Samples

Ten surveys were carried out in the Americas (Colombia, Mexico, USA), Europe (Belgium, France, Germany, Italy, The Netherlands, Spain), and Asia (Japan). All surveys were based on multistage, clustered, area probability household samples. Interviews were carried out face to face by trained lay interviewers. Sample sizes ranged from 2372 (Netherlands) to 9282 (USA). Response rates ranged from 45.9% (France) to 87.7% (Colombia), with an average response rate of 70.8%.

Internal subsampling was used to reduce respondent burden and cost by dividing the interview into two parts. Part 1 included the core diagnostic assessment of mental disorders. Part 2 included additional information relevant to a wide range of survey objectives, including assessment of chronic physical conditions and childhood family adversities. All respondents completed part 1. All part 1 respondents who met criteria for any mental disorder and a probability sample of other respondents were administered part 2. Part 2 respondents were weighted by the inverse of their probability of selection for part 2 of the interview to adjust for differential sampling. Analyses in this paper were based on the weighted part 2 subsample (n=18 303). Additional weights were used to adjust for differential probabilities of selection within households, adjust for non-response, and match the samples to population socio-demographic distributions.

Mental disorder status

All surveys used the World Mental Health (WMH) Survey version of the World Health Organization's Composite International Diagnostic Interview (WHO–CIDI) ^{Reference Kessler and Ustun8} (WMH–CIDI; current version, CIDI 3.0), a fully structured diagnostic interview, to assess disorders and treatment. Methodological evidence collected in the WHO–CIDI field trials and later clinical calibration studies showed that all the disorders considered herein were assessed with acceptable reliability and validity in the original CIDI, ^{Reference Wittchen9} and the WMH–CIDI. ^{Reference Haro, Arbabzadeh-Bouchez, Brugha, de Girolamo, Guyer and Jin10} Disorders were assessed using the definitions and criteria of the DSM–IV. ¹¹ Composite International Diagnostic Interview organic exclusion rules were imposed (i.e. the diagnosis was not made if the respondent indicated that the episode of depression or anxiety symptoms was due to physical illness or injury or use of medication, drugs or alcohol).

This paper includes anxiety disorders (generalised anxiety disorder, panic disorder and/or agoraphobia, post-traumatic stress disorder and social phobia) and major depressive episode, which occurred in the previous 12 months and before age 21 years. Depression and anxiety disorders were aggregated into a single category, on the basis of prior findings from the WMH surveys that anxiety disorders and major depression have independent associations of a comparable magnitude with headache and with other chronic physical conditions. ^{Reference Scott, Bruffaerts, Tsang, Ormel, Alonso and Angermeyer7}

Headache

Severe headache was ascertained as part of a chronic physical disorder checklist, of the kind commonly used in national health surveys. ^{Reference Edwards, Winn, Kurlantzick, Sheridan, Berk and Retchin12} Respondents were asked whether they had ever had frequent or severe headache in the previous 12 months. Prior research has demonstrated reasonable correspondence between self-reported chronic conditions (e.g. diabetes, heart disease and headache) and records of general practitioners, ^{Reference Kriegsman, Penninx, Van Eijk, Boeke and Deeg13} but lifetime history of severe headache is subject to recall bias. Although we assessed whether recall bias associated with current depression or anxiety disorder influenced the association of childhood family adversities with adult onset of frequent or severe headaches, the problem of retrospective recall may admittedly remain.

Childhood family adversities

The following kinds of childhood family adversities were included in the present study: physical abuse, sexual abuse, neglect, parental death, parental divorce, other parental loss, parental mental disorder, parental substance misuse, parental criminal behaviour, family violence and family economic adversity. Those respondents who met the criteria specified for a given adversity before the age of 18 years were coded as having experienced childhood family adversity (online Appendix DS1).

Training and field procedures

Experienced CIDI trainers in the USA trained bilingual supervisors in each country. The WHO translation protocol was used to translate instruments and training materials. Some surveys were carried out in bilingual form (Belgium). Other surveys were carried out exclusively in the country's official language. Persons who could not speak these languages were excluded. Quality control protocols, described in more detail elsewhere, ^{Reference Kessler, Bergland, Chiu, Demler, Heeringa and Hiripi14} were standardised across countries to check on interviewer accuracy and to specify data cleaning and coding procedures. The institutional review board of the organisation that coordinated the survey in each country approved and monitored compliance with procedures for obtaining informed consent and protecting participants.

Statistical analysis

The association of childhood family adversities and early-onset mental disorders with adult-onset frequent or severe headache was studied with discrete-time survival analyses, using retrospectively reported age at onset of headache (reported in whole years). The start of the period at risk of adult-onset headache was set at age 21 years. Persons who reported that frequent or severe headache developed before age 21 were excluded.

Cox proportional hazards models were constructed to estimate risk of adult-onset headache as a function of number and type of childhood family adversities and early-onset depression/anxiety disorder status while adjusting for potential confounders (gender, current age and smoking). Country was included in the analyses as a stratifying variable, which allowed each country to have a unique hazard function. This was necessary because of large differences in prevalence rates of childhood family adversities, mental disorders and chronic physical disorders between countries, and to allow for the slight differences in measurement of childhood family adversities among countries (e.g. data from European countries did not include childhood neglect or parental divorce).

In the Cox proportional hazard regression analyses, time to age at onset of headache from age 20 was the dependent variable. Persons who had not developed headache were censored at their current age. The associations are expressed as hazard ratios measuring relative risk. Childhood family adversities were analysed in four categories of number of adversities (none, 1, 2, 3+), with no adversities as the reference group. Childhood family adversities and early-onset mental disorders were included in the models both separately and simultaneously.

To account for the possibility of differential recall of childhood family adversities among those with a current depressive or anxiety disorder, we performed an additional analysis that adjusted for current (12-month) depression or anxiety disorder. We also performed an additional analysis adjusting for educational attainment. Because the inclusion of education in the models made virtually no difference in results, we report results unadjusted for education (detailed information on education was not collected in France, so to have reported all results adjusted for education would also have meant excluding France from analysis). We screened for interaction of childhood family adversities with early-onset depression/anxiety disorder in predicting headache onset, but the interaction was non-significant so only main effects are reported in this paper.

Ninety-five per cent confidence intervals were calculated for all hazard ratios. Statistical significance level was set at 0.05 for a two-sided test. The analyses were performed using the SURVIVAL procedure in SUDAAN (version 9.0.1, SAS–Windows). statistical software to account for the complex sample design. In all multivariate models, a complete-case approach was used, but the WMH survey used multiple imputation methods to impute missing values for the childhood family adversities. The survival analyses employed sample weights to adjust for differences in probability of selection. The unadjusted cumulative percentage with adult onset of frequent or severe headache by age for persons with v. without early-onset depression/anxiety disorder, and for persons with 0, 1 or 2+ childhood family adversities were graphed with product-limit (Kaplan–Meier) life-table estimates developed using SAS PROC LIFETEST (version 9.1.3, Windows).

Types of association

Because of evidence of clustering of childhood adversities in previous studies of the association between such adversities and mental disorders, ^{Reference Kessler, Davis and Kendler15} we assessed a broad range of childhood family adversities in our study. Early-onset mental disorders and childhood family adversities were independently associated with adult-onset frequent or severe headache. This association was strongest for persons with three or more childhood family adversities. This finding is the first of its kind to show in a large cross-national community sample that the association between childhood family adversities and adult onset of headache is independent of mental disorders. Although mental disorder was still independently associated with headache, distinct associations of both types of predictors with headache have not been demonstrated in previous studies. ^{Reference Tietjen, Brandes, Digre, Baggaley, Martin and Recober4,Reference Sachs-Ericsson, Kendall-Tackett and Hernandez5}

In our analyses that decomposed the association of specific childhood family adversities with headache risks, not all the childhood family adversities were found to predict increased risk. Sexual and physical abuse, parental mental and substance disorders and family antisocial behaviours showed the most robust associations with risk of adult-onset headache. Parental death, divorce, and economic adversity were not significantly related. This is different from a previous case–control study, which found significant associations between migraine and parental divorce and chronic financial difficulties in the family after adjustment for education, self-reported general physical health, and depression. ^{Reference Sachs-Ericsson, Kendall-Tackett and Hernandez5}

Sexual abuse and physical abuse have been the most frequently studied childhood family adversities in relation to adult headache disorders. Our study confirmed previous research showing an association between childhood abuse and headache. ^{Reference Golding16,Reference Walling, O'Hara, Reiter, Milburn, Lily and Vincent17} Additionally, the association remained significant when early-onset mental disorders were included in the analysis. This is in keeping with studies showing that headache was associated with childhood family adversities after controlling for depression ^{Reference Scott, Bruffaerts, Tsang, Ormel, Alonso and Angermeyer7} and mental disorders in general. ^{Reference Goodwin, Hoven, Murison and Hotopf18}

Limitations

Our study has several significant limitations relevant to interpretation of its results. Although we used age-at-onset data for childhood family adversities and adult-onset headache to assess temporal ordering, the cross-sectional methodology used imposed a significant shortcoming. Since lifetime history of frequent or severe headache and of childhood family adversities were both ascertained by retrospective self-report, there were multiple opportunities for recall bias to affect both in a similar fashion, thereby introducing spurious correlation. We sought to control this potential bias by carrying out an analysis which controlled for current depression and anxiety disorder. This analysis yielded similar results to analyses which did not control for current mental disorder. None the less, the potential for recall bias to influence the results reported here cannot be excluded, especially for childhood family adversities. Although a comprehensive review indicates that adult retrospective reports of adverse childhood experiences have a worthwhile place in research, substantial false negatives and measurement errors remain possible. ^{Reference Hardt and Rutter19} Prospective research that encompasses corroborative tools of assessment of childhood adversities is needed to address this methodological weakness of our study. We also have not tackled the potentially different meaning and impact of childhood adversity across the participating countries. Parental divorce, for example, may exert dissimilar impact on individuals who grow up in countries with different social and religious attitudes towards broken families. Likewise, the impact of family economic adversity or parental death may vary according to the quality of social service available in a country. These and other cultural differences need to be examined in future country-specific analyses that take social context into account. Regarding the assessment of headache, it was based on a question about history of frequent or severe headache, so that information about the subtypes, chronicity and frequency of attacks of headache was lacking. Consequently, our findings may not be readily compared with research that focused on specific types of headache, especially migraine, which may differ from other kinds of chronic headache in aetiology and the association with depression and childhood family adversities. ^{Reference Breslau, Lipton, Steward, Schultz and Welch1,Reference Peterlin, Ward, Lidicker and Levin20}

A broad developmental perspective

Recognising the limitations of this research, our results suggest that childhood family adversities may predispose individuals to adult-onset headache via mechanisms in addition to Axis I psychopathology. Prior research has found that inadequate coping style and maladaptive personality factors are observed among people who have experienced childhood family adversities. ^{Reference Marlowe21–Reference Romans, Belaise, Martin, Morris and Raffi23} Neuroimaging studies indicate that the brain areas activated during distress caused by social adversities such as exclusion are also activated during physical pain. ^{Reference Eisenberger, Lieberman and Williams24} However, it should be noted that our analyses did not identify interaction or synergistic effects of the number of childhood family adversities with the presence of early-onset depression/anxiety disorder.

In summary, this research found that the association of risk of developing headache was independently associated with both number of childhood family adversities and the presence of early-onset depression/anxiety disorder. Although the aetiology of headache remains poorly understood, these results suggest that a broad developmental perspective concerning risk factors for development of severe or frequent headache may be needed to understand how life stressors influence risks in combination with depression and anxiety disorders.