Major sex differences in migraine prevalence among occupational categories: a cross-sectional study using UK Biobank

Migraine represents one of the most prevalent neurological conditions worldwide. It is a disabling condition with high impact on the working situation of migraineurs. Interestingly, gender-related differences regarding an association of migraine with important occupational characteristics has been hardly studied. The current study scrutinizes gender-specific differences in the prevalence of migraine across a broad spectrum of occupational categories, shedding also light on associations with important job-related features such as shift work, job satisfaction, and physical activity. The study included data from 415 712 participants from the UK Biobank cohort, using the official ICD10 diagnosis of migraine and other health conditions as selection criteria. Prevalence ratios of migraineurs compared to healthy controls among different occupational categories and job-related variables were estimated using log-binomial regression analyses. Statistical models were adjusted for important sociodemographic features such as age, BMI, ethnicity, education and neuroticism. To better highlight specific differences between men and women we stratified by sex. We detected a differential prevalence pattern in relation to different job categories between men and women. Especially in men, migraine appears to be more prevalent in highly physically demanding occupations (PR 1.38, 95% CI [0.93, 2.04]). Furthermore, migraine is also more prevalent in jobs that frequently involve shift or night shift work compared to healthy controls. Interestingly, this prevalence is especially high in women (shift work PR 1.45, 95% CI [1.14, 1.83], night shift work PR 1.46, 95% CI [0.93, 2.31]). Our results show that higher migraine prevalence is associated with physically demanding jobs and shift working.

Migraine is a highly prevalent neurological disorder that affects predominantly 25 women [1][2][3]. Due to its highly debilitating symptoms, migraine has an important 26 impact on the daily life of people, interfering negatively with private events, such as 27 time spent with friends and family, as well as with the work productivity. In a study 28 published in 2018 [2], it has been estimated that migraine is the cause of 45.1 million 29 risk factors that have been highlighted in recent studies are heavy workloads, emotional 47 stress, shift working and sleep disturbances [8][9][10]. A cross-sectional study [11] showed 48 also that an ethnic minority with high unemployment rate manifested higher migraine 49 diffusion than a Spanish reference population, especially in women. Accounted risk 50 factors are familiar stress, psychological overload and few hours of sleeping. Another 51 evidence of the relationship between migraine and work-related triggers is a longitudinal 52 study [12], in which the cohort comprised employees in a French gas and electricity 53 company. The aim of this study was to examine the trajectory of migraine in relation 54 with retirement. It has been found that retirement is significantly associated with a 55 decrease in headache prevalence, particularly among subjects with high amount of work 56 stress or with stress-prone personality, such as high hostility or type A personality. A 57 Danish cross-sectional study [13] highlighted more practical issues related to work, like 58 the physical work load and the physical activity. In particular, it has been found that 59 migraine is positively associated with seated/standing and walk jobs. It has been 60 detected also an increased risk to develop migraine in women practicing heavy physical 61 work. From this brief review of the literature it emerges that finding possible 62 correlations between migraine pathogenesis and work-related issues is a complex process, 63 since many interconnected factors may increase the risk for the development of this 64 disorder. Previous studies addressed some of these factors, but a more comprehensive 65 analysis is needed. 66 A growing body of evidence suggests that work stress may predict a variety of health 67 problems, such as depression, cardiovascular and muscoloskeletal diseases [5,[14][15][16], but 68 little is known about a possible correlation between migraine and work-related 69 factors [4]. Also, all issues related to work activity are country-specific to a certain 70 extent, and the studies we referred to addressed the association between some 71 work-related issues and migraine, taking into account the cultural and geographical 72 specificity. Therefore, it is important to investigate to what extent these epidemiological 73 findings can be generalized to broader target populations. Also, many studies focused 74 on particular sectors of employment or companies, without addressing the broader issue 75 of the migraine prevalence among the different occupational categories. Another specific and effective preventive strategies, lowering the impact of migraine on society. 82 The aim of this study is therefore to investigate the gender-specific association between 83 occupational categories and work-related features such as physical activity involved, 84 shift work, job satisfaction and migraine, using the large population-based UK Biobank 85 cohort. involves mainly walking or standing, whether it involves heavy manual or physical work, 109 or whether it involves shift or night shift work. To further address the work-related 110 individual situation, we took into account variables that describe work/job satisfaction. 111 For our project we used the values of variables at the baseline, i.e. the first visit of the 112 participants at the centers for the collection of data.

113
Covariates 114 Several basic demographic characteristics were considered as relevant covariates as 115 described in the following. Information about the sex has been acquired from central 116 registry at recruitment, and in some cases it has been updated by the participants.

117
Ethnic background and educational status have been self-reported by the participants in 118 a touch-screen questionnaire. The vast majority of participants was white British, 119 therefore the participants were divided into two categories, i.e. "White" and "Others". 120 Regarding the educational status, we distinguished just between with or without 121 University/College degree, since in previous studies it has been shown that higher 122 education acts as a protective factor for migraine [17]. Furthermore, we considered the 123 age at recruitment, i.e. the age of the participant when attended an Initial Assessment 124 Centre visit. Since weight can exert an influence on both migraine onset and job-related 125 issues, the BMI was also considered, calculated from height and weight measured during 126

3/24
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 29, 2021.  neuroticism. Work perception may be influenced by personality traits. As highlighted in 128 previous studies [18,19], negative personality traits have great importance when tackling 129 work-related issues and neurologic disorders, and for this reason it has been taken 130 neuroticism into account.

131
Covariates such as ethnic background, educational status, age, BMI and neuroticism 132 are associated both with the exposure (job and job-related issues) and the outcome 133 (migraine), therefore they were considered confounders. We adjusted for them in our   Those participants were excluded as well, as no sound conclusion could be drawn while 143 including them in the analyses. In any part of the flow chart in which the number of 144 excluded participants was reported, the amount of participants that refused to answer 145 or that were not sure about their own health condition have been included as well.

146
The UK Biobank provides information from the hospital registers, ensuring that 147 disorders or illnesses have been at some point diagnosed by a physician. We used the 148

4/24
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 29, 2021

5/24
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 29, 2021. ; https://doi.org/10.1101/2021.07.27.21261191 doi: medRxiv preprint Statistical analysis 165 In order to investigate the prevalence of migraineurs in the various job categories as 166 compared to healthy controls, we calculated the prevalence ratio (PR) with 95% 167 confidence interval (CI). We performed the calculations for PR and 95% CI using two 168 models: (i) adjusting for sociodemographic confounders such as age, BMI, ethnic 169 background (ii) as model (i) including education level and neuroticism as well [21]. The 170 analysis has been stratified by sex, to further investigate the differences between men 171 and women. We performed log-binomial regression analysis for the two models [22,23]. 172 In three cases it was not possible to perform the analysis for the first model: 173 occupational category and heavy manual or physical work for women and shift work for 174 men. Convergence criteria were not met as the Hessian matrix was singular. In these 175 three cases we added the variable neuroticism as predictor in the model. Only in these 176 three cases the first model had this change, while the fully adjusted model is equal to 177 the others.

178
As recommended by leading epidemiologists and statisticians, a thorough discussion 179 of effect sizes and confidence intervals has been provided, as the most solid way to draw 180 sensible conclusion from statistical analyses combined with biological scientific 181 background. For these reasons, no statistical test has been performed nor any threshold 182 of statistical significance has been established. P-values have not been provided as 183 possibly misleading [24][25][26][27][28][29][30][31][32]. Accordingly, any finding deemed as significant is intended 184 to be clinically or epidemiologically (and not statistically) significant.

185
The statistical analysis has been conducted via SPSS (IBM SPSS Statistics version 186 26). Plots and diagrams have been created using GraphPad Prism 9.12 and L A T E X.

188
The flow chart in Figure 2 summarizes the selection process of individuals finally 189 included in the analysis. From the initial sample population of 502 488 participants, 86 190 776 subjects were excluded (17.3% of the initial sample) as they have been diagnosed by 191 a disease or disorder as listed in the exclusion criteria or because they refused to report 192 their health condition. The final sample was divided into two main categories, i.e. 193 migraineurs and controls. Table 1 shows the main demographic and socioeconomic 194 characteristics of the final sample population. Analysis of data reported in Table 1 195 allows detecting interesting features. Migraine is more prevalent in women rather than 196 in men, with a 3:1 ratio, in line with what is already known from an epidemiological 197 perspective [1]. The neuroticism score on average is higher in migraineurs as compared 198 with controls. Subjects with high neuroticism score tend to respond worse to stressors 199 and this feature may play an important role in migraine onset, as stress and poor stress 200 response are common risk factors for this neurologic disorder [17]. In general, we can 201 also observe that this sample population is constituted mostly by middle aged, slightly 202 overweight, white British people, whereby variables such age, BMI and education level 203 are well balanced in the study arms.
204 Table 2 reports the descriptive statistics of job-related outcome variables. Since not 205 all the selected participants underwent all the various questionnaires, at the beginning 206 of each main category it has been reported the overall number of migraineurs and 207 controls who actually answered that particular job-related questionnaires. Data on 208 occupational categories are available from 1 569 migraineurs and 277 241 controls.

209
Female migraineurs appear to be more prevalent in associate professional and technical 210 occupations, personal services occupations and sales and customer service occupations. 211 In contrast to this, male migraineurs appear to be less prevalent in categories such as 212 managers and senior officials and professional occupations, while they appear to be 213 6/24 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 29, 2021. ; To assess the influence of job-related physical strain on migraine, we considered the 216 variables in our analyses that described the frequency of heavy manual or physical work 217 as well as the intensity of walking or standing during work. In the first case, 1 410  is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 29, 2021. ;  Table 3 and Figure 3 display prevalence ratios (PR) and corresponding 95% confidence 229 intervals (95% CI) for the different occupational categories, as classified according to 230 SOC2000, and stratified according to the sex of the participants. Although point 231 estimates of the PRs are limited in the precision because of wide 95% CI (see Figure 3) Figure 3. Forest plot illustrating the prevalence of migraine among occupational categories. Means with error bars plot of the data from Table 3. The point estimate is the prevalence ratio (fully adjusted model) of migraineurs compared with controls in the different occupational sectors as defined by SOC2000.

9/24
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 29, 2021. ; . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 29, 2021.

11/24
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 29, 2021.   Table 4. The point estimate is the prevalence ratio (fully adjusted model) of migraineurs compared with controls among job categories, differentiated by the relevance and frequency of (a) heavy manual or physical work and (b) walking or standing.

12/24
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 29, 2021. ; . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 29, 2021. ; is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 29, 2021.

15/24
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 29, 2021. ; https://doi.org/10.1101/2021.07.27.21261191 doi: medRxiv preprint . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 29, 2021. ; https://doi.org/10.1101/2021.07.27.21261191 doi: medRxiv preprint Table 6 and Figure 6 display the outcomes of our analyses on the relationship 252 between prevalence of migraine and job satisfaction. Participants were asked to rate 253 how satisfied they were with the work they were doing. The general trend for both man 254 and women is that migraine is less prevalent among participants happier with their job. 255 In particular, migraine appears to be less prevalent among women moderately happy 256 with their job (fully adjusted PR 0.76, 95% CI [0.52, 1.12]). Men belonging to the 257 "extremely unhappy" group show higher prevalence of migraine, but being the 95% CI 258 very wide it is difficult to draw a solid conclusion on the actual size of this effect. 259 Figure 6. Forest plot illustrating the prevalence of migraine among participants divided by different degrees of job satisfaction. Mean with error bars plots of the data from table 6. The point estimate is the prevalence ratio (fully adjusted model) of migraineurs compared with controls among different levels of job satisfaction.

17/24
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 29, 2021. ; To our knowledge, this is the first study that thoroughly addresses the prevalence of 261 migraine in a broader spectrum of occupational categories. While other studies focused 262 only on a very particular work sector (e.g. health professionals) [5][6][7][8][9][10][11][12], we considered 263 nine different job categories (according to SOC2000) as well as important job related 264 features such as work-related physical activity, shift work and job satisfaction. Furthermore, stratifying by sex we were able to perform more precise analyses, 266 highlighting sex-specific differences on the prevalence pattern of migraine among the 267 various job-related features. The fact that we considered official diagnosis from hospital 268 registry is another important feature of this study, as it is a more solid and trustworthy 269 way to divide patients and controls as compared with analyses based on self-reported 270 diagnosis.

271
Our results show striking differences between men and women in the prevalence of physically demanding. Thus, both may be a significant source of stress, which is a risk 283 factor for migraine [3,4,20]. The fact the migraine is not highly prevalent in these 284 groups is probably due to the fact that these types of job are also characterized by a 285 high level of control and regulation. This may act as a protective factor against the 286 onset of migraine, as it has been shown in a previous study that low control in the work 287 environment is a risk factor [6]. This argument is supported by our analyses on work 288 satisfaction, showing that a good work environment is inversely associated with 289 migraine.

290
For men the situation appears to be the opposite. Physically demanding occupations 291 are characterized by a higher prevalence of migraineurs, while the others, more 292 psychologically demanding, have a lower prevalence. This is also confirmed by the 293 analyses on work-related physical activity, as discussed below. In particular, managers 294 and senior, professional, associate professional and technical occupations are 295 characterized by very low prevalence of migraine. These occupational categories 296 comprise many different types of job, such as managers in all sectors, science and 297 technology, health, teaching and research, business and public service professionals as 298 well as assistants and technical occupations related to these areas. These jobs require 299 highly educated and skilled workers. Thus, it can be hypothesized that they may have a 300 better access to healthier environments and therapeutic interventions that can improve 301 the health in general and reduce the risk of migraine. However, this is also true for 302 women, so this cannot explain the asymmetry between the prevalence of migraine for 303 men and women in this various job categories. Additional sex-specific mechanisms may 304 act to account for this asymmetry. With the data available for this research we were not 305 able to further address this issue.

306
The analyses on the work-related physical activity show that frequent heavy manual 307 physical work is related to higher prevalence of migraine among men, while for women 308 there is no significant difference compared with the healthy controls. This interaction 309 between migraine and physical activity is also supported by the analyses on jobs that 310 involve mainly walking or standing. Higher prevalence of migraine has been reported 311

19/24
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 29, 2021. ; https://doi.org/10.1101/2021.07.27.21261191 doi: medRxiv preprint among men and women for whom it's required to walk or stand for many hours and 312 often at their work place. There is no conclusive evidence in the scientific literature 313 about the impact of physical exercise on migraine. Is it a risk or a protective factor? A 314 recent review [31] addressed the issue of the relation between migraine and physical 315 exercise. There is a limited amount of evidence that proves that intensive physical 316 exercise may trigger migraine attacks, however there are extensive studies that prove 317 that regular aerobic exercise may be an effective preventive strategy against migraine. 318 In general, the review reported studies on regular and intensive physical exercise, not 319 related to the work place, where the intensity of the physical activity and the duration 320 may vary consistently. It could be possible that these irregular working rhythms are the 321 source of a kind of stress that may be at the origin on migraine onset for men. Another 322 feature related to these elementary occupations that should be taken into account is the 323 general lifestyle of these workers, as many habits and sociodemographic factors can 324 increase the migraine prevalence (e.g. obesity, caffeine consumption among others) [3]. 325 The level of the impact of shift work on migraine has not been fully established, as 326 the available data are few and conflicting [9,32]. One key argument in favor of the 327 possible migraine risk due to shift work is the variation of the circadian rhythm. Shift 328 and night shift work are characterized by irregular schedule and this affects negatively 329 the sleep-wake balance and thus making irregular the circadian rhythm. This may be at 330 the origin of migraine pathogenesis. Our analyses show that migraine is more prevalent 331 in jobs characterized by frequent shift and night shift work. In particular, the prevalence 332 is higher for women and for night shift work, supporting the sleep-wake imbalance 333 theory as the possible explanation of the causing factor of migraine. Not many 334 participants answered the shift work questionnaires, thus our analyses are characterized 335 by lower precision, due to the smaller sample sizes, as compared with previous analyses. 336 Some limitations of this study derives from some features of the UK Biobank itself. 337 The sample population is characterized by older people (the mean is around 56 with a 338 standard deviation of 8 years), while migraine is mostly prevalent in people between 35 339 and 39 [1]. Also, not all the participants answered all the work-related questionnaires, 340 and this forced us to perform analyses with different degrees of precision, due to the 341 variation in sample sizes. Another limitation is due to the fact that there is no data 342 available on when the participants were actually diagnosed with migraine (or any other 343 condition considered for the selection of the participants). In particular, some 344 participants may have been not affected by migraine when their data have been 345 collected. One last limitation of this project is its cross-sectional nature. This allowed us 346 to give an estimation of the prevalence of migraine among different work-related groups, 347 but not to assess risk or incidence. Any kind of hypothesis about the possible underlying 348 causative path between the examined exposures and the onset of migraine in this study 349 is purely speculative, even though supported by parallel and independent evidence from 350 other studies. Nonetheless, the study of the prevalence of migraine among the different 351 groups here considered is a powerful tool that can be used by policy makers and health 352 institutions to improve the work environment and therefore the health of the workers.

354
Previous studies showed how particular job-related features exert a differential impact 355 on women and men in terms of migraine pathogenesis. This research shows that 356 migraine is more prevalent in physically demanding occupations, especially in men.

357
Migraine is also more prevalent in jobs that involve frequently shift or night shift work, 358 and the prevalence is even higher in women. Study results support the assumption of 359 gender-specific differences in work-related stressors associated to migraine. 360

20/24
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 29, 2021. ; https://doi.org/10.1101/2021.07.27.21261191 doi: medRxiv preprint