Physical inactivity and headache disorders: Cross-sectional analysis in the Brazilian Longitudinal Study of Adult Health (ELSA-Brasil)

Study design and population

This is a cross-sectional analysis of the baseline data from the Brazilian Longitudinal Study of Adult Health (ELSA-Brasil). The ELSA-Brasil is a prospective multicenter cohort study, which recruited 15,105 civil servants from six capitals from three regions of Brazil (São Paulo, Rio de Janeiro, Belo Horizonte, Salvador, Porto Alegre, and Vitória). Baseline data were retrieved from workplace-based interviews and clinic visits for biochemical sampling and assessments, conducted between August 2008 and December 2010.

Inclusion criteria were active or retired employees from six public institutions, aged between 35 and 74 years. Exclusion criteria were current or recent pregnancy (<4 months before the interview), intention to quit working at the institution, severe cognitive or communication impairment, and living outside of a study center’s corresponding metropolitan area (for retired participants). The ELSA-Brasil cohort population’s sociodemographic characteristics showed higher monthly income, higher educational levels, and more access to health care than the Brazilian Census’ population (16). Nevertheless, ELSA-Brasil’s population included participants within a range of occupations classified as unskilled, technical/clerical and faculty and professional staff, which allowed us to stratify socioeconomic levels across the sample.

Approvals from all institutional review boards of the institutions involved in the ELSA-Brasil study and signed informed consent was provided. Further information regarding ELSA-Brasil study’s design is detailed elsewhere (16,17).

Outcomes

The primary outcome was the association between headache disorders and physical inactivity in the LTPA and CPA domains separately and combined. The association between physical inactivity and headache attack frequency was set as a secondary outcome.

Headache disorders diagnosis

All participants in the study (n = 15,105) who answered “yes” to the question “In the last 12 months, did you have a headache?” at the ELSA-Brasil baseline evaluation were invited to answer a detailed headache questionnaire based on the International Classification of Headache Disorders (ICHD 2nd edition (18)), which has been validated and previously used in Brazil (19). Briefly, it investigates pain frequency, duration, quality, location, intensity, triggering factors and accompanying symptoms, such as nausea or vomiting. We classified individuals who answered “yes” to the question about headache and fulfilled all criteria for migraine as “definite migraine”. We classified individuals who answered “yes” to the question about headache and fulfilled all criteria for migraine but one as “probable migraine”. We classified individuals who answered “yes” to the question about headache and fulfilled all criteria for TTH as “definite TTH”. We classified individuals who answered “yes” to the question about headache and fulfilled all criteria but one for TTH as “probable TTH”. Other headaches were defined as headaches that did not fulfil criteria for primary headaches. Participants experiencing no headache in the past 12 months were classified as the “no headache” group.

Since the diagnoses of headaches were not mutually exclusive, when we had the possibility of two types of main primary headaches (TTH or migraine), we always chose that classified as definite. In cases in which both probable TTH and migraine occurred, we always considered migraine as the main type due its relevance to PA (20).

The data on headache attack frequency was collected through a closed-ended question with the following response options: “once in a while”, “1 to 2 per month”, “once a week”, “more than once a week”, and “daily”.

Physical activity levels

In the baseline wave of the ELSA-Brasil study, only PA levels in the LTPA and CPA domains were collected. The PA data were retrieved using the international Physical Activity Questionnaire (IPAQ) long-form, which has been translated and validated for the Brazilian population (21). In the IPAQ, LTPA refers to PA of moderate and vigorous intensity related to recreation, sport, exercise, or leisure. CPA only considers moderate bicycling and walking related to travel to and from work, to do errands, or to go from place to place (21). Physical activity levels in the LTPA and CPA domains were analysed pooled together (combined LTPA + CPA) and separately to provide specific information on each PA domain. PA levels were computed by multiplying the weekly frequency (number of days) by the duration (minutes per day) of PA performed. Only physical activities performed for at least 10 min per week (min.wk⁻¹) were computed. PA level definitions within the LTPA, CPA, and combined (LTPA + CPA) domains, were set based on the WHO’s PA guidelines for adult’s health. That is, to be considered physically active, one should accumulate the minimum amount of 150 min.wk⁻¹ of moderate or 75 min.wk⁻¹ of vigorous PA (10). Because the current WHO PA guidelines are more inclusive and recognize that PA levels under 150 min.wk⁻¹ are still better than none (10), we set as “somewhat active” those who did not meet this minimum recommended amount. Thus, PA levels in every PA domain were categorized into three levels: “inactive”, for those not performing the minimum accountable amount of weekly PA (i.e. <10 min.wk⁻¹); “somewhat active”, for participants reporting 10–149 min.wk⁻¹ of moderate or 10–74 min.wk⁻¹ of vigorous PA; and “active”, for participants engaging in ≥150 min.wk⁻¹ of moderate and/or ≥75 min.wk⁻¹ of vigorous PA.

For the LTPA domain, the frequency and duration of both moderate and vigorous PA intensities were computed and categorized separately. Regarding LTPA intensity, the same PA levels categories were set for moderate (active: ≥150 min.wk ⁻¹; somewhat active: 10–149 min.wk⁻¹; inactive: <10 min.wk⁻¹) and vigorous PA intensities (active: ≥75 min.wk⁻¹; somewhat active: 10–74 min.wk ⁻¹); inactive: <10 min.wk⁻¹).

Confounder variables

Our adjusted analyses included sociodemographic variables (age, sex, household income, educational level, ethnicity, and marital status); cardiovascular risk variables (hypertension, diabetes, metabolic syndrome (MS), obesity, and smoking); psychiatric variables (depression and generalized anxiety disorder (GAD)); and medications for migraine prophylaxis. The cardiovascular risk profile of our sample was obtained through standardized anthropometric and laboratory procedures and testing. We used data on blood pressure, fasting glycemia, total cholesterol and fractions (i.e. LDL, HDL), triglycerides, glycosylated haemoglobin, insulin, HOMA-IR index, and smoking status. Hypertension diagnosis was based on previous clinical history and/or systolic blood pressure ≥140 mmHg and/or diastolic blood pressure ≥90 mmHg and/or use of medication to treat hypertension. Diabetes was diagnosed based on previous medical history of diabetes and/or use of medication to treat diabetes, and/or a fasting plasma glucose ≥126 mg/dl, and/or a 2-h plasma glucose ≥200 mg/dl), and/or a haemoglobin A1C (HbA1C) ≥6.5%. The homeostasis model assessment for insulin (HOMA-IR) was estimated for insulin resistance based on the formula: Fasting glucose (mg/dl) × insulin (mIU/ml)/405 HOMA-IR values. Dyslipidaemia and metabolic syndrome diagnoses were defined according to National Cholesterol Program-Adult Treatment Panel III (NCEP ATP III) criteria (22); that is, LDL cholesterol ≥130 mg/dl and/or lipid-lowering drugs, and three criteria: Waist measurement >88 cm for women or >102 cm for men, HDL cholesterol <50 mg/dl for women or < 40 mg/dl for men, an SBP ≥130 mmHg or ≥85 mmHg, serum triglyceride levels ≥150 mg/dl, and fasting plasma glucose ≥110 mg/dl. Obesity was defined according to WHO criteria based on body mass index (BMI): >30 kg/m²) (23).

Psychiatric comorbidity diagnoses were determined by an adapted Brazilian-Portuguese version of the Clinical Interview Schedule – Revised (CIS-R), applied by trained interviewers (24). The CIS-R is a structured interview for measurement and diagnosis of non-psychotic psychiatric morbidity in the community and a suitable instrument to be adopted in epidemiological studies (16,17).

Regarding migraine prophylactic medications, we included medications with evidence level A and B classed by the American Academy of Neurology’s guidelines (25). Level A: Antiepileptic drugs (divalproex sodium, sodium valproate, topiramate), beta-blockers (propranolol, metoprolol, timolol); Level B: Antidepressants (amitriptyline, venlafaxine) and beta-blockers (atenolol, nadolol).

Statistical analysis

Descriptive statistics of socioeconomic, cardiovascular risk profile, and physical activity data are reported as mean with standard deviation (SD) or 95% confidence interval, median with interquartile range (IQR), or proportion, according to variable features. The Clopper–Pearson method was adopted for calculating the 1-year prevalence estimates of any headache disorder and each type of headache in the overall sample.

Comparisons between headache disorder subtypes and no headache groups for variables of interest were performed by χ² test (categorical variables), one-way ANOVA, or Kruskal Wallis (continuous variables), depending on sample distribution characteristics.

Univariate general linear models (GLM) and multinomial logistic regressions analyses were conducted to explore the associations between PA levels and headache disorders. These analyses included crude and multivariable-adjusted models, which controlled for the effects of confounder variables: Sociodemographic variables (age, sex, household income, educational level, and ethnicity); cardiovascular risk variables (hypertension, diabetes, MS, obesity, and smoking); psychiatric comorbidities (depression and GAD), and use of migraine prophylactic medication.

A series of GLMs for each PA domain (i.e. LTPA, CPA, and combined LTPA + CPA) were performed for testing the associations between PA levels in minutes per week (continuous dependent variable) and headache disorders (independent variable), controlled for the effects of confounder variables. Separated GLM analyses tested the sex interaction effects in the associations between PA levels and headache disorders, adjusted for confounder variables. One-way ANOVA contrasts were computed to test for a linear trend in PA levels (in all domains) across the headache attack frequency groups (“once in a while”, “1 to 2 per month”, “once a week”, “more than once a week”, and “daily”).

Multinomial logistic regression models computed the odds ratio for associations between headache diagnosis groups (“no headache” set as the reference group) and categorical PA levels “inactive”, “somewhat active”, and “active” (“active” group set as the reference group) for each PA domain (LTPA, CPA, and combined LTPA + CPA). In the LTPA domain, separate logistic regression models computed the odds for headache disorders according with moderate and vigorous PA intensities (“active” as reference group). For the associations between categorical PA levels and headache attack frequency, “once in a while” was set as the reference group.

Logistic regression models stratified by sex were performed to evaluate sex patterns for the associations between PA levels and headache disorders. In these later analyses, the same confounders variables, except sex, were also included in the adjusted models.

Finally, to investigate the effect of retirees’ data in the context of the association between headaches and CPA, we performed a sensitivity analysis excluding retired participants (n = 1426) to check whether the associations between headache disorders and PA levels in the CPA alone and combined CPA + LTPA domains would change.

Missing data for the primary outcome variables represented < 2.8% of the sample at most, thus was within acceptable limits (<5%). For all tests, a p-value < 0.05 was considered statistically significant. Statistics were computed by SPSS software (IBM SPSS Statistics for Windows, Version 20.0.; Armonk, NY).

LTPA and primary headache disorders

The data on PA levels in the LTPA domain suggest that physical inactivity could be a risk factor for major primary headache disorders such as migraine and TTH. The strong trend for the association between PA levels and headache attack frequency regardless of headache subtype suggests that physical inactivity could also participate in the chronification processes of headache disorders. Alternatively, recurrent pain itself may rather constitute a barrier to PA in leisure time, a plausible hypothesis supported by evidence from another population-based study (3), as well as smaller observational studies in patients with migraine (26), TTH (27), and with both conditions (28).

Regarding migraine, on the one hand the more consistent associations between migraine/probable migraine and LTPA may suggest that people with migraine may avoid PA, especially of vigorous intensity. Vigorous exercise is a common migraine attack trigger (29–31) and headache aggravated or provoked by routine physical effort is a “canonical” characteristic of migraine (20). The recent body of evidence suggests the involvement of physiological (32,33) and psychological (34–36) mechanisms underlying exercise-triggered migraine attacks and exercise avoidance in this population. Because we controlled for the effects of depression and anxiety, our data suggest that a more specific psychological construct could influence LTPA levels in people with migraine, such as intentional avoidance (34), kinesiophobia (36), or cephalalgiophobia (37), which in this context are grounded on the fear of an impending migraine attack triggered by physical exercise.

On the other hand, it could imply that physical inactivity in this population could lead to higher migraine susceptibility and more frequent headache attacks owing to the dysregulation of mechanisms modulating pain processing and perception that are activated by regular PA. In agreement, small clinical studies have suggested significant effects of moderate aerobic exercise training on putative molecules involved in pain modulation and migraine pathophysiology (38–42), with evidence for superior therapeutic effects of vigorous over moderate aerobic exercise in people with migraine (43).

Regarding TTH, our data diverge from other population studies (2,4), including in Brazil (8), which showed no significant associations between LTPA levels and TTH, or rather showed physical inactivity in the leisure time associated with TTH in men (2) but not women. This discrepant data may be due to the more robust control of confounder variables strongly associated with headache disorders in our analysis (e.g. obesity and psychiatric comorbidities). Higher LTPA levels in men than women with TTH and the association of physical inactivity only with women with TTH in our study suggest that sex factors such as hormones and premenstrual syndrome may impact LTPA in these group. Other putative factors reported in smaller observational studies and not assessed in our analysis are the co-existence of TTH with migraine and neck pain. Køll et al. (2017) (28) assessed LTPA levels in patients from a tertiary clinic (n = 148). They found a high prevalence (67%) of patients with co-existing migraine, TTH, and neck pain. The patients with co-existing conditions significantly associated with low PA levels assessed by IPAQ, higher stress levels, and lower psychological wellbeing scores. They also reported migraine as the most burdensome condition that hampered LTPA participation, followed by TTH and neck pain (28). Interestingly, this same research group showed that after an intervention program with aerobic exercise, patients with co-existing migraine, TTH, and neck pain reported higher ability to engage in PA due to reduced burden of TTH and neck pain (44). Physical inactivity may associate with TTH through a combination of higher stress-mediated tenderness of pericranial muscles involving myofascial mechanisms and/or impaired function of descending pain modulation mechanisms at central level (28,44–46).

CPA and primary headache disorders

A novelty in this study, physical inactivity in the CPA domain associated with lower migraine and probable migraine in women, while “somewhat active” associated with increased probable TTH in men. Also, higher PA levels in the CPA domain associated with more frequent headache attacks, and men showed higher PA levels than women with TTH. These findings may indicate a detrimental effect of CPA on major primary headache disorders, possibly due to the influence of other factors not accounted for in our models, as well as owing to sex- and aetiologic-specific characteristics of migraine and TTH. For example, the ELSA-Brasil cohort represents a population living in large metropolitan areas, therefore exposed to the interference of environmental factors such as air pollution. Air pollution exposure has been associated with migraine attack frequency (47) and prevalence (48). Although pollution and odours have been reported as perceived triggers of headache attacks by both migraine and TTH patients (49), migraine patients rather than TTH patients often report osmophobia as a symptom (50). As such, we speculate that in migraine patients this could result in reverse association due to avoidance of potential perceived environmental triggers, whereas the lack of avoidance in men with TTH would allow them to perform higher PA levels but render them more exposed to the trigger effects of air pollution. In agreement, CPA levels also associated with higher odds for hypertension in women in another study with the ELSA-Brasil cohort (51).

Possible negative effects of CPA on headache disorders would cast doubt on the therapeutic effects of recommending CPA as adopted by current WHO PA guidelines embodied in its “every move counts” awareness campaign (52), at least for populations with these specific headache disorders living in large metropolitan areas.

Conversely, the linear trend of higher PA levels in the CPA domain with higher headache attacks could mean that people with increasing headache attack frequency may associate it with a poorer health status and try to comply with a healthier lifestyle by including CPA more often in their daily routine. In fact, the ELSA-Brasil cohort represents a population with more access to programs aiming at health promotion, and more awareness of healthier behaviours, including CPA. These hypotheses need to be tested in further investigations.

Combined PA levels and primary headache disorders

Even combining the LTPA and CPA domains, reduced PA levels were positively associated with definite migraine, probable migraine, TTH, as well as with headache attack frequency. Men showed higher PA levels than women in all headache disorders but migraine. In the analysis of PA levels according to WHO recommendation categories, physical inactivity showed no association with headache disorders. However, not meeting WHO PA recommendations was still associated with higher migraine, probable migraine, and TTH. Similar findings were observed among women, but not in men. Physical inactivity or PA below WHO recommendations was associated with frequent headache attacks.

These findings may reflect the balance between inverse and positive associations depending on the PA domain, and partly explain the lack of association of physical inactivity in the combined PA domain with any headache disorder and headache attack frequency. Nevertheless, these data strengthen the notion that people with major headache disorders such as migraine and TTH, could only comply with PA levels below current WHO PA recommendations on a multidimensional domain. The more consistent associations of lower PA levels (regardless of sex) and unmet levels of recommended PA with migraine are akin to behavioural aspects involving exercise avoidance in this headache entity, which may impact PA behaviour in general (20,26,34,36,37). Accordingly, Bond et al. (2015) (53) assessed PA through accelerometers and reported lower daily PA levels in women with comorbid migraine and obesity (n = 25) compared to women with obesity without migraine (n = 25) during a 7-day period. Nevertheless, TTH may also impact PA levels. Kikuchi et al. (2007) (54) measured daily PA levels by actigraphy and computerized ecological momentary assessment in 34 patients with TTH for 7 days. The authors found that higher headache attack intensity caused less daily PA. However, this latter study did not include a control group to compare daily PA levels with people without headache disorders 54.

The data on combined PA underscores the necessity of assessing PA domains separately in people with headache disorders to establish specific associations and their clinical implications. Unlike many associations between PA and health outcomes for other major chronic non-communicable diseases (i.e. cardiovascular diseases, diabetes, cancer, etc.) (10), PA and headache disorders show no linear dose-response effects, in particular in the LTPA domain. For example, heavy PA at work was associated with higher migraine prevalence in women in the Danish population (5).

Further studies should evaluate the impact of other PA domains (e.g. occupational and housework) and sedentary behaviour (e.g. sitting time), in order to fully understand the associations of specific and multidimensional PA domains with headache disorder subtypes. Moreover, studies should also explore the associations of headache disorders with other PA modalities and intensities, as patients may feel safer and may be more adherent to light/mild PA modalities. Such studies should adopt prospective designs and collect PA data through objective measures (e.g. actigraphy, accelerometers) to avoid recall bias and provide accurate estimates on specific and multidimensional PA domains. These data could help to identify optimal dose-response effects and PA intensity for each headache subtype and contribute to the development of headache-tailored PA guidelines to help reduce the headache burden.

The impact of physical inactivity on major primary headaches is relevant in terms of public health, as their prevalence and disability peak at the population’s most productive age, contributing to lower quality of life and poorer lifestyle behaviours in young adults and the middle aged, with consequential personal and socioeconomic burden (55).

In practical terms, this study provides a ready interpretation with immediate clinical application in terms of PA recommendations. That is, based on our results, recommending a weekly amount of 150 minutes of moderate PA in leisure time for people with migraine and TTH should be encouraged by health care professionals and clinicians managing this clinical population, with caution on recommending PA in commuting time in large metropolitan areas for people with migraine. For women with migraine, accumulating 75 min of vigorous LTPA per week should be rather encouraged. This opens an avenue to implement more feasible PA modalities of shorter duration (e.g. 15 min/day, 5 days/week), such as high intensity interval training (HIIT). Importantly, interventions with PA activity should also address patient psychoeducation approaches to highlight the benefits of complying with PA guidelines and manage negative beliefs about exercise to reduce avoidance (35,36).

Limitations and strengths

Our study carries limitations and strengths worth mentioning. As a cross-sectional data, it is not possible to establish causality between variables. This is not a population-based study, thus the generalizability of our findings is limited. However, the ELSA-Brasil study is a large cohort composed of a population of middle-aged civil servants from large metropolitan areas with LTPA data in accordance with previous findings from the Risk and Protective Factors Surveillance System for Chronic Non-communicable Diseases (VIGITEL) study, a large system of surveillance of risk factors by telephone for variables measured using a similar strategy in both studies (56,57). The data on PA levels, albeit retrieved by validated questionnaire, is not based upon objective measures and therefore is susceptible to recall bias. Although the IPAQ contains five PA domains, in the baseline wave of the ELSA-Brasil study only LTPA and CPA were collected. Sitting time was not collected either. Thus, data on a broader, multidimensional PA dimension is still lacking. Caution while interpreting the data on headache attack frequency is advised and chronic migraine/TTH diagnosis were not possible, as responses were based on categorical options rather than number of days, and specific diagnosis criteria for the chronic form of these headache types were not evaluated. Additionally, although we controlled for the effects of cardiovascular and psychiatric comorbidities in the analyses, we did not include nutritional factors, which were found to be associated with PA behaviour in the ELSA-Brasil study (58). Finally, as the ELSA-Brasil study is focused on cardiovascular diseases and diabetes, only participants 35 to 74 years of age at baseline were included. Therefore, individuals younger than 35 years, in which headache disorders are highly prevalent (6,8), were not considered in this study.

On the other hand, the strengths of this study are as the robust headache diagnosis data retrieved by validated questionnaire adopting IHS criteria, the use of a validated questionnaire to allow comparison with international PA parameters, and the control of confounding variables that could influence the relationship between PA and headaches, such as cardiovascular risk variables, common psychiatric comorbidities and migraine preventive medication.

In summary, physical inactivity is associated with the occurrence of higher major primary headache disorders in the ELSA-Brasil cohort. The distinct associations regarding headache subtype, sex, PA domain and intensity, and headache frequency call for additional studies aiming at identifying specific associations and their implications for headache-tailored physical activity recommendations.