Study aims and hypotheses
Study aim 1 is to examine the effects of state medical cannabis laws on receipt of clinical guideline-discordant opioid and clinical guideline-concordant non-opioid, non-cannabis treatment among patients with chronic non-cancer pain. We will use difference-in-differences design with a policy trial emulation approach  adapted from comparative effectiveness research to identify the comparison group. We expect state medical cannabis laws to reduce receipt of opioid and non-opioid, non-cannabis treatment among patients with low back pain, headache, fibromyalgia, arthritis, and/or neuropathic pain.
Study aim 2 is to examine the effects of state medical cannabis laws on receipt of treatment for opioid use disorder, opioid overdose, cannabis use disorder, and cannabis poisoning among patients with chronic non-cancer pain. We will use the same difference-in-differences design with policy trial emulation approach as in aim 1. We expect state medical cannabis laws to decrease utilization for opioid use disorder and opioid overdose and to increase utilization of treatment for cannabis use disorder and cannabis poisoning among patients with low back pain, headache, fibromyalgia, arthritis, and/or neuropathic pain.
Aim 1–2 hypotheses related to policy implementation rules
In aims 1–2, we will analyze how specific policy implementation rules modify laws’ effects on outcomes. Hypotheses pertaining to specific policy implementation rules are shown in Table 1.
Study aim 3 is to characterize implementation of state medical cannabis laws for treatment of chronic non-cancer pain. Through interviews with state decision-makers and healthcare system leaders, we will collect in-depth information on leaders’ perceptions of Proctor’s eight implementation outcomes  as well as characterize key implementation strategies such as presence of state initiatives designed to support use of medical cannabis for treatment of chronic non-cancer pain and healthcare system policies related to medical cannabis treatment.
Study aim 4 is to characterize physician and patient perspectives of state medical cannabis laws as they pertain to the treatment of chronic non-cancer pain and to quantitatively measure four implementation outcomes: acceptability, appropriateness, adoption, and penetration. Through representative surveys of physicians and chronic pain patients in states with medical cannabis laws, we will examine perceived acceptability and appropriateness of medical cannabis for treatment of chronic non-cancer pain; the proportion of physicians who recommend medical cannabis to chronic non-cancer patients, refer patients to a recommending physician, or recommend against medical cannabis for chronic non-cancer pain management; and the proportion of people with chronic non-cancer pain who report using medical cannabis for pain management. Surveys will also measure barriers and facilitators to the use of cannabis for chronic non-cancer pain. Stratified analyses will explore whether relevant attitudes, for example patient perceptions of medical cannabis access, differ depending upon the policy implementation rules of the medical cannabis law in respondents’ state of practice (physicians) or residence (patients).
The study uses a concurrent-embedded design , in which one primary method (difference-in-differences analyses, aims 1–2) guides the research, and secondary qualitative (aim 3) and survey (aim 4) methods play a supportive role. Aims 1–3 begin concurrently. Aim 3 interview results will inform aim 4 survey development; aim 4 will be conducted concurrently with the later phases of aims 1–2.
The study sample includes 32 states: 17 control states without medical cannabis laws (AL, GA, ID, IN, IA, KS, KY, MS, NE, NC, SC, SD, TN, TX, VA, WI, WY) and 15 intervention states that enacted medical cannabis laws in 2012 or later and do not have recreational cannabis laws (AR, CT, FL, LA, MD, MN, MO, NH, NY, ND, OH, OK, PA, UT, WV). We excluded states that enacted medical cannabis laws prior to 2012 due to concern about recall bias in aim 3 and excluded states that have enacted both medical and recreational cannabis laws since 2012 due to our study’s focus on medical cannabis laws; recreational cannabis laws could lead people to “self-medicate” with cannabis obtained through recreational channels and contaminate aim 1–2 quantitative analyses.
The study period for the overarching study is 2009–2022, a period chosen to include 3 years of pre-law data for the states with the earliest (2012) medical cannabis law enactment dates in the sample. In quantitative aims 1–2, each of the 15 intervention states with medical cannabis laws will have a unique 7-year study period, with 3 years of data pre- and 4 years of data post-law. The rationale for this approach is that it is critical to examine the effects of state medical cannabis laws among continuous cohorts of patients with chronic non-cancer pain in order to attribute observed effects to medical cannabis laws as opposed to the changing composition of the study sample. But, requiring continuous presence of individuals in the aims 1–2, insurance claims data across the entire time period would substantially reduce sample size; we therefore will only require continuous enrollment for the 7-year study period relevant for each state. Aim 3 qualitative interviews will seek to characterize implementation timing, barriers, and strategies from the date a state’s law was enacted through the time interviews are conducted in 2021/2022. Aim 4 surveys will characterize physicians’ and patients’ attitudes and behaviors at the time the surveys are fielded in 2022.
Aim 1–2 state medical cannabis law data
Our study team assembled a longitudinal state medical cannabis law database using legal research and legislative history techniques, including full-text searches of the Westlaw database and identification of state session laws and regulatory materials. The longitudinal database includes each law’s effective date, date the first dispensary opened, and time-varying measures of the policy implementation rules of interest (see Table 1). For quality control purposes, we compared our findings with publicly available materials compiled by the Prescription Drug Abuse Policy System  and the National Conference on State Legislatures . When we found inconsistences between our results and these materials, we consulted the text of the relevant law and sought clarification from legal experts in the relevant state.
Aim 1–2 administrative insurance claims data
Aims 1–2 will use Medicare and OptumLabs Data Warehouse administrative claims. The Medicare data includes inpatient, emergency department, outpatient, and prescription drug insurance claims for the approximately 34 million adults aged 65+ and nine million adults aged 18–64—who qualify for Medicare coverage by virtue of disability—covered by fee-for-service Medicare each year [51, 52]. The OptumLabs data used for this study include inpatient, emergency department, outpatient, and prescription drug insurance claims for approximately 30 million privately insured adults aged 18–64. Both data sources include data from all 50 US states. Information in these two claims data sources includes diagnosis codes; procedure codes; type, dose, and duration of prescriptions; and service dates, allowing for identification of individuals diagnosed with one of the five chronic non-cancer pain conditions of interest and the pharmacologic and non-pharmacologic pain treatments they receive. Unique patient identifiers allow tracking of individuals over time and across treatment settings. Patient demographic information includes age, sex, state, and five-digit zip-code of residence. Unique provider identifiers allow tracking of providers over time. Provider characteristics include specialty and treatment setting.
Aim 3 qualitative interview data
Aim 3 qualitative data will be collected through semi-structured interviews with key state policy and healthcare system leaders in the 15 intervention states. The guide will include three cross-cutting domains relevant for both groups of interviewees: (1) perceptions of policy implementation rules; (2) perceived barriers and facilitators to implementation of state medical cannabis laws for the treatment of chronic non-cancer pain; and (3) perceptions of Proctor’s eight implementation outcomes  in the context of medical cannabis law implementation for chronic non-cancer pain. The interview guide for state policy leaders will include two additional domains focused on (a) eliciting leaders’ perceptions of factors influencing the design of medical cannabis policy implementation rules and (b) characterizing state medical cannabis law implementation initiatives. The interview guide for state healthcare system leaders will include two domains in addition to the cross-cutting domains above, which will focus on (a) healthcare system leaders’ perceptions of how their state’s medical cannabis law has influenced treatment of chronic non-cancer pain and (b) characterizing healthcare system policies or other initiatives related to medical cannabis.
The interview guide will be developed by the study team and refined based on feedback from the study’s advisory board, which includes national and state experts in pain management, medical cannabis, addiction medicine, and drug policy. Videoconference interviews will be conducted by a single master’s-level study team member trained in qualitative interviewing techniques. Table 2 delineates our qualitative research design within the Consolidated Criteria for Reporting Qualitative Studies (COREQ) framework , including additional details regarding interview guide development and data collection.
Aim 4 survey data
Aim 4 data will be collected through surveys of representative samples of three groups: (1) all primary care and pain specialist physicians, (2) primary care and pain specialist physicians registered to recommend medical cannabis in their state, and (3) people with chronic non-cancer pain in the 15 intervention states. Physician surveys will be administered by the study team using a multi-contact, mixed-mode strategy following an adapted Dillman tailored design method , which includes three email survey waves and three postal survey waves. Patient surveys, which will be administered by NORC using their nationally representative AmeriSpeak panel, will be conducted by telephone or online, depending upon AmeriSpeak panel members’ preference . Survey domains and item development will be informed by aim 3 interview results. See the measures section below for preliminary domains. Table 3 characterizes our survey research design within the Survey Reporting Guideline (SURGE) framework , including additional details regarding survey development and data collection.
The aim 1–2 study sample will include individuals continuously enrolled in Medicare or private insurance for their state’s 7-year study period and individuals in control states who are continuously enrolled for the same 7-year periods as the intervention states. The sample includes people were diagnosed with a chronic non-cancer pain condition—low back pain, fibromyalgia, chronic headaches, arthritis, or neuropathic pain—during the study period.
The aim 3 study sample will include key state policy and healthcare system leaders in the 15 intervention states. We will begin by interviewing (1) the individual with primary responsibility, as determined by statute, for implementation of the state’s medical cannabis law, and (2) the Chief Medical Officers of two large integrated healthcare systems in each state (when more than two exist, we will select one in an urban area and one in a rural area, if possible). Additional interviewees will be identified through purposive snowball sampling . Interviewees will receive a standard recruitment email explaining the study aims and inviting them to participate in the study. We plan to conduct interviews with 3–4 policy leaders and 3–4 healthcare system leaders in each intervention state.
In aim 4, we will survey three samples. First, we will survey a random sample of 2000 primary care physicians and 2000 pain specialist physicians practicing in the 15 intervention states. The sample will be drawn from the National Provider Plan and Provider Enumeration System (NPPES), a Centers for Medicare and Medicaid Services (CMS) physician database. Second, we will survey 2000 primary care physicians and 2000 pain specialist physicians registered to recommend medical cannabis in the 10 intervention states that require physicians to register with the state in order to recommend cannabis to patients. The sample will be drawn from state databases obtained by our study team. Third, we will survey a representative sample of 1500 people with chronic non-cancer pain in the 15 intervention states using the NORC AmeriSpeak panel , a nationally representative, probability-based survey panel. We will identify people with chronic non-cancer pain using the National Health Interview Survey metric on national prevalence .
Aim 1–2 independent variables are dichotomous indicators of medical cannabis laws that will “turn on” (from 0 to 1) in the first full year the law is implemented. Aim 1–2 effect modifiers, or “moderators,” are dichotomous indicators of each of the specific policy implementation rules (Table 1), except dispensary and registered physician proximity. These variables will be measured as the distance in driving miles from the centroid of patients’ residential zip code to the nearest dispensary/registered physician.
Aim 1 dependent variables include measures of receipt of opioids, non-opioid pain medications, and non-pharmacologic therapies (e.g., physical therapy). Our study team identified non-opioid pain medications and non-pharmacologic therapies recommended by clinical guidelines to treat low back pain, headache, fibromyalgia, arthritis, and/or neuropathic pain. We will create person-year level measures indicating whether an individual received any opioid medication, any guideline-concordant non-opioid pain medication—with guideline-concordance defined as receipt of a medication recommended for that individual’s specific chronic non-cancer pain diagnosis—and any guideline-concordant non-pharmacologic therapy. Among individuals who received these treatments, we will measure the number of treatments per year. Aim 2 dependent variables include person-year measures of any inpatient, emergency department, or outpatient visits for opioid use disorder, opioid overdose, cannabis use disorder, and cannabis poisoning. Among individuals with any utilization, we will measure number of visits per person-year.
Aim 3 qualitative interviews will characterize the implementation of medical cannabis laws as they pertain to chronic non-cancer pain in the 15 intervention states. Key themes will be identified within the domains described in the data collection section above. These key themes will inform development of aim 4 survey domains by identifying issues related to medical cannabis law implementation that warrant further exploration through surveys of representative samples of physicians and patients. Preliminary aim 4 domains include:
Survey 1, primary care and pain specialist physicians: perceived acceptability and appropriateness of medical cannabis for chronic non-cancer pain; perceptions of demand for cannabis among chronic non-cancer pain patients; perceived effectiveness of cannabis as a treatment for chronic non-cancer pain; and self-reported chronic non-cancer pain treatment practices, including but not limited to recommendation of medical cannabis for management of chronic non-cancer pain (a measure of adoption). As this survey will be fielded among a representative sample of physicians, we will measure medical cannabis law penetration as the percent of all physicians who report recommending medical cannabis to patients with chronic non-cancer pain.
Survey 2, physicians registered to recommend cannabis to patients: measures of adoption including self-reported practices regarding recommending, referring, and monitoring cannabis for chronic non-cancer pain; perceived barriers and facilitators to delivering cannabis treatment; factors that influence treatment modality decisions, e.g., cannabis versus opioids versus non-opioid treatments.
Survey 3, people with chronic non-cancer pain: We will measure the policy implementation outcome measure of adoption as self-reported use of medical cannabis to manage chronic non-cancer pain. The survey will also measure self-reported use of prescription opioids, non-opioid prescription medications, and non-pharmacologic therapies to manage chronic non-cancer pain; perceptions of barriers and facilitators to cannabis treatment; and policy attitudes, for example whether or not insurers should cover medical cannabis.
Aim 1–2 difference-in-differences study
We will use a difference-in-differences design with a policy trial emulation approach to compare trends in outcomes pre/post medical cannabis laws in intervention states to changes in outcomes in control states over the same period. We will identify individual patients with chronic non-cancer pain for inclusion in the analytic comparison groups by adapting methods used in comparative effectiveness research to account for variation in law implementation date across states and taking advantage of longitudinal data on each person [58,59,60,61]. This method, as applied to public policy evaluations, has recently been described as “policy trial emulation” . This approach is conceptualized as a series of all possible initiation trials over time, where each year a state implemented a medical cannabis law represents the start of a new trial, and comparison individuals are selected relative to that start year; data across trials are pooled.
As noted in the study sample section above, each intervention state has its own 7-year study period and sample of individuals continuously enrolled during that 7-year period. For each intervention state, we will create a control group using the year of law implementation to define the “time zero”: individuals in states without medical cannabis laws continuously enrolled during a given intervention state’s 7-year study period will be weighted by fitting a model predicting the probability of living in the medical cannabis law state as a function of observed patient characteristics, measured before the law’s implementation year. Individuals in intervention states receive a weight of 1 and comparison individuals receive a weight proportional to (1-p)/p, were p = probability of living in a medical cannabis law state. Comparison individuals in states with no medical cannabis laws can be used in multiple trials for intervention states implementing medical cannabis laws at different time points; variance estimation will account for those repeated observations and for the clustering of individuals over time and within states [62, 63].
We will use a flexible regression framework, fit in the weighted sample, to evaluate medical cannabis laws’ effects on outcomes. Consider Y any of the outcome measures, f(∙) a function to transform Y (e.g., logit), law an indicator for the intervention group, and Post an indicator of whether the observation is “post” medical cannabis law adoption. The basic difference-in-differences model is: (1) f(Y) = α + β1Law + γ2Post + δlaw*post + ε. δ is the coefficient of interest: the effect of exposure to a state medical cannabis law on outcome Y. In addition to main models estimating the average effects of state medical cannabis laws over the entire post-law period, we will separately estimate effects for years 1–4 post-law by allowing δ to vary by year post-implementation. Effect modification will be measured with a three-way interaction between law, post, and the modifier (as well as all two-way interactions of those 3 variables). While aim 3–4 measures of policy implementation outcomes are primarily designed to support interpretation of aim 1–2 model results, we will explore the possibility of including state-level implementation outcome measures (e.g., the proportion of physicians in a given state in the aim 4 survey sample who report recommending cannabis for chronic non-cancer pain, a measure of penetration) as effect modifiers in aim 1–2 analyses.
Aim 3 qualitative analysis
After each interview, the interviewer will create a summary memo to help identify preliminary themes within the data. Interview transcripts will be analyzed using a staged approach, starting with general coding and then including more specific codes as data analysis proceeds and researchers develop and refine a working model for the relationships within the data. The study team will create an initial codebook based on the research questions and summary notes from interviews. The codebook will be refined through input from the study’s advisory board. Then, two coders will pilot the codebook using a randomly selected sub-sample of transcripts. Discrepancies between the two coders will be resolved through a discussion and consensus process with the full study team. If additional themes emerge during the pilot phase, they will be added to the codebook. In the case of significant disagreements, we will solicit additional review from the advisory board. The finalized codebook will be applied to all transcripts. Qualitative research software will be used to organize text segments, first descriptively and then by themes and sub-themes. See Table 2 for more details.
Aim 4 survey analysis
All analyses will incorporate sampling weights adjusting for known sampling deviations and survey nonresponse. We will calculate the proportion, with 95% confidence intervals, of respondents endorsing each survey item response option. Statistical significance of potential sub-group comparisons, for example planned comparisons of survey responses among respondents living in states with differing medical cannabis policy implementation rules, will be assessed using chi-square tests. See Table 3 for more details.