10, 2020

State	Effective date	Data source
Alaska	2015/02	APIS
Arizona	2020/11	ProCon.org
California	2016/11	APIS
Colorado	2012/12	APIS
Illinois	2020/01	ProCon.org
Maine	2017/01	APIS
Massachusetts	2016/12	APIS
Michigan	2018/12	APIS
Montana	2020/11	ProCon.org
Nevada	2017/01	APIS
New Jersey	2020/11	ProCon.org
Oregon	2015/07	APIS
South Dakota	2020/11	ProCon.org
Vermont	2018/07	APIS
Washington	2012/12	APIS

Sources: Alcohol Policy Information System (APIS, 2019) and ProCon.org. (2020b).

Notes: Since Alcohol Policy Information System only updated recreational marijuana law (RML) effective dates through March 2019, we collected RML effective dates after March 2019 from ProCon.org.

State-level time-varying covariates include concurrent polices (i.e., state prescription drug monitoring programs and mandates, and pain clinic laws), general economy indicators (i.e., unemployment rate, median household income, and poverty rate), other health-related measures (i.e., primary care physician supply and binge drinking rate), and state population. Please see Supplementary Table A2 for the descriptive summary of the study variables.

3.3 |. Analytic strategies

To estimate the effects of MMLs and RMLs on opioid prescribing, we used a differences-in-differences (DD) approach, operationalized through a two-way fixed-effects model:

$$Y_{s,t}={\beta }_0+{\beta }_{1}mm{l}_{s,t}+{\beta }_{2}rm{l}_{s,t}+\gamma X_{s,t}+{\delta }_s+{\eta }_t+{\theta }_{s,t}+{\epsilon }_{s,t,}$$

where s denotes a state and t denotes a month in a year. Y_s,t represents the opioid-related outcomes. mml_s,t and rml_s,t are the DD indicators for state implementation of MMLs and RMLs. X_s,t is a vector of state-level covariates. We included state fixed effects δ_s and year-month fixed effects η_t to account for the unobserved time-invariant state heterogeneity and the national secular trend and common shocks in opioid prescribing. We also included state-specific linear time trends θ_s,t to account for state-wide confounding factors that evolve at a constant rate over time. Figure 1 lends support to the parallel-trend assumption regarding our policy indicators—that is, in the prepolicy period, changes in MME per enrollee in policy states overtime are not different from those in control states.

FIGURE 1.

Pre- and posttrend in MME per enrollee in policy states relative to the control states.

Source: Truven Health MarketScan Commercial Claims and Encounters Database.

Notes: Estimates were simultaneously estimated, controlling for state-level covariates, state fixed effects, year-month fixed effects, and state-specific linear time trends. Standard errors were clustered at the state level. Month zero refers to the first month in which an MML or RML took effect. One month prior to a law effect month was the excluded dummy in the estimation

Abbreviations: MME, morphine milligram equivalent; MML, medical marijuana laws; RML, recreational marijuana laws

4 |. RESULTS

4.1 |. Estimated effects of MMLs and RMLs on MME per enrollee

State implementation of MMLs and RMLs was associated with reductions in MME per enrollee (Table 3; Supplementary Table A4). Specifically, the implementation of MMLs was associated with a reduction of 2.41 MME per enrollee, which can be translated to a 7% relative reduction. The implementation of RMLs was associated with a reduction of 4.43 MME per enrollee, equivalent to a 13% relative reduction.

TABLE 3.

Effects of medical marijuana laws (MML) and recreational marijuana laws (RML) on morphine milligram equivalent (MME) per enrollee

Marginal effect	MME per enrollee
MML in effect	−2.41** (0.80)
RML in effect	−4.43** (1.47)
Baseline predicted mean	33.37
# Observations	4200
R-squared	0.80

Source: Truven Health MarketScan Commercial Claims and Encounters Database.

Notes: Standard errors in parentheses. Baseline predicted mean was calculated as the average of predicted values when setting mml_i,t and rml_i,t to 0, and leaving the other covariates as the observed values.

^*p < 0.05;

^**p < 0.01;

^***p < 0.001.

4.2 |. Event study results

An event analysis with lag and lead policy indicators allows us to estimate the differential effects of MMLs and RMLs on the basis of the current month relative to the effective date (Model, 1993; Figure 1).

We discerned no prepolicy difference in MME per enrollee between states with and without marijuana laws, which lends weight to the parallel-trend assumption of the DD approach. After MMLs and RMLs taking effect, we observed immediate and sustained policy effects on MME per enrollee.

4.3 |. Heterogeneity of estimated effects of MMLs and RMLs in different age groups

When examining the changes in MME per enrollee across age groups (Figure 1; Supplementary Table A5), we found that the reduction associated with MMLs was concentrated in the age 45–54 group, whereas the reduction associated with RMLs was concentrated in the age 35–44 and 45–54 groups.

4.4 |. Sources of changes in MME per enrollee

In patients with chronic or acute pain conditions, MMLs and RMLs were shown to reduce MME per enrollee through reducing both MME per pain patient prescribed opioids (intensive margin) and the number of pain patients prescribed opioids per 1000 enrollees (extensive margin; Table 4; Supplementary Table A6). In addition to the policy effects in pain patients, the implementation of MMLs was associated with a reduction in MME per nonpain patient prescribed opioids but not the number of nonpain patients prescribed opioids per 1000 enrollees. In comparison, the implementation of RMLs was associated with a reduction in the number of nonpain patients prescribed opioids per 1000 enrollees while not MME per nonpain patient prescribed opioids. Some primary diagnoses of the nonpain patients were abdominal pain, chest pain and headache. We classified patients with these conditions as nonpain patients, because for them opioids were not considered appropriate nor commonly prescribed. The nonpain patients may self-medicate with marijuana once MMLs or RMLs were in effect. Both pain patients and nonpain patients may be better able to access marijuana once RMLs were in effect because RMLs would allow those who were previously unqualified for medical marijuana use or unable to find medical marijuana dispensaries to access marijuana as states with RMLs do not require doctors’ recommendations and allow more dispensaries.

TABLE 4.

Effects of medical marijuana laws (MML) and recreational marijuana laws (RML) on other opioid-related outcomes

Marginal effects	Morphine milligram equivalent (MME) per pain patient prescribed opioids	MME per nonpain patient prescribed opioids	# Pain patients prescribed opioids per 1000 enrollees	# Nonpain patients prescribed opioids per 1000 enrollees
MML in effect	−96.19** (34.67)	−159.72*** (44.20)	−0.23** (0.09)	0.03 (0.06)
RML in effect	−149.25* (64.64)	−88.95 (88.61)	−1.37*** (0.22)	−0.83*** (0.15)
Baseline predicted mean	1436.52	1729.37	12.94	8.60
# Observations	4200	4200	4200	4200
R-squared	0.80	0.73	0.94	0.93

Source: Truven Health MarketScan Commercial Claims and Encounters Database.

Notes: Standard errors in parentheses. Baseline predicted mean was calculated as the average of predicted values when setting mml_i,t and rml_i,t to 0, and leaving the other covariates as the observed values.

^*p < 0.05;

^**p < 0.01;

^***p < 0.001.

4.5 |. State-specific policy effects

The effects of MMLs and RMLs on MME per enrollee varied across states, with some of the largest reductions in Colorado (an RML state), Maryland (an MML state), Oregon (an RML state), and Delaware (an MML state; Supplementary Table A7). The state-specific policy effects were not precisely estimated in other MML and RML states.

4.6 |. Effects of the physical availability

An MML or RML taking effect does not necessarily mean that the targeted population immediately have legal access to marijuana if the law only protects marijuana use but not commercial production, sales or home cultivation.⁸

In the main model, we defined an MML or RML to be in effect as long as the law provides legal protection for marijuana users. Here we define an MML or RML to be in effect if (1) a law or agency rule explicitly allowing medical or recreational marijuana home cultivation has been in effect for at least 2 months (the minimum time to grow useable marijuana is about eight weeks⁹), or (2) there is at least one active medical or recreational marijuana dispensary, whichever comes first (Supplementary Table A8 and A9).

We found that the physical availability of recreational marijuana was associated with a reduction of 8.91 MME per enrollee—that is, a 27% relative reduction (Table 5; Supplementary Table A10). The effect of the physical availability of recreational marijuana was larger than that of the RML in the main model. However, the physical availability of medical marijuana had no discernable effect on MME per enrollee. This suggests that for medical marijuana users, legal protection may be more important than legal availability of medical marijuana.

TABLE 5.

Effects of legal physical availability of medical marijuana and legal physical availability of recreational marijuana on morphine milligram equivalent (MME) per enrollee

Marginal effects	MME per enrollee
The physical availability of medical marijuana	0.02 (0.57)
The physical availability of recreational marijuana	−8.91*** (1.65)
Baseline predicted mean	32.56
# Observations	4200
R-squared	0.80

Source: Truven Health MarketScan Commercial Claims and Encounters Database.

Notes: Standard errors in parentheses. Baseline predicted mean was calculated as the average of predicted values when setting legal physical availability of medical marijuana in effect and legal physical availability of recreational marijuana in effect to 0 and leaving the other covariates as the observed values.

^*p < 0.05;

^**p < 0.01;

^***p < 0.001.

5 |. DISCUSSION & CONCLUSIONS

This study advances our understanding of the impact of both MMLs and RMLs on opioid prescribing in the privately insured population. We found that the implementation of MMLs was associated with a 7% relative reduction in MME per enrollee and the implementation of RMLs was associated with a 13% relative reduction in MME per enrollee (Table 3; Supplementary Table A4). The reduction associated with MMLs was concentrated in the age 55–64 group, whereas the reduction associated with RMLs was largely in the age 35–44 and 45–54 groups (Figure 2; Supplementary Table A5). With respect to the source of reductions, MMLs and RMLs may have reduced opioid prescribing at both intensive and extensive margins for patients with chronic or acute pain conditions (Table 4; Supplementary Table A6).

FIGURE 2.

The effects of medical marijuana laws (MML) and recreational marijuana laws (RML) on morphine milligram equivalent (MME) per enrollee in different age groups

Source: Truven Health MarketScan Commercial Claims and Encounters Database

Similar effects of MMLs and RMLs on opioid prescribing have been found in Medicaid, Medicare, and privately insured populations (Bradford et al., 2018; McMichael et al., 2020; Shi et al., 2019; Wen & Hockenberry, 2018). We also found that RMLs were associated with larger reductions in MME than MMLs. The physical availability of recreational marijuana was associated with further reductions in MME beyond the effect of only providing legal protection for recreational marijuana use.

Our study further suggests that MMLs and RMLs may directly benefit middle-aged people in the opioid crisis. The reduction in MME per enrollee aged 45 to 54 associated with RMLs was the largest across all age groups, and the economically significant reductions associated with both laws were concentrated in the middle-aged population. Those results were consistent with the studies that found the use of marijuana among middle-aged adults increased significantly from 2006 to 2016 (Han et al., 2017; Han & Palamar, 2018). The increasing use of marijuana by middle-aged people might have steered them away from opioid-related harms.

However, our study has the following limitations. First, due to the nature of our claim data, we were unable to observe the individual drug substitution mechanism. We only have the records of prescribed opioids that were covered by the employer-sponsored insurance. Therefore, we did not know whether those individuals used marijuana or not, or whether those individuals also bought opioids with cash. It is also possible that random attrition of employers could affect the estimates. Thus, our results did not mean to imply causality. It merits future research to explore the mechanisms between marijuana liberalization and changes in opioid prescribing in the privately insured population.

Second, the differential changes in enrollment across states, particularly from 2014 to 2015 following the implementation of the Affordable Care Act, could potentially bias our main findings and result in spurious correlations. Therefore, we conducted four additionally robustness checks (Supplementary Table A11–A14; Supplementary Figure A2–A3) and did not detect significant differences in enrollment between states with and without MMLs and RMLs.

Third, we also do not know the medical potential of marijuana. Marijuansa is still a schedule I drug at the federal level and there was no large-scale clinical trial to investigate its medical use. State MMLs based on limited clinical evidence suggest its use on certain pain conditions, such as neuropathic pain, due to its analgesic effect in humans (Abrams et al., 2011), which cannot rule out the possibility that medical marijuana can be used to treat other conditions for which opioids were prescribed. More studies, perhaps individual-level surveys, are needed to further discuss the mechanisms of possible opioid and marijuana substitutions among privately insured population.

Finally, we were unable to test the underlying mechanisms through which state policies may influence individual behavior. Future research may look into the potential pathways such as price and risk perception.