Effectiveness of Oxygen and Other Acute Treatments for Cluster Headache

Results From the Cluster Headache Questionnaire, an International Survey

Stuart M. Pearson, MA; Mark J. Burish, MD, PhD; Robert E. Shapiro, MD, PhD; Yuanqing Yan, PhD; Larry I. Schor, PhD


Headache. 2019;59(2):235-249. 

In This Article


The Cluster Headache Questionnaire (CHQ) is a self-administered internet-based survey conceived and constructed by authors SMP and LIS, with authors MJB and RES asked to provide input as neurologists and assist in analysis and interpretation, and author YY asked to provide statistical analysis. The CHQ consists of 152 items organized into 8 separate sections: (1) Sign up and Verification; (2) Symptom Screening; (3) Demographics; (4) Experience; (5) Medications/Treatment; (6) Beck Depression Inventory; (7) Hopelessness Depression Symptom Questionnaire; and (8) End of Survey – Contact Options. Sections 1–5 were newly created by the authors and were tested on 10 cluster headache respondents and reviewed by 1 neurologist prior to release of the final version; however, these questions were not otherwise validated. The scope of this manuscript focuses primarily on "Medications/Treatment" in Section 5. Several of the questions for Sections 2 and 5 are shown in Supplemental Figure 1. The study was performed in Qualtrics, which enabled the authors to construct, distribute, collect, and securely store responses. Qualtrics is an online survey company that provides web-based survey software, encrypted cloud-based data storage, and controlled user access. The University of West Georgia holds a Qualtrics software license and datacenter.

Informed Consent. Respondents were given a summary of the intent and purpose of the research as well as a brief summary of each section. In the frame following, respondents were required to verify their age (18 years or older) as well as agree to participate in the survey. Due to concerns about suicidality, international suicide prevention resources were embedded in the CHQ and respondents could skip some questions that were deemed to be potential triggers. At the end of the survey, respondents could elect to share their contact information for further follow-up, but this was not required.

Distribution and Data Collection. Recruitment, distribution, and data collection consisted of 3 concurrent efforts: direct email through the Clusterbusters member listserv, web-site hosting through Clusterbusters and the International Headache Society, and advertising on Google via Google AdWords as well as Reddit forum. Clusterbusters is a nonprofit organization with a mission statement that includes "research, education, support, and advocacy related to cluster headache"[11] and thus their website would be expected to select for respondents interested in more information on several aspects of cluster headache. The survey was open without a password. It was voluntary and accessible internationally by anyone with internet access; however, the survey utilized cookies and recorded IP addresses to identify unique survey respondents and prevent multiple submissions. The survey was presented in the same order to all respondents and displayed a progress bar; respondents could use a "back button" to review their responses before submission, and progress was saved (based on IP address) allowing respondents to close their browser or navigate away. No incentives were offered for taking the survey. IRB approval was obtained in January 2016 from the University of West Georgia, the survey was piloted in February 2016 with 10 cluster headache respondents and a neurologist, and the survey was open online from March 2016 to April 2018.

Participants. For inclusion, participants must have: (1) stated that they were at least 18 years of age; (2) stated that they had been diagnosed with cluster headache by a medical professional; (3) completed at least 90% of the survey including all inclusion/exclusion questions; and (4) filled out the English version (other versions were generated in Google translate but have not been fully verified by native speakers). For exclusion, participants answered several questions that addressed the full International Classification of Headache Disorders (ICHD) 3-beta criteria for cluster headache and probable cluster headache,[12] including all autonomic features except for rhinorrhea, and all other criteria except criterion E ("not better accounted for by another ICHD-3-beta diagnosis"). Because we asked respondents about their longest period of remission in the last year, the definitions of episodic and chronic cluster headache reflect the new ICHD-3 criteria that were released during this study (ie, 3 months of headache freedom for episodic cluster headache).[1] Chronic cluster headache was defined as a remission period lasting less than 3 months in the last year; episodic cluster headache was defined as all respondents who stated that they were episodic, as well as all respondents who stated they were chronic but the headache remission period was 3 months or longer. Authors MJB and RES reviewed these questions and excluded all respondents who did not meet the criteria for cluster headache or probable cluster headache. However, the authors did not corroborate a formal clinical diagnosis of cluster headache. The diagnoses of cluster headache and probable cluster headache were never combined in the analysis and were always examined independently.

Development. Qualtrics provides an adaptive/responsive display framework in conjunction with response validations, thus not all respondents received all 152 questions. Questions were grouped on each screen so that an individual screen could contain between 1 and 30 questions. At a minimum, the survey prompted and required all respondents to answer "yes," "no," or "decline to answer" for each treatment subcategory. Section 5 (Medications/Treatment) divided interventions into 4 subsections: preventive medications, abortive medications, unregulated treatments, and surgical/neuromodulation treatments. For the abortive medication section, the first question displayed a list of common abortive medications in "Check Box" format: (1) Triptans; (2) 100% Oxygen; (3) Cafergot/Ergotamine; (4) Intranasal Ketamine; (5) Lidocaine Nasal Drops; (6) DHE–IV (Migranal); (7) Intranasal Capsaicin; and (8) Opiates, as well as 3 additional "other" boxes for respondent write-in (see Supplemental Figure 1). Questions about triptans referred not to specific medications or specific routes of delivery but to the class of triptans (almotriptan, eletriptan, frovatriptan, naratriptan, rizatriptan, sumatriptan, and zolmitriptan). Questions about oxygen mentioned 100% oxygen but did not specify a flow rate, flow duration, or type of delivery mask or cannula used. For the purposes of the survey, oxygen was included as a "medication." Intranasal capsaicin was included as an abortive despite being recommended as a preventive medication that may take several days for effect[13] because some patients have found immediate relief and use it in an abortive fashion. Questions for each medication were asked as follows:

  1. Effectiveness: This question carried forward and displayed only the previously checked medications. Respondents then evaluated the effectiveness of each medication as: (1) completely ineffective; (2) minimally effective; (3) somewhat effective; (4) very effective; and (5) completely effective. No further explanation or definitions of these choices were provided.

  2. Access: This question asked respondents "how difficult it was to obtain" each medication as: (1) no difficulty; (2) slight difficulty; (3) some difficulty; (4) extreme difficulty; and (5) unable to get. No further explanation or definitions of these choices were provided.

  3. Adverse events: Respondents were subsequently asked to evaluate the psychological or emotional complications as well as the physical or medical complications of each medication in 2 separate questions. For both questions, respondents chose from (1) none; (2) minimal complications; (3) some complications; or (4) severe complications. These questions also displayed text/write-in option allowing respondents to describe if they marked "severe" complications. No further explanation or definitions of these choices were provided.

One category of abortive medications – caffeine and energy drinks – was created after the survey was closed because a free text box was allowed and there were a high number of entries for caffeine, coffee, espresso, and energy drinks. We excluded combination medications such as caffeine plus aspirin or acetaminophen plus caffeine plus butalbital. While the study did ask about abortive neuromodulation devices such as sphenopalatine ganglion stimulation and vagus nerve stimulation as acute treatments, there were less than 25 responses for each and the type of device used could not be verified, thus they were not included in the analysis.

Statistical analysis: All statistical analyses were performed in R, version 3.4.2 (www.r-project.org). For categorical variables, a multinomial test was used to test the null hypothesis of equal value. To test the relationship between nominal and ordinal variables, we calculated Freeman's Theta and performed a Cochran–Armitage test for trend. A Bonferroni correction was used to adjust P values, and adjusted P values less than .05 were considered statistically significant. A separate Bonferroni correction was calculated for each analysis (ie, a specific Bonferroni correction was calculated for each figure). Two-tailed tests were used throughout the study.

To compare medications, we reduced the Likert scales from 5 categories to 2 categories to increase statistical strength, limited our analysis to the 3 most commonly used treatments in the survey (oxygen, triptans, and opioids), then performed a generalized linear mixed-effects model using these 3 medications as the predictors. The significance was then evaluated by likelihood ratio tests; post hoc pairwise comparisons of oxygen, triptans, and opioids were examined using Tukey's method through the "lsmeans" package in R version 3.4.2.[14] The categories were reduced from 5 to 2 before analyzing the data after discussion between authors. The reduced categories were as follows: (1) high effectiveness (completely effective and very effective) vs low effectiveness (somewhat effective, minimally effective, and completely ineffective); and (2) high complications (severe complications and some complications) vs low complications (minimal complications and no complications).

No statistical calculation of power was performed prior to the study. Sample size was based on a previous study.[15] Due to the rarity of the disorder, the study was open for 2 years to ensure sufficient time for the widest reach internationally. In this study, we analyzed 2 subsets of respondents: (1) probable cluster headache, decided after the survey started but before analysis began, and (2) respondents 65 years and older, a subset decided before the survey began in part to investigate the difference in insurance coverage in this age group. All other analyses were planned either before the survey began (by authors SMP and LIS) or after the survey started but before analysis began (by authors MJB, RES, and YY).

Missing data are as follows. For cluster headache respondents, 8 respondents did not provide an answer for chronic vs episodic (n = 1596), 3 respondents did not answer duration of headache (n = 1601), and 1 respondent did not answer the Beck's Depression Inventory II (n = 1603). There were several respondents that answered questions on complications and access to medications that did not answer the question about effectiveness, suggesting missing data on effectiveness for 25 triptans, 11 oxygen, 6 dihydroergotamine, 24 cafergot/ergotamine, 3 ketamine, 8 opioid, 8 capsaicin, 3 caffeine and energy drinks, and 16 lidocaine. The full range of missing data for medications, however, is unknown: in our survey design, a blank response could mean the respondent did not try a medication, but could also mean that they forgot that they tried a medication. For probable cluster headache respondents, 1 respondent did not provide an answer for age of onset (n = 588), and 2 respondents did not provide an answer for chronic vs episodic (n = 587).