Abstract
The number of epidemiological headache studies is rapidly increasing, and has accentuated the need to improve the quality of the way studies are performed and how results are interpreted. The aims of this review were to summarize the main findings from the Head-HUNT study, to describe strengths and limitations of the Head-HUNT approach and to discuss the significance of some of our findings. Head-HUNT included a large sample size of 51 383 participants that gave the opportunity to make analyses also of relatively rare conditions. The wide range of health-related information made it possible to adjust for many potential confounding variables. Blood samples for future genetic headache studies are available for a non-selected large group of individuals. The data show that among several factors that have been evaluated, age and gender were the two most clearly related to migraine, whereas analgesic overuse and the presence of some comorbid conditions were most strongly associated with the prevalence of chronic headache. Interesting relationships to blood pressure and thyroidstimulating hormone levels were also found. Methodological problems in headache epidemiological studies are discussed and, in particular, problems with causal inferences. Despite the limitations of the head-HUNT study, we believe that the results so far have provided clues to causes and preventive factors of headache that should be explored in other populations and in prospective studies.
Introduction
Epidemiological studies have been important in the detection of causes and risk factors for various disorders, e.g. cholesterol for cardiovascular disease and tobacco use for lung cancer. In the same way, one may hope that epidemiological studies may increase our knowledge about risk factors for headache so that the occurrence of headache can to some degree be prevented (1). Headache epidemiology is still a young discipline and less than 40 years have passed since Waters performed the first population-based headache study (2). The majority of studies so far have been descriptive, i.e. giving the distribution of migraine and other headaches in a population along with data on attack features and the consequences of headache. However, there is a rapid increase in the number of analytical epidemiological studies trying to elucidate potential causes by evaluating associations between exposure to potential risk factors and headache. This high activity in headache epidemiology accentuates the need to improve the way we perform such studies and interpret the results.
One of the largest epidemiological studies to date is the Nord-Tr⊘ndelag Health Survey (‘Helseunders⊘kelsen i Nord-Tr⊘ndelag’ = HUNT) (3). Among a wide range of topics, the study also included a headache questionnaire which formed the basis for the Head-HUNT study. In addition to describing the prevalence and consequences of headache, an important aim was to provide clues to causes and mechanisms for the most common headache types. So far, the relationship between headache and demographic factors [e.g. socio-economic status (SES)], analgesic overuse, other self-reported complaints [e.g. chronic musculoskeletal complaints (MSCs), psychiatric symptoms and gastrointestinal (GI) complaints] and various measurements [e.g. blood pressure (BP) and thyroid-stimulating hormone (TSH)] have been investigated (4–21).
The aim of this review was to summarize the main findings of the Head-HUNT study, to describe the strengths and limitations of the Head-HUNT approach and to discuss the significance of some of our findings.
Materials and methods
Study population
The two HUNT studies performed so far both included all the inhabitants in the Nord-Tr⊘ndelag County aged ≥20 years. In the 1984–86 HUNT (HUNT 1) a total of 74 599 persons participated (88%) and this study mainly focused on cardiovascular risk factors and diabetes mellitus. The questionnaire did not include headache items, but 59 471 persons responded to a question on use of analgesics, and BP was measured in all.
The 1995–97 HUNT (HUNT 2) study was much more comprehensive, collecting more information from each participant and covering a wide range of topics in two questionnaires. The first questionnaire (Q1) was enclosed with the invitation letter, completed at home and delivered during attendance at the health examination which included BP measurements and blood sampling. The second questionnaire (Q2), which also included headache questions, was filled in after the examination and returned by mail. A total of 51 383 subjects (56%) out of all the 92 566 invited inhabitants completed the headache questionnaire (3). Details of the non-participants are described elsewhere (3, 22). In HUNT 2 a number of blood analyses were performed, e.g. including a haemochromatosis screening of all participants and TSH measurements of 28 058 subjects. HUNT 2 also included a separate study including headache questions and interviews among adolescents (the Head-HUNT-Youth Study). Results from this part of the study are given elsewhere (23–25).
Headache diagnosis
The headache questions in Q2 and the prevalence of headache are published elsewhere (3, 26). The headache questions were designed mainly to determine whether or not the person had headache, determine frequency of headache and diagnose migraine according to a modified version of the migraine criteria of the Headache Classification Committee of the International Headache Society (27). Subjects who answered ‘yes’ to the question ‘Have you suffered from headache during the last 12 months?’ were classified as headache sufferers. Based on data from the subsequent 12 headache questions, they were classified into two groups of either migraine or non-migrainous headache. The diagnoses were mutually exclusive. A headache which did not fulfil the criteria for migraine was classified as a non-migrainous headache. Based on a question about headache frequency during the last year, headache frequency was divided in three categories; <7 days/month, 7–14 days/month and >14 days/month.
The classification of the subjects has been described in detail previously and has been validated by interview diagnoses (26). In short, for migraine, the positive predictive value (PPV) was 84% and the negative predictive value (NPV) was 78%; for non-migrainous headache, the PPV was 68% and the NPV was 76% (26). For the questionnaire-based diagnosis of migraine with aura (MA) the PPV and specificity were 100%, whereas the NPV was 91% and sensitivity 42% (26).
Statistics
There are no established rules or guidelines on how to summarize multiple results from epidemiological headache studies. In our previous papers, each variable or sets of variables were analysed with multiple logistic regression and the results were given with prevalence odds ratios (ORs) and with 95% confidence interval (CI) as measures of the association.
In general, the value of the estimated association may indicate the relative influence of each factor on the headache prevalence. By using logistic regression, the relative contribution of each headache-associated factor can to some degree be evaluated by looking at the estimated proportion of variance (R 2).
Statistical analyses were performed using the Statistical Package for the Social Sciences (SPSS), version 13.0 (SPSS Inc., Chicago, IL, USA).
Resumé of the main findings
1-year prevalence
The overall age-adjusted 1-year prevalence of headache in Nord-Tr⊘ndelag County in HUNT 2 was 38% (46% in women and 30% in men). The prevalence of migraine was 12% (16% in women and 8% in men) and for non-migrainous headache 26% (30% in women and 22% in men). For chronic headache (>14 days per month) the prevalence was 2.4% (2.8% in women and 1.7% in men) (3), whereas the prevalence of possible medication overuse headache (MOH) was 1% (1.3% in women and 0.7% in men) (8).
Relation to demographic factors (gender, age and SES)
The women : men ratio was higher for migraine than for non-migrainous headache (2.1 vs. 1.4) (3), whereas the sex ratios for chronic headache and possible MOH were 1.6 and 1.9, respectively (3, 9). With respect to distribution among different age groups, the prevalence varied more for migraine than for non-migrainous headache (3). For both headache types, the prevalence peaked in the fourth decade of life (3). Using the prevalence among individuals ≥80 years old as a reference, the prevalence in the age group 30–39 years was 4.8 times higher for migraine, but only 2.3 times higher for non-migrainous headache. The prevalences of chronic headache and possible MOH were nearly constant at 2.4% and 1%, respectively, across the age groups (3, 9).
In the HUNT 1 population 22 685 adults did not use analgesics and these were considered less likely to suffer from headache and thus to be ‘at risk’ of developing headache in a prospective analysis. Among these, those with low SES defined by educational level or type of occupation had a doubled risk of chronic headache in HUNT 2 compared with those with high SES (4). The risk of chronic headache decreased with increasing income, but only among men (4). Also in the cross-sectional analyses (n = 49 948), low SES (defined by education level, occupation type or income) was associated with nearly twice as high prevalence of chronic headache compared with those with high SES (4).
Relation to self-reported complaints
Chronic MSCs (defined as MSC for at least 3 months during the past year), GI complaints reported in Q1 in HUNT 2 and anxiety and depression measured by Hospital Anxiety and Depression Scale (HADS) in Q1 all had an almost similar relationship with headache and, for all these complaints, migraine and non-migrainous headache was 2–3 times more likely among individuals with such complaints than among those without (7, 13, 14). Frequency of headache had a higher impact than headache type on the association and the strongest relationship was found for chronic headache. The prevalence of chronic headache was 5–7 times higher among individuals with at least one of these complaints than among those without (7, 13, 14).
Both migraine and non-migrainous headache were approximately 1.5 times more likely among individuals who reported asthma and/or chronic bronchitis than among those without such symptoms (16). The strongest association was found for individuals with chronic headache (16).
Relation to other self-reported health information
Use of tobacco was also associated with higher headache prevalence. Prevalence of current smoking increased with increasing headache frequency, most evident among those with age <40 years, with an OR of 1.7 (95% CI 1.3, 2.2) among those with chronic headache compared with headache-free individuals (15). Use of alcohol was related to lower ORs for headache. However, alcohol overuse measured by CAGE (cut down, annoyed by criticism, guilty about drinking, eye-opener drinks) was 1.5 times more likely among individuals with chronic headache than among those without headache (15).
Among users of female hormones and analgesics there were also higher headache prevalence rates. Among postmenopausal women using hormone replacement therapy (HRT) and among premenopausal women using oestrogen-containing oral contraceptives (OCs), the prevalence of migraine and non-migrainous headache was significantly higher (P < 0.001) than among the women who never had used HRT or OCs (10, 11). Individuals who reported use of analgesics daily or weekly in the 1984–86 HUNT study had markedly increased risk of chronic headache, most evident for migraine >14 days per month (OR = 13.3, 95% CI 9.3, 19.1) (8). In the cross-sectional analysis, chronic headache was more than seven times more likely among those with analgesic overuse than among those without (9).
The prevalence of sick leave increased with increasing frequency of migraine and non-migrainous headache and the prevalence of sick leave of >8 weeks during the past year was more than three times higher among individuals with headache >14 days/month than among those without headache (19).
Relation to measured variables
High BP and high TSH values were both associated with low headache prevalence. Individuals with a systolic BP of ≥150 mmHg in 1984–86 HUNT had a 30% lower risk (OR = 0.7, 95% CI 0.6, 0.8) of having non-migrainous headache at follow-up compared with those with systolic BP <140 mmHg (5). For migraine, an inverse relationship with systolic BP was found only among women. For diastolic BP, the risk of non-migrainous headache decreased with increasing values. Although the picture was less clear in the cross-sectional analyses, high systolic BP was associated with low headache prevalence (5).
Similarly, high TSH values were associated with low headache prevalence, most evidently among women with no history of thyroid dysfunction (6). Among these individuals, headache was less probable (OR = 0.5, 95% CI 0.3, 0.7) if TSH was ≥10 mU/l than in women with normal TSH (0.2–4 mU/l) (6).
High body mass index (BMI) defined as ≥35 kg/m2 was associated with 1.7 (95% CI 1.3, 2.3) times higher prevalence of chronic headache compared with normal BMI (<25 kg/m2) and there was a linear relationship between headache prevalence and an increase in body weight from 1984–86 HUNT to 1995–97 HUNT (20).
Moderately low haemoglobin values were associated with low headache prevalence. Among 2385 women aged 20–55 years there was a linear trend of decreasing prevalence of headache and migraine with decreasing haemoglobin (18). Migraine was less likely among women with haemoglobin values <11.5 g/dl (OR = 0.4, 95% CI 0.2, 0.9) compared with those with normal values (11.5–15.5 g/dl) (18).
Phenotypic haemochromatosis and the C282/C282Y genotype were both associated with an 80% increase (OR = 1.8, 95% CI 1.1, 2.9) in headache prevalence, but only among women (12).
Val158Met polymorphism at the COMT gene was not associated with migraine, whereas non-migrainous headache tended to be less likely among women with Val/Val genotype than among women with the other genotypes (17).
Comparison of different variables
The cross-sectional analyses of the 51 383 participants in 1995–97 HUNT are summarized in Table 1, presenting OR and R 2 of each factor. The total OR and R 2 when two or more factors are included simultaneously are also presented.
Estimated proportion of variance (R 2) and odds ratio (OR) of headache (all types), non-migrainous headache, migraine and chronic headache (>14 headache days per month) among 51 383 participants in 1995–97 HUNT
Gender; Men = reference, SES/education: High = ≥13 years = reference, low = ≤9 years.
Chronic MSCs = no musculoskeletal complaints for at least 3 months during the past year = reference. Anxiety and/or depression: Hospital Anxiety and Depression Scale (HADS) <11 = reference. GI complaints: No reflex symptoms, diarrhoea, constipation or nausea = reference.
TSH 0.2–4 mU/l = reference, systolic BP <140 mmHg = reference; use of female hormones: no use of oral contraceptives or hormone replacement therapy = reference; Smoking: never smoked = reference. Alcohol overuse: Yes = at least one positive answer in CAGE (cut down, annoyed by criticism, guilty about drinking, eye-opener drinks).
Use of analgesics: no use of analgesics daily or almost daily for ≥1 month during the past year = reference. Sick leave = no sick leave past year = reference vs. >8 weeks' sick leave.
Among the demographic factors, age has a somewhat higher influence on migraine prevalence than gender (R 2 = 0.08 and 0.06, respectively). Together, gender and age give a rise of R 2 to 0.14. In contrast, age and gender do not stand out as important factors for explaining the prevalence of chronic headache. For chronic headache, self-reported complaints (chronic MSCs, anxiety and/or depression and GI complaints) have a much higher impact and including all these complaints in the analyses resulted in a R 2 of 0.14. Among the 1014 individuals who reported both chronic MSCs, GI complaints as well as anxiety and/or depression, chronic headache was 38 times more likely (OR = 38.2, 95% CI 27.3, 53.5) than among the 50 369 without all these three complaints. Among other self-reported health-related information, regular use of analgesics during the past year had a strong impact on the prevalence of chronic headache (R 2 = 0.14).
In general, the measured variables did not have a very strong influence on the headache prevalence. However, among the small subgroup of 101 individuals with both TSH >8 mU/l and a systolic BP ≥160 mmHg the prevalence of headache was 50% lower (OR = 0.5, 95% CI 0.3, 0.8) compared with those with TSH between 0.2 and 4.0 mU/l and systolic BP <140 mmHg. This difference in headache prevalence did not influence R 2, probably because these values occurred only in very small subgroups, e.g. TSH >10 mU/l was found in only 243 out of 28 058 individuals investigated.
Discussion
The overall 1-year prevalence of headache was 38% (migraine 12% and non-migrainous headache 26%) and of chronic headache 2.4%. The migraine prevalence of 12% is consistent with data from other population-based studies in western countries (28). Because headache sufferers were classified into two groups of either migraine or non-migrainous headache, and the diagnoses were mutually exclusive, potential problems with so-called ‘double-counting’ were avoided. However, according to the two Danish population surveys >90% of migraineurs also have tension-type headache (29, 30). Thus, the coexistence of multiple headaches in the same individuals will have a great influence on the prevalence of non-migrainous headache, especially tension-type headache.
The data from Head-HUNT showed that among several factors that have been evaluated, age and gender were the two factors most strongly associated with migraine, whereas analgesic overuse and the presence of comorbid conditions were the most strongly associated with chronic headache. A thorough discussion of each result is given in the different papers (3–19), but in the following we will highlight some results in order to illustrate general principles.
Methodological considerations
Bias
The head-HUNT study was incorporated into a large health survey and was questionnaire based. Among several potential pitfalls using such methodological approach, misclassification bias is probably one of the most important.
Comparing the questionnaire-based headache diagnosis with the gold standard diagnosis, i.e. neurologist interviews, showed that our questionnaire-based headache diagnoses were not optimal (26). Some degree of misclassification could probably be explained by change in the headache condition itself, because the self-administrated headache questionnaire was filled in 5–9 months prior to the diagnostic interview (26). On the other hand, since we enriched our validation group with individuals with moderate to severe headache and included only a few participants with no or mild headache, the agreement between the questionnaire-based headache diagnosis and diagnosis by neurologist may have been overestimated (26).
The low sensitivity (42%) of the questionnaire-based diagnosis of MA makes the instrument unsuitable for determining the prevalence of MA in the population. The high specificity (100%), however, indicates that those who fulfil the criteria of MA most probably have MA, making this group suitable for association studies. Such a strategy was used evaluating the association between depression/anxiety disorders and migraine with and without aura (21). A similar high sensitivity and a much higher specificity were obtained in another large-scale headache epidemiological study (American Migraine Study II), which also used only a few questions for the diagnosis (31). This indicates that there is room for considerably improvement to our headache questions in this respect.
In order to minimize misclassification between the main headache types, we used for most studies a simplified headache classification with only two subgroups: migraineurs and non-migrainous headache sufferers. The bias caused by misclassification can either exaggerate or underestimate the true difference between headache groups. Most likely, differences between the migraineurs and the non-migrainous headache sufferers were underestimated due to the presence of migraine patients in the group of non-migraineurs and vice versa, making the two groups more similar than they really were. A similar problem is encountered in the differentiation between headache sufferers and the reference population. The latter included those who did not consider themselves as ‘headache sufferers’. If the discrimination between diseased and non-diseased had been based on the more neutral question ‘do you have headache’ the reference group would have been much smaller, because headache is experienced at least occasionally by the great majority of the population (29, 30). The differentiation between those with and without headache may be difficult in epidemiological studies, as illustrated by one previous study, which showed that 34% (13 out of 38) of the individuals who answered negatively to the direct question whether they had had headache or not, reported relatively frequent headache when asked to keep a headache diary (32). Population-based data from Denmark based on face-to-face interview have shown that <5% of the population never had headache and that 44% of the population had tension-type headache <15 days per year (29, 30). There is no consensus on the ‘ideal’ headache-free or ‘headache-poor’ reference group in epidemiological studies, but it seems reasonable to include those with no headache or infrequent episodic tension-type headache (<12 headache days per year). This will result in a reference group of approximately 50% of the general population, according to some studies in western countries (28–30). The reference group in the Head-HUNT study (62% of the population) included those who did not consider themselves to ‘suffer’ from headache. Although we have no confirming data, we may assume that this group mostly consists of those with infrequent episodic tension-type headache.
It is a potential problem of the Head-HUNT study and of many other large-scale epidemiological studies that multiple headache types often coexist in the same individual (29, 30, 33, 34). Personal interview and examination by a neurologist (the ‘gold standard method’) is required to make reliable diagnoses (34). This expensive and time-consuming approach has been used in only a few population studies (29, 30, 35) and such studies have shown that >90% of migraineurs also have tension-type headache (29, 30). In studies in which only one diagnosis is made, the aim can only be to differentiate between those who suffer from migraine and those who have only other headaches (non-migrainous headache) (34). However, because headache patients in a questionnaire may choose to report their most frequent headache, the potential for misclassification is probably highest for individuals with only rare migraine attacks during the past year or with migraine attacks that occurred more than 1 year ago (33). Such misclassification could influence the results in, for example, genetic studies on migraine. For such studies, reliable information on lifetime prevalence of migraine is the most relevant.
In addition to misclassification bias, selection bias may influence the results. In HUNT 1 and HUNT 2, all adults in the county of Nord-Tr⊘ndelag were invited and problems related to sampling methods could therefore be avoided. In HUNT 2, 70% participated, of whom 80% responded to the headache questions, giving an overall participation rate of 56%. The influence of the non-participants has been discussed in more detail elsewhere (3). The fact that Head-HUNT was not the primary objective of HUNT makes selective participation due to headache unlikely. An indication of this is the fact that a similar disorder, MSC, did not differ in prevalence between the 51 383 participants in Head-HUNT and the 13 177 individuals who answered Q1, but not the headache questions in Q2. Previously, a high correlation between headache and neck or shoulder pain has been found (7). Hence, we may assume that the headache results are fairly representative of all the 64 560 who participated in HUNT 2. However, individuals who responded to the headache questions tended to be younger, were more likely to be women and had higher socio-economic status than the non-responders. Thus, generalization of our results to those who did not participate must be done with some caution.
In studies dealing with subjective complaints such as headache, musculoskeletal pain and psychiatric symptoms the results may be influenced by a tendency to answer in a similar way all questions regarding complaints (‘reporting bias’) (36). This may create strong associations that may reflect personality traits rather than biological mechanisms (37). Reporting bias has been invoked to explain, for example, why Scottish men who viewed their lives as more stressful were also more likely to report chest symptoms even though they had no increase in objectively registered cardiac ischaemia (38). We suspect that some of the marked associations between headache and self-reported complaints may at least partly be due to this phenomenon. To verify that such associations are due to biological mechanisms, studies would be required where these disorders and complaints are registered in a more objective way than by self-report in a questionnaire. Possibly, this is less of a problem when using a validated diagnostic instrument such as HADS to diagnose anxiety and depression. Therefore, although the association between headache and some of these self-reported complaints (e.g. chronic MSCs) are impressive, the weaker associations between headache and objectively measured variables such as TSH, BP and genetic polymorphisms may be even more interesting because the potential problem of reporting bias does not arise in these analyses. Conceivably, reporting bias is also less of a problem with self-reported variables that are not complaints, e.g. tobacco or medicine consumption.
Bias due to confounding variables may influence the measured association when such variables are associated with both headache and a suspected causal factor. For example, headache prevalence is strongly associated with age and, similarly, thyroid disease and blood pressure, so all analyses had to be adjusted for age. One of the major advantages with large-scale health studies such as the HUNT study is the wide range of health-related information which makes it possible to adjust for many potential confounding variables. However, one can not rule out the possibility that there may be other unmeasured factors or factors which are incompletely registered that could influence our findings (residual confounding).
Role of chance
Bias and confounding factors may give systematic errors. Sources of error in estimation may also be random, which includes the role of chance. In the head-HUNT study the large study population reduced the role of chance and, as a consequence, increased the precision of measurement and estimation. It also facilitated extensive subgroup analysis which was necessary to obtain sufficiently large groups of individuals with relatively rare findings, e.g. TSH >10. A large sample size is also necessary in polymorphism–disease association studies because of a relatively high risk that positive findings may be due to coincidence (38).
When many hypotheses are tested, some statistically significant findings may occur by coincidence. In our studies, there were several factors that were not associated with headache prevalence (e.g. serum ferritin (12) and Val158Met polymorphism at the COMT gene (17)). It has also been convincingly shown that when many hypotheses are tested of which a large majority are false, much more than 5% of the analyses will have a P < 0.05 (38). By increasing the power and by setting the level of significance lower than the usual 0.05 (e.g. 0.01 or 0.001), these sources of error can be greatly reduced. However, the problem can not be fully overcome in a single study and only by performing studies in other populations by other groups will it be possible to sift the evidence to identify the true associations. On principle grounds one may also argue that hypothesis testing is relatively irrelevant in the analysis of epidemiological data, and we emphasize that our results are mainly hypothesis generating (39).
Causality
A main motivation for performing analytic epidemiological studies is to discover new mechanisms and causes that may be amenable to treatment or preventive measures.
Epidemiological studies can indicate associations between causes and effects but cannot provide direct proof (40). In 1965, Hill provided a list of criteria that characterize causal relationships (41). One of his criteria for causality was valid ‘relation in time’ (i.e. that the exposure must precede the development of disease). In cross-sectional studies potential cause and disease is registered at the same point in time. Therefore, only statistical associations can be studied, without a distinction between cause and effect. This is a problem with regard to factors that may be changed as a result of headache. This may be relevant not only for use of analgesics (9) but also for OCs or HRT, which to some degree are used to reduce various complaints (10, 11). Because our genes do not change as a result of headache, this potential problem is avoided in polymorphism–headache association studies.
Prospective cohort studies are necessary to establish that a valid temporal relationship is present. Our findings are mainly cross-sectional because we did not have data in HUNT 1 on a headache-free-population ‘at risk’. However, using ‘use of analgesics’ to define a relatively headache-free population we demonstrated that low BP and low SES were prospectively associated with development of headache. Admittedly, use of analgesics is an indirect and imprecise way to identify a headache-free population at baseline. True prospective data will become available after the next HUNT study (HUNT 3, 2006–2008).
High BP was associated with low headache prevalence (5). Another criterion for causality is consistency among different studies (41). With respect to consistency, it should be mentioned that our group also demonstrated a similar inverse relationship between BP values and chronic MSCs in all parts of the body (42). Furthermore, a decreasing frequency of migraine with increasing BP has been found by other groups (43–45). In one of these studies, the relationship was found only for systolic BP, whereas an opposite relationship was found for diastolic BP (45). Biological plausibility, another Hill criterion, is also present, since data from humans and rats suggest an interaction between the cardiovascular and pain regulatory systems, a phenomenon termed ‘hypertension-associated hypalgesia’ (46). In accordance with this, low pain sensitivity has been reported in hypertensive animals and humans as well as in groups deemed to be at an increased risk for development of hypertension (46).
High TSH values may be another factor that may be protective against headache (6). Regarding consistency, we found a similar inverse relationship between TSH values and chronic MSCs in all parts of the body (47) and, regarding biological plausibility, there are some clinical observations noting that patients with hypothyroidism seem to be relatively insensitive to nociceptive stimuli and have higher pain thresholds than controls (48, 49).
Strength of the association is also a criterion of causality, since weak associations are more likely to result from undetected bias. The value of the estimated association was indicated by the size of OR. As demonstrated in Table 1, the strongest association was found between chronic headache and analgesic overuse and comorbid conditions such as chronic MSCs, GI complaints and psychiatric symptoms. Further prospective studies are needed to clarify whether analgesic overuse and comorbid conditions are causes and not only consequences of chronic headache, or whether they share a common underlying biological explanation (50).
The relative contributions of the different associated factors were also evaluated by the estimated proportion of variance (R 2). Previously, R 2 has been used, for example, to indicate factors which were most important to explain the variation in prevalence estimates of headache and migraine inmeta-analyses using linear regression (51). In the analyses of headache studies using logistic regression, use of R 2 has, to our knowledge, never been reported.
We should point out that if a preventable or treatable cause of headache can be detected, it may be of great importance even if it only explains a small proportion (e.g. 1%) of all headaches, due to the high headache prevalence in the general population. The fact that only a small proportion of variation of headache is ‘explained’ by the factors listed in Table 1 (according to R 2) may indicate that we have not had access to the most relevant causal factors, e.g. hereditary factors.
Concluding remarks
Head-HUNT included a large population, which allowed us to investigate possible associations to relatively rare conditions such as haemochromatosis and to measured variables such as TSH and BP. The wide range of health-related information made it possible to adjust for many potential confounding variables. Blood samples for future genetic headache studies are available for a large non-selected group of individuals.
Even though many factors associated with primary headaches are known, there is still little robust knowledge about their causes. Analytical epidemiological studies may prove to be important instruments to demonstrate possible causes. We therefore recommend headache researchers to try to include headache questions when large population-based health studies are planned. The headache questions, although used before in other studies, must be validated in a completely random sample of the participants with a face-to-face-interview by a neurologist. The validation interview should be done in close temporal proximity to the main study so that one can be certain that differences in the diagnosis are caused by the difference in method and not by a change in the headache condition itself (1). Cross-sectional studies may be an efficient first step to identify factors that cause or prevent headache (52). Since chance findings are a problem with the associations demonstrated in epidemiological surveys, additional studies to ‘sift the evidence’ generated, e.g. in the head-HUNT study, are needed. As a next step, associations found in cross-sectional surveys should be evaluated in prospective studies of a headache-free cohort at baseline. Prospective studies are time-consuming and, so far, cross-sectional headache studies dominate. It is an important challenge in the future to perform longitudinal epidemiological headache studies. To improve the quality and comparability of headache epidemiological studies, international guidelines on how to perform such studies are urgently needed.
Acknowledgements
The Nord-Tr⊘ndelag Health Study (The HUNT study) is a collaboration between The HUNT Research Centre, Faculty of Medicine, The Norwegian University of Science and Technology (NTNU); Norwegian Institute of Public Health; and the Nord-Tr⊘ndelag County Council.