Marijuana, Opioids, and Some Thoughts on Empirical Research on Marijuana Policy
Back in February, I discussed research suggesting that legalizing marijuana could help combat the nation’s opioid epidemic. See Marijuana and the Opioid Crisis. The theory is that once marijuana is legalized some people will use marijuana in lieu of opioids to treat chronic pain. It is the reverse of the gateway hypothesis that was once endorsed by the DEA—that version suggested that using marijuana might actually increase the use of opioids (and other hard drugs).
Since I made the post, the Journal of the American Medical Association (JAMA) – Internal Medicine, has published two new studies that appear to bolster the claim that marijuana legalization could curb opioid harms. Both studies can be downloaded for free from the JAMA website by following the hyperlinks below. I’ll briefly describe each of the studies; I’ll then offer a few comments on their value and conclude with a broader observation about the role I think empirical research ought to play in the marijuana policy domain.
The first study, Association of Medical and Adult Use Marijuana Laws with Opioid Prescribing for Medicaid Enrollees (by Hefei Wen and Jason Hockenberry), found that between 2011-2016, legalizing marijuana was correlated with a roughly 6% lower rate of opioid prescriptions for Medicaid patients. The drop was slightly bigger for states that passed recreational marijuana laws during this period (all of them already had medical laws in place), and slightly smaller for states that passed only medical marijuana laws. The 6% average figure translates into about 40 fewer prescriptions for opioids for every 1,000 patients in the Medicaid program.
The second study, Association Between US State Medical Cannabis Laws and Opioid Prescribing in the Medicare Part D Population (by Ashley Bradford, et al.), similarly found that between 2010-2015, legalizing medical marijuana was correlated with a 9% lower rate of opioid prescriptions for Medicare Part D patients. In particular, the authors found that states with medical marijuana laws on average filled 9% fewer opioid prescriptions than did states without such laws. As with the Medicaid study, Bradford et al. find some variation among marijuana laws. Namely, the drop in opioid prescriptions was steeper in states with legal medical marijuana dispensaries (14.4%) compared to those without (6.9%).
Now, let me make three observations.
First, the Medicaid and Medicare studies nicely complement earlier research (discussed in the February post linked above), and thereby lend credibility to claims that legalizing marijuana could curb opioid harms. While the earlier research found a negative correlation between legalizing marijuana and opioid deaths, it failed to find any negative correlation between legalizing marijuana and opioid use (qua supplies). For reasons discussed in my earlier post, this non-result is somewhat puzzling—after all, marijuana legalization is supposed to curb opioid deaths by steering people away from opioids, not, e.g., by making opioids safer to use. The Medicaid and Medicare studies arguably fill this gap by demonstrating a negative correlation between legalizing marijuana and opioid use (for which prescriptions are a reasonable proxy). To be sure, the Medicaid and Medicare studies are not the first to demonstrate this link (both studies acknowledge earlier research). But the Medicaid and Medicare studies arguably improve upon the design of that earlier research, e.g., by trying to account for some differences in marijuana laws that had previously been neglected.
Nonetheless—and this is my second observation—the reader should take the latest studies with a grain of salt. Even though I think the Medicaid and Medicare studies are comparatively well-designed, I still found problems with them that could undermine their validity.
Let me highlight just one of those problems: the failure to satisfactorily control for key differences in marijuana policies across the states (and over time). For example, although both studies claim to control for whether a state has legal and operational marijuana dispensaries, they disagree about when particular states met those criteria. The Medicaid study, for example, claims that New York state opened its first legal dispensary in January 2016, whereas the Medicare study puts the date 6 months later—in June, 2016. (The Supplements to both articles list these dates.) Furthermore, it’s not clear how the study authors came up with some of the dates. For example, the Medicare study suggests that Colorado opened its first legal medical marijuana dispensary back in July 2005. But that’s perhaps 5 years too early, at least by my reckoning. To be sure, Colorado had dispensaries back in 2005; but the legality of those dispensaries was highly doubtful until the legislature passed the Medical Marijuana Code in 2010.
Apart from imprecisely measuring one key variable (dispensaries), the latest studies also fail to control for a variety of other potentially important variations in state policy that could affect their results. Consider just one such variable: whether chronic pain is considered a Qualifying Condition under a state medical marijuana law. The book covers Qualifying Conditions on pages 101-110 and 203-209. As discussed on those pages, states have adopted different lists of Qualifying Conditions (see Figure 5.2 on page 204 in particular), and not all consider chronic pain to be one. It follows that if State A considers chronic pain a qualifying condition, but State B does not, we should expect State A’s medical marijuana program to have a bigger impact on opioid use than State B’s, ceteris parabis. After all, the theory is that marijuana provides a substitute for opioids in treating pain. But as far as I can tell, the Medicaid study doesn’t control for the inclusion of chronic pain as a qualifying condition; and the Medicare study suggests—erroneously, I think—that “Chronic pain is listed as an approved condition (either directly, or by implication) in every state [Medical Cannabis Law].” (Bradford et al., page E2) (emphasis added). To be sure, the vast majority of states now list chronic or intractable or severe pain as a qualifying condition. But several states have only recently done so (e.g., Connecticut, Illinois, Louisiana, and New Jersey)—i.e., they did so only after the study period (2010-2015 or 2011-2016) was concluded. Other states have listed chronic/intractable pain as a Qualifying Condition for longer periods of time, but some require pain patients to jump through special hurdles (like exhausting other treatment options, including opioids) before trying marijuana (see book p. 105 n.2)—and those hurdles might limit the use of marijuana as a substitute for opioids in treating pain.
Ideally, researchers would come to some consensus on 1) which policy variables to control for in studies; and 2) how to control for them. I suspect researchers now agree that the presence of dispensaries matters, but so could a variety of other policy choices that differ across states and that are (as yet) rarely noted in empirical research: lists of qualifying conditions, patient registration requirements, legal protections granted to patients, limits imposed on physicians who recommend the drug to patients, and so on. Even after agreeing upon which of these policy differences might matter, researchers will still need to agree upon how to measure those differences (i.e., how to operationalize them). Regarding dispensaries, for example, should researchers care only when they first opened? When they first became legal (under state law)? Or should researchers also (or instead) care about their ability to supply the state’s overall market, e.g., their number relative to the state’s qualifying patient population?
Ultimately, the problem is that when a study leaves out a potentially important variable—or fails to precisely measure it—we can’t necessarily trust the study’s results. It’s the classic problem of spurious correlation. The study may find a correlation between A and B, but that correlation might be caused by some third (or fourth, or fifth . . .) variable that was omitted.
I am skeptical of much empirical research on marijuana policy (among other subjects) because of this problem. So while I think the latest studies (including the Medicaid and Medicare studies) continue to refine and improve the analyses, I don’t think anyone should put too much weight on them in making important policy choices—or at the very least, should not base those decisions on empirical studies alone.
Which leads me to my third, and final, and broadest point: policymakers shouldn’t neglect arguments based on reason, logic, and values when making decisions in this area. Empirical research can make policy choices somewhat easier, but it takes a long time to build consensus around empirical research, and on many policy questions, that consensus never arrives (or doesn’t survive long, as new doubts are constantly raised about old research). So policymakers can’t—or at least, shouldn’t—put off making hard choices waiting for social scientists / scientists to make those choices for them. In other words, if someone is offering sound and reasonable arguments why legalizing marijuana is (or is not ) a good idea, policymakers should listen to those arguments, and not let the presence or absence of empirical research drown them out.
Anyway, check out the studies above. There is also a useful commentary on the first study at the JAMA website by Kevin P. Hill, The Role of Cannabis Legalization in the Opioid Crisis. NPR has also covered the studies at Opioid Use Lower in States that Eased Marijuana Laws.