This is an open-access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on http://www.jmir.org/, as well as this copyright and license information must be included.
Although population studies have greatly improved our understanding of migraine, they have relied on retrospective self-reports that are subject to memory error and experimenter-induced bias. Furthermore, these studies also lack specifics from the actual time that attacks were occurring, and how patients express and share their ongoing suffering.
As technology and language constantly evolve, so does the way we share our suffering. We sought to evaluate the infodemiology of self-reported migraine headache suffering on Twitter.
Trained observers in an academic setting categorized the meaning of every single “migraine” tweet posted during seven consecutive days. The main outcome measures were prevalence, life-style impact, linguistic, and timeline of actual self-reported migraine headache suffering on Twitter.
From a total of 21,741 migraine tweets collected, only 64.52% (14,028/21,741 collected tweets) were from users reporting their migraine headache attacks in real-time. The remainder of the posts were commercial, re-tweets, general discussion or third person’s migraine, and metaphor. The gender distribution available for the actual migraine posts was 73.47% female (10,306/14,028), 17.40% males (2441/14,028), and 0.01% transgendered (2/14,028). The personal impact of migraine headache was immediate on mood (43.91%, 6159/14,028), productivity at work (3.46%, 486/14,028), social life (3.45%, 484/14,028), and school (2.78%, 390/14,028). The most common migraine descriptor was “Worst” (14.59%, 201/1378) and profanity, the “F-word” (5.3%, 73/1378). The majority of postings occurred in the United States (58.28%, 3413/5856), peaking on weekdays at 10:00h and then gradually again at 22:00h; the weekend had a later morning peak.
Twitter proved to be a powerful source of knowledge for migraine research. The data in this study overlap large-scale epidemiological studies, avoiding memory bias and experimenter-induced error. Furthermore, linguistics of ongoing migraine reports on social media proved to be highly heterogeneous and colloquial in our study, suggesting that current pain questionnaires should undergo constant reformulations to keep up with modernization in the expression of pain suffering in our society. In summary, this study reveals the modern characteristics and broad impact of migraine headache suffering on patients’ lives as it is spontaneously shared via social media.
Migraine affects approximately 12% of adults in the Western world [
Infodemiology is a branch of science that deals with the occurrence, distribution, and analysis of electronic information that is used to inform the public of disease patterns and discourse, and of their relationship to the health status of a population. A key feature of infodemiology is the potential to collect and analyze data in near real time [
Major aims of this study included using social media to assess migraine headache impact in real time to avoid memory bias, and to identify a set of current suffering descriptors that were not prompted by an experimenter. We report that Twitter, used as an instrument for infodemiology [
A continuous cross-sectional sample of 21,741 tweets was collected between Saturday, April 30 (12:00:00 am) and Friday, May 6, 2011 (11:59:59 pm). According to the official Twitter website, a relative number of 1 billion tweets were posted weekly during that season, with an average of 200 million tweets per day from a total of 100 million active users [
This study was certified as exempt from human subjects review by the University of Michigan Committee on Human Research (reference No. HUM00054476).
During the specified seven consecutive days, two investigators alternated in eight-hour shifts compiling all messages posted with the word “migraine” in the Twitter public search engine [
Subsequently, a coding system was used for in-depth interpretation and categorization of each tweet. The categories were: “migraine headache” (a user self-reporting having an actual migraine headache attack), “commercial” (advertising treatments or drugs), “metaphor” (the term migraine was used metaphorically), “not related” (the term migraine does not describe an actual physical experience of headache), “re-tweet” (a re-post of a previous tweet), “third person’s” migraine headache (information is related to another person’s migraine headache attack), general “discussion” (general discussion on migraine), “blanks” (missing data), and “inconclusive” (when not possible to identify the meaning of the word migraine in the tweet). In addition, when available, information about the self-reported migraine headache impact on the users’ sleep, work, social, school, mood, or debilitation was compiled using the same methods described earlier in the text. The following free and public information was also extracted: profile name, gender, and geographic location. All the acquired tweets were automatically translated to English by the Twitter website via Google Translate; however, only tweets originally written in English were used for linguistic analysis to avoid any translation bias.
We analyzed the time/date of occurrence of all the global self-reported migraine headache messages posted on Twitter using Greenwich Mean Time (GMT) for representation of the user time zone, ranging from 0h to 23h. However, the actual geographic location is discretionary information for the user and not always provided. Hence, to achieve a more comprehensive understanding of such temporal behavior, we isolated and then reported times in the United States (the largest representative group) by converting the geographic location time to the appropriate standard time, when available. For instance, an eight o’clock posting from a user on the East Coast in the United States was computed together with a similar posting from another user at local eight o’clock on the West Coast. To make sure every US tweet was corrected to the standard time zone, the US Census Bureau database [
All the data was compiled in Microsoft Excel, which was used to calculate basic descriptive statistics. Frequencies were reported for each category that was collected.
In a systematic manner, three pain specialists oriented and supervised 54 undergraduate students, four graduate students, and six research assistants on the reading, interpretation, and classification of tweets into the criteria described in the Methods section. For each posting, the following was taken into consideration: semantics, users’ demographics, impact of the attacks, geographic location, and time pattern.
Among the non-physical pain categories, advertising treatments or drugs (commercial) were the most prevalent, with 8.99% (1955 out of the 21,741 total tweets collected) prevalence. Re-tweets had a similar prevalence (8.85%, 1923/21,741 tweets), followed by general discussion (6.72%, 1462/21,741 tweets) or third person’s migraine (2.05%, 445/21,741 tweets), and metaphor (1.20%, 261/21,741 tweets). A total of 5.23% of the tweets were inconclusive (1137/21,741 tweets), 1.99% not at all related (434/21,741 tweets), and 0.44% were blanks (96/21,741 tweets). Only 64.52% of all the collected tweets (14,028/21,741) posted using the word “migraine” were an actual self-report of physical pain suffering and other migraine-related symptoms (
Classification of Tweets in Categories. Only 65% (14,028) of the 21,741 tweets were classified as self-reported migraine headache attacks.
Based on the self-reported gender, it was found that 73.47% of the 14,028 migraine headache tweets were from females (10,306 tweets), 17.40% were from males (2441 tweets), 0.01% from transgender (2 tweets), and 9.12% was not provided (1279 tweets). The presence of transgendered in the assessment reflects a new trend in research studies, where the possibility of free self-expression leads to a more accurate gender representation of our modern society and cohort (
Gender distribution disclosed by users who reported their migraine headache attacks (n=14,028). 73.47% female (10,306 subjects), 17.40% male (2441 subjects), 0.01% (2 subjects) self-reported as transgender, and 9.12% was not provided (1279 subjects).
When possible, each interpretation of the tweets was also classified based on their migraine impact. The majority of the actual self-reported migraine headache posts generated an impact on patients’ internal status: personal impact, mostly “mood” with 43.91% (defined as any changes in the natural and emotional state of mind of the individual) (6159 tweets out of the 14,028 migraine tweets), followed by an impact on “sleep” with 5.61% (meaning difficulties falling and staying sleep) (787 tweets). Another percentage related to personal impact included “debilitating” with 3.61% (defined as a physically incapacitating migraine headache) (507 tweets). The external ongoing impact of the self-reported migraine attacks (productivity impact) similarly and instantly affected “work” productivity with 3.46% (impact on work productivity and/or absenteeism) (486 tweets), “social” life with 3.45% (denoting influence and/or absenteeism in current social activities) (484 tweets), and finally “school” with 2.78% (impact on school productivity and/or absenteeism) (390 tweets). Missing data in this category was 37.18% (5215 tweets) (
Impact and expression of migraine headache suffering (n=14,028). Personal impact (Internal) was predominantly on mood (43.91%, 6159 tweets). External impact was on productivity and absenteeism at work (3.46%, 486 tweets), social events (3.45%, 484 tweets), and school (2.78%, 390 tweets).
Based on the physical tweets originally posted in English, we compared the adjectives used to describe a real ongoing migraine attack with the McGill Pain Questionnaire (MPQ) [
Most common pain descriptors used (n=1378). The most frequently used word from the McGill Pain Questionnaire (MPQ) was "horrible" (6.97%, 96 uses). Additional migraine headache adjectives (“Not McGill” words) included "worst" (14.59%, 201 uses) and profanity, the "F-word" (5.30%, 73 uses) being most frequently used.
The collected database is geographically diverse, since it was composed of real-time tweets from all around the world. To better visualize and understand the origin of the messages, we divided the posts based on self-reported geographic location. When in doubt about the precise location, Google Maps [
Percentage of migraine headache tweets by continent (n=5865). Most represented was North America with 65.57% (3840); United States alone represented 58.28% of overall data (3413 tweets).
To have a better understanding of the online temporal behavior of sharing and expression of actual ongoing migraine headache suffering on the database, we organized each tweet based on the time and day when it was posted. As an initial step, we divided each global tweet by the day of the week it was posted. Our analysis demonstrated a higher global prevalence of self-reported migraine attack tweets on Tuesday (2559 of the 14,028 tweets) and Thursday (2155). It was followed by Wednesday (2074), Saturday (1933), Monday (1909), and Sunday (1752). A lower global prevalence of self-reported migraine headache tweets was observed on Friday (1646).
In an effort to improve our temporal data interpretation, the posts were divided according to the reported GMT, since it is the traditional method used by the Twitter website. By plotting the information on a timeline graphic, it was possible to observe a peak of the global migraine headache-related tweets at 14:00 GMT on Monday. This valuable information could easily lead to misleading interpretations if not adjusted for the original standard time of each specific geographic region where the tweets were generated. Consequently, we selected the United States (since it had the majority of tweets) and converted each single tweet to their particular local time. The United States observes DST during the spring season in several regions, which could also lead to erroneous interpretation if not corrected for this particular time when appropriate, during this period of the year. Ultimately, 9:00h and 20:00h on Monday across the United States were the actual peaks of prevalence when most Americans were sharing on social media the occurrence of their migraine headache attacks (
Temporal patterns of migraine headache tweets in the United States. Tweets were converted to local times and corrected for daylight savings time. The averaged flow of migraine tweets accumulated at 10h, persisted and gradually peaked later at night (22h). During Saturday and Sunday (dashed line in the top-left graph), the highest peak occurred at 18h.
An unbiased evaluation of spontaneous sharing and expression of an ongoing migraine headache suffering is crucial for clinicians and researchers to understand the pattern of the disorder, and most importantly, the population under study. Here, we report the use of Twitter as a research tool to assess epidemiology and linguistics of migraine suffering in real-time in our modern society. Our results showed not only a significant overlap with other traditional epidemiologic studies, but also generated unique information about
The methodology used here provided an effective, but laborious and time-consuming approach, to analyze reports of real-time migraine headache attacks on social media. This step was extremely important to avoid any sort of erroneous interpretation in our study. The use of current generalized algorithm search tools available on the Internet to effortlessly analyze altogether a sample of migraine tweets could inevitably lead to misleading conclusions, since they currently struggle to precisely exclude tweets that contain the word “migraine” used for commercial advertisement, general discussion, re-tweets, and metaphors [
In this study, 73.47% of those who self-reported migraine headache were females (10,306 subjects of the 14,028 users who self-reported migraine) and 17.40% were males (2441 subjects). These results demonstrate and reinforce a higher prevalence of migraine headaches among females, which is consistent with population-based studies performed around the world [
Migraine is a disabling form of primary headache [
What inevitably differentiates human from animal pain research is our ability to articulate the suffering experience and how we communicate it constantly evolves. In the case of migraine, the pulsating nature of the attacks has commonly been described in scientific literature as “throbbing”, and is even included in the International Headache Society criteria for migraine diagnosis [
Real-time expression of migraine suffering occurs daily at the global level via social media. It is worth highlighting that it was at 14:00 GMT on the Monday that our planet had the highest flow of migraine headache attack postings on Twitter. The majority of those postings originated from North America, where the United States represented more than half of the total global stream. When each single tweet across the United States was converted to its particular local time, and corrected for DST during spring season when necessary, the largest point prevalence of migraine headache suffering was clustered on Monday with one peak in the morning at 9:00h and another one at 20:00h (
This study showed that the spontaneous flow of communication on Twitter reflects multiple patterns of human interaction for sharing the ongoing suffering experience, and proved to be a rich and instant resource of knowledge regarding the actual impact of migraine attacks in our modern society.
Daylight Savings Time
Greenwich Mean Time
McGill Pain Questionnaire
We thank Ilkka Kristian Martikainen for early assistance in data classification, and the following professors and mentors at University of Michigan School of Dentistry (UMSoD): Robert M Bradley, Paul H Krebsbach, and William V Giannobile for valuable advice and encouragement. We would also like to acknowledge UMSoD (Under)Graduate Class of 2014 for their esteemed contribution as co-authors: A Berger, A Forest, A Green, A Petraszko, A Rohra, A Ruhlig, A Seal, B Corbett, B Jamo, B Kendziorski, B Mansfield, B Powell, D Coviak, D Fujawa, D Hammaker, D Patel, E Mencarelli, I McComb, J Birchmeier, J Harris, J Jeong, J Musselwhite, J Schell, J Sung, K Kilgore, K Swanson, L Bierwirth, L Ehardt, L Harness, L Koch, L Lungu, L McGarvey, L Persichetti, M Gariboy, M Hall, M Irani, M Lala, M Shallal-Ayzin, M Solverson, M Towler, N Poel, O Zorlu, P Dulay, P Iyer, P Kim, P Rutowski, R Abu-Zahra, R Kakar, R Mooar, S Garber, S Nematbakhsh, S Zafar, T Kwun, W Yahn (Undergraduate Students), C McWatters, B Royle, X Tang, R Wiesen (Graduate Students).
This work was supported by the following grants (DaSilva AF): National Institute of Health – National Institute of Neurological Disorders and Stroke – K23 NS062946, Dana Foundation’s Brain and Immuno-Imaging Award, and the Migraine Research Foundation Research Grant Award.
AFD, TDN, and MFD developed the study concept and design. TDN, MFD, MD, HvH, SL, CA, LK, MCB, and UMSoD (Under)Graduate Class of 2014 acquired the data. AFD and TDN analyzed and interpreted the data. AFD, TDN, and TD drafted the manuscript. JKZ, MD, HvH, SL, CA, LK, and MCB were responsible for administrative, technical, or material support. All authors had full access to all the data in the study. AFD is guarantor and takes responsibility for the integrity of the data and the accuracy of the data analysis.
None declared.