Medication Overuse Headache in Chronic Migraine Patients Using Cannabis: A Case-Referent Study

Objective : To examine whether cannabis use predicts medication overuse headache (MOH) in chronic migraine (CM) patients. Methods : Electronic chart review was conducted by combining the terms “chronic migraine”, “medication overuse”, “cannabis”, “CBD”, “THC” for patients seen at our headache clinics from 2015 to 2019. Of 729 charts identified, 368 (150 using cannabis; 218 not using cannabis) met our inclusion criteria, i.e., adult CM patients with ≥ 1-year CM duration. The following variables were extracted from each patient’s chart: MOH diagnosis as dependent variable, and predictor variables as age, sex, migraine frequency, current CM duration, current cannabis use duration, overused acute migraine medications, current MOH duration, and types of cannabis products used. Logistic regression was employed to identify variables predicting MOH while controlling for remaining predictors. Agglomerative hierarchical clustering (AHC) was conducted to explore natural clusters using all predictor variables. Results : There were 212 CM patients with MOH ( cases ) and 156 CM patients without MOH ( referents ). Current cannabis use statistically significantly predicted cases with MOH – odds ratio 6.0 (3.45, 10.43), p < 0.0001 . Current cannabis use, opioid use, and MOH were significantly associated. AHC revealed two major natural clusters. Cluster I patients were younger with less migraine frequency, higher MOH burden, more current cannabis and opioid users than cluster II. Conclusion : Cannabis use significantly contributes to the prevalence of MOH in CM. Bidirectional cannabis-opioid association was observed – use of one is a presumed migraine trigger 31 . Based on these results and our findings, we that cannabis Our results did not show opioid-sparing effects of cannabis use in chronic migraine patients with or without MOH. In contrast, we found increased association between current cannabis use and current opioid use. However, longitudinal studies will be the appropriate design to examine this relationship so as to explore cause and effect. There is conflicting evidence in opioid-sparing effects of cannabis in the general population. Some cross-sectional studies demonstrate the protective effect of cannabis use from developing opioid dependence 32 and speculate that cannabis may potentiate opioid efficacy leading to reduced opioid dosage 33 . A study in migraine patients has found that 43% patients substituted their opiates/opioids to cannabis 34 . Other prospective studies indicate that cannabis increases the risk of opioid dependence or opioid use disorder 18,35 . Our study showed mild association between current cannabis use and current benzodiazepine use. Another retrospective study has reported increased benzodiazepine discontinuation among mixed cohort of patients with pain and non-pain conditions 36 . In our study, no association was found between current cannabis use and current butalbital use. Our results are the first to explore relationships and MOH risk among these four dependence-causing drugs commonly used by migraine patients i.e. cannabis-opioid, cannabis-Zhang


Introduction
Medication overuse headache (MOH) is a consequence of regular overuse of acute headache medications in patients with pre-existing primary headache disorders, such as migraine (ICHD-3). MOH presents as either a new headache or worsening of a pre-existing headache 1 . Prevalence of MOH in the general population is 0.5-2.6%; among patients with chronic daily headache, the prevalence of MOH is estimated to be 11-70% 2 . The Nord-Trøndelag Health Survey (Helseundersøkelsen i Nord-Trøndelag: HUNT), a community-based study from Norway involving nearly 50,000 participants, showed that 54% of chronic migraine patients had comorbid MOH 3 . A study from Latin American countries reported that up to 95% of MOH patients have migraine 4 . Treatment of MOH requires tapering and discontinuing the offending medication, and, typically, the addition of preventive treatment 5 . However, withdrawal of the offending medication can be challenging because of the initial increase in pain and other withdrawal side effects 6 .
Cannabis is the most widely used drug in the world, with an estimated 192 million adults who used cannabis globally in 2018 (3.9% of people aged   7 .
Cannabis dependence is the second most common drug use disorder after opioid dependence afflicting 22.1 million people worldwide in 2016. It affects the endocannabinoid system in the brain, which plays a role in pain processing [8][9][10] .
There is moderate evidence to support the use of cannabis or cannabinoids for the treatment of chronic pain in adults [11][12][13] and as opioid-sparing agents [14][15][16] .

Zhang 6
However, other studies show cannabis use to be a risk factor for abuse or dependence of prescription opioids and other drugs [17][18][19] . There is currently limited evidence suggesting that cannabis could be helpful for treatment of migraine 20 . There are no randomized controlled clinical trials that support this hypothesis. However, there is emerging anecdotal clinical evidence that use of cannabis may lead to medication overuse 17,21 . Based on clinical observation, patients with chronic migraine and MOH appear to be concomitantly using cannabis products.
Our study is the first of its kind to assess the risk of MOH in chronic migraine patients who use cannabis. In addition, we sought to utilize unsupervised data reduction methods to explore data-driven as well as clinically meaningful natural clusters of chronic migraine patients based on variables such as MOH comorbidity, use of cannabis and opioids, age, migraine frequency, duration of chronic migraine and MOH. These approaches allowed us to gain in-depth insight into the MOH risk and other related variables associated with cannabis use. Furthermore, we explored whether there was opioid-sparing effect of cannabis use in chronic migraine with or without MOH. The findings contribute to the growing body of work that will help guide physician recommendations for chronic migraine patients who are already self-medicating with cannabis.

Electronic Chart Search
This research used data provided by STARR, "STAnford medicine Research data Repository" (STARR), a clinical data warehouse containing live Epic data Research Repository Cohort Discovery Tool (part of the STARR platform), we created a cohort for the chronic migraine patients. Our search terms included "chronic migraine", "medication overuse", "cannabis", "CBD", "THC", and "marijuana." We reviewed electronic charts of patients seen from January 1, 2015 to January 1, 2019. Two headache specialists (NZ, YWW) reviewed the charts and labeled each chart as included, excluded, or undecided. The undecided charts were reviewed by both authors, and disagreements were resolved by discussion.

Inclusion and Exclusion Criteria
We included adult chronic migraine patients aged 18 and above with a minimum of one-year chronic migraine duration, using and not using cannabis products at time of clinic visit. We excluded children under the age of 18 years, patients with Zhang 8 episodic migraine, chronic migraine shorter than one-year duration, secondary headache disorders (such as post-traumatic headache).

Data Extraction
The data that was extracted included age, sex, diagnosis of chronic migraine, average number of headache days per month during the last 3 months, duration

Ethical approval
The study received full ethics approval from the institutional review board at Stanford University (Protocol 50215). De-identified data was stored securely in Stanford encrypted server. Our study is in accordance with The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement: guidelines for reporting observational studies 22 .

Power Analysis and Sample Size Estimation
Sample size was estimated using logistic regression 2-tailed test whether a dichotomous variable (use or no use of cannabis) was a significant predictor of a binary outcome (presence of absence of MOH), with other covariates (e.g. age, sex, monthly migraine frequency, duration of current CM)at an α error Zhang 9 probability of 0.05, a 1-β error probability of 80% (power), low squared multiple correlation coefficient (R 2 ) of 0.04 between the main predictor (cannabis use) and other covariates, and an estimated 50% proportion of cannabis users. The null hypothesis that probability of MOH when a chronic migraine patient is not using cannabis was estimated to be 0.54 based on the HUNT study 3 . We estimated for our alternative hypothesis that probability of MOH when a chronic migraine patient is using cannabis to be 0.68 (odds ratio = 1.90). Based on these assumptions, there is 80% chance of correctly rejecting the null hypothesis that a particular category of the main predictor variable (using cannabis) is not associated with the value of the outcome variable (MOH), with 345 patients. After adjusting for an assumed 7% of missing data, our final sample size was made 368 patients. Sample size estimation was conducted on G*Power 3.1.9.6 23,24 .

Statistical Analysis
Data were summarized using descriptive statistics. Differences in variables between cases and referents were analyzed using Welch's t-tests and chisquared tests with Yates correction. Logistic regression was employed to identify variables predicting CM patients with MOH (cases) from CM patients without MOH (referents). Predictor variables (age, sex, monthly migraine frequency, duration of current CM, current use of cannabis, opioid, butalbital, benzodiazepine) were tested in one block to determine their predictive capacity by examining their adjusted odds ratio (OR) statistics. Opioid, butalbital, benzodiazepine use was included in predictor variables as these are Zhang 10 dependence causing drugs that are sometimes prescribed in migraine management. The regression model's goodness-of-fit was tested using Cox & Snell R-square 25 , Nagelkerke R-square 26 , and Hosmer and Lemeshow test 27 .
Correlogram based on Spearman rank correlation was used to assess the associations among all variables. Sex (female, male), current use of cannabis (yes, no), opioid, butalbital, and benzodiazepine were dummy coded as "1" and "0" for analysis. Significance threshold was set at p-value of 0.05. Missing data were excluded from analysis.
Additionally, unsupervised data-driven agglomerative hierarchical clustering (AHC) analyses was performed to explore MOH-consistent natural clusters within the total patient population using all predictor variables (age, sex, monthly migraine frequency, duration of current CM, current use of cannabis, opioid, butalbital, benzodiazepine). Clustering analysis was performed using Ward's agglomeration method with Squared Euclidean distance metric as measure of dissimilarity. A dendrogram was created to visualize the AHC clustering and select the major clusters. Statistical analyses were done using Statistical Package for Social Sciences (version 21.0; SPSS Inc, Chicago IL) and XLSTAT 2020 (Addinsoft).

Data Availability Statement
The data that support the findings of this project are available from the contributing author, YWW, upon reasonable request.

Results
Of the 729 charts reviewed, a total of 368 patients (212 cases and 156 referents) were included in the study. The remaining 361 patients were excluded as per our aforementioned exclusion criteria. Clinical characteristics of included patients are displayed in Table 1. There were no statistically significant differences in age, sex ratio, monthly migraine frequency, and duration of current CM. There was 3 times greater number of cases (CM patients with MOH) currently using cannabis than referents (CM patients without MOH) (p = 0.00001; chi-squared test). On average, cases were using cannabis about 4 times longer than referents (19 versus 5 months). In cases and referents, overused acute migraine medications included triptans (sumatriptan, rizatriptan), non-steroidal anti-inflammatory medications (ibuprofen, naproxen, ketorolac), acetaminophen, combination medications (acetaminophen/aspirin/caffeine, acetaminophen/codeine, butalbital/acetaminophen/caffeine, butalbital/aspirin/caffeine), opioid medications (tramadol, oxycodone, oxycodone/acetaminophen, hydrocodone, hydrocodone/acetaminophen, hydromorphone, morphine), and benzodiazepines (alprazolam, lorazepam, clonazepam, diazepam). The median duration of MOH in cases was 2 years. The different forms of cannabis products used by cases included, inhalation products (joints, electronic vaping devices), orally ingested products (cookies, tablets, gummies, tinctures, mints), topical products (oils, ointments, creams, patches). Cases exhibited a significantly higher usage of inhaled and ingested cannabis products compared to referents (Figure 1, Table   1). There was 5% of missing data in the following datasets: duration of current

Discussion
We found that the presence of cannabis use significantly increased the odds of medication overuse headache in patients with chronic migraine. This finding brings up two important questions: 1) Does cannabis use in migraine patients lead to the development of MOH? 2) Can cannabis be used to treat MOH?
The mechanism behind the development and maintenance of MOH is not well understood 28 . Effective migraine therapies that can help reduce the risk of MOH and to treat existing MOH is much needed. A study in rodent model of migraine showed that administration of THC reduces migraine-like pain 29 . Pini et al.
conducted the first and only randomized active-controlled crossover study that evaluated the use of cannabinoids for treatment of medication overuse headache 30 . They found that synthetic cannabinoid nabilone (a CB1 receptor agonist) was more effective than ibuprofen at reducing pain intensity and daily analgesic intake in individuals with medication overuse headache 30 . Using cannabis for the acute treatment of headache brings up the concern of whether cannabis can lead to the development of MOH just as other migraine abortive therapies have the potential of doing so. In a recent study that evaluated cannabinoid receptor agonists in a preclinical model of medication overuse headache, the authors explored whether the exposure of rats to cannabinoids would result in latent trigeminal sensitization and vulnerability to typical migraine triggers 31 . They found that cannabinoid receptor agonists (including delta-9-THC) does produce a state of latent sensitization characterized by increased sensitivity

Zhang 14
to stress, which is a presumed migraine trigger 31 . Based on these results and our findings, we speculate that cannabis consumption leads to increased sensitization that can exacerbate the progression of medication overuse headache.
Our results did not show opioid-sparing effects of cannabis use in chronic migraine patients with or without MOH. In contrast, we found increased association between current cannabis use and current opioid use. However, longitudinal studies will be the appropriate design to examine this relationship so as to explore cause and effect. There is conflicting evidence in opioid-sparing effects of cannabis in the general population. Some cross-sectional studies demonstrate the protective effect of cannabis use from developing opioid dependence 32 and speculate that cannabis may potentiate opioid efficacy leading to reduced opioid dosage 33 . A study in migraine patients has found that 43% patients substituted their opiates/opioids to cannabis 34 . Other prospective studies indicate that cannabis increases the risk of opioid dependence or opioid use disorder 18,35 . Our study showed mild association between current cannabis use and current benzodiazepine use. Another retrospective study has reported increased benzodiazepine discontinuation among mixed cohort of patients with pain and non-pain conditions 36 . In our study, no association was found between current cannabis use and current butalbital use. Our results are the first to explore relationships and MOH risk among these four dependence-causing drugs commonly used by migraine patients i.e. cannabis-opioid, cannabis-

Zhang 15
benzodiazepine, and cannabis-butalbital association. Of these four drugs, cannabis and opioid use significantly contributed to MOH prevalence in chronic migraine patients while adjusting for the other variables. It may be noteworthy to consider that cannabis using CM patients are at increased risk of opioid use, and the consumption of both cannabis and opioid increases the prevalence for MOH. Many migraine patients are already self-medicating with cannabis 34 . In a 2018 electronic survey study on patterns of medicinal cannabis use, 88% of headache sufferers were using cannabis to treat probable migraine 34  That migraine-related patient characteristics were comparable between cases and referents in our study was a strength for our study. In addition, the regression model's satisfactory goodness-of-fit, optimum sample size, and adjustment for potential confounders (age, sex, migraine frequency, chronic migraine duration) are strengths of this study. With the rise of legal consumption of cannabis products, it is important to fully understand risk of MOH in cannabis using migraine patients.
Limitation of this study include the fact that our hospital-based study from a tertiary headache clinic may not be representative of the general chronic migraine population in the community. However, the demographic involving mostly of female middle-aged patients may offer some degree of representativeness to target population of chronic migraine 38 . By virtue of being a retrospective design, our study was limited to challenges which are inherent to retrospective chart studies, e.g. lack of data to determine temporal association between cannabis use and MOH development or dose response between cannabis use and risk of MOH. Additional possible sources of confounders such as psychological comorbidities, alcohol use and cigarette smoking were not consistently available and hence not studied. Likewise, causative analysis cannot be confirmed based on our results. For example, in the absence of psychological

Zhang 17
profiling, the association results from our study might be partly due to dependency-prone personality rather than the biological effects of cannabis or opioids.

Conclusion
Medication overuse headache is significant issue especially among patients with chronic migraine. Our study showed that cannabis use significantly increases the odds of MOH in CM patients.     Correlogram displayed that higher associations were found between cannabis use, MOH, and opioid use. Two major natural clusters were identified. Cluster I (top dendrogram, first left branch) exhibited higher MOH burden, higher cannabis use, higher opioid use, younger age and fewer migraine frequency than cluster II (top dendrogram, first right branch). Figure 1 Cannabis usage types among cases and referents. Cases (blue bar) featured signi cantly higher usage of inhaled and ingested cannabis types compared to referents (orange bar). *** = p-value less than 0.0001, * = p=value less than 0.05. Abbreviations: CM = chronic migraine, MOH = medication overuse headache.  Correlogram of Variables. Correlogram displayed that higher associations were found between cannabis use, MOH, and opioid use.

Figure 4
Heatmap displaying agglomerative hierarchical clustering. Two major natural clusters were identi ed. Cluster I (top dendrogram, rst left branch) exhibited higher MOH burden, higher cannabis use, higher opioid use, younger age and fewer migraine frequency than cluster II (top dendrogram, rst right branch).

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.