Demographic enquiry is essential to characterize the sample. Data are needed in order to compare those who have been selected with the population of interest from whom they are drawn and of whom they are intended to be representative. While, ideally, these data will reflect all factors that may influence prevalence and/or burden of headache, this objective is necessarily limited by the availability of data characterizing the entire population. National (but perhaps not regional) statistics are commonly available for gender and age distributions. Even when they are not, these are of such prime importance in headache epidemiology that they must be known in the sample. Social situation (especially wealth), habitation (urban or rural) and ethnicity and/or culture may be important influencers of prevalence or burden, and are therefore of some interest.
Diagnosis must follow ICHD criteria  because these are the common language of definition and description of headache disorders . There is no alternative, even though ICHD criteria were not designed for epidemiological enquiry and are not particularly well-suited to it . For epidemiological purposes, these criteria must be built into a structured questionnaire, although this is not how diagnoses are usually made in clinical settings. Open questions are difficult to interpret and categorize, and do not permit algorithmic determination.
Certain criteria distinguishing between migraine and TTH pose particular problems in population surveys . First, empirically it has been found difficult to gather correct responses on headache duration , requiring patients to consider untreated attacks, which they may never have or last had long ago. This results in a high proportion of probable diagnoses because duration criteria appear unfulfilled . Second, there are no easy lay explanations of photo- and phonophobia, which are technical concepts, and even more difficult is to specify what degrees of photo- and phonophobia fulfil migraine criteria in ICHD . False-positive responses, more likely when answers are forced (without the response option of “don’t know”, which is diagnostically unhelpful), push diagnoses towards migraine. HARDSHIP includes a “not sure” option and applies a rule that “not sure” implies absence of the symptom. The reasoning is that presence of a symptom creates a definite awareness of it, and only its absence allows uncertainty. This suggested approach requires further empirical testing.
MOH is diagnosable in cross-sectional studies only as an association of medication overuse with frequent headache (there is no evidence available of causation) . Therefore, all such cases are probable MOH, and it is important to recognize this limitation during analyses and interpretation.
Symptoms of common headache disorders include pain, and, of migraine, nausea, vomiting and photo- and/or phonophobia. Symptom burden is addressed in HARDSHIP by questions 14, 15, 20, 21/23, 24, 29–32, 36 and 37. Pain can be quantified at individual level as a product of intensity, frequency and duration, and at population level as the product of the average among individuals and prevalence. Nausea, photophobia and phonophobia are almost impossible to quantify, but their occurrence can be recorded and frequencies expressed.
Disability attributed to headache is also difficult to quantify completely. Common proxies are lost time and reduced productivity, for which well-validated instruments exist [19, 20]. HARDSHIP (questions 38–44 and 58–62) imports the Headache-Attributed Lost Time (HALT) index . Reliability of recall is an issue here. Burden questions have commonly been limited to a 3-month timeframe [19, 20] as a compromise between the limits of recall and the purpose of enquiry. When the latter is the assessment of an individual patient for therapeutic reasons, the period must be long enough to be representative of that individual. In large-group studies, this is quite unnecessary: different considerations apply, because population- rather than individual-representativeness is sought. Variations of HALT that record over shorter timeframes of one month (HALT-30) and one week (HALT-7) are being tested empirically .
Enquiry into headache yesterday (effectively HALT-1) (HARDSHIP questions 34–45) avoids recall problems almost altogether [18, 22]. It cannot describe the proportion of the population with an active headache disorder, but it yields very accurate information on burden in each individual and, potentially, a rather precise estimate of population burden on a particular day and, therefore, on any day (assuming no major seasonal variation). A large sample is necessary, because 1-day prevalence of episodic headache disorders is obviously much lower than 1-year prevalence. This module probably should not be used except in an unscheduled interview (face-to-face or telephone) ; if it is received by a person with headache on that day, he or she may well postpone answering it until their next headache-free day.
Interictal burden (HARDSHIP questions 64–66) arises because headache attacks are unpleasant, and those who experience them frequently are likely to worry about when the next may occur, and/or attempt to eliminate possible triggers through lifestyle compromise. Interictal burden, which is continuous, is likely to affect subjective wellbeing and may be sufficient to impair quality of life. It is perhaps adequately, if not specifically, captured by measures of subjective wellbeing and quality-of-life measures. HARDSHIP imports, as modules, WHOQoL-8  (questions 90–97) and the four questions on subjective wellbeing taken from the UK-ONS 2012 survey  (98–101).
Cumulative burden (HARDSHIP questions 51–57), accruing over a lifetime, cannot be fully assessed until late in a lifetime. Furthermore, attribution may be uncertain. Nevertheless, a consequence of recurring inability to work may be decreased probability of promotion, and a consequence of lost school-time may be reduced career opportunities. These may be heavy burdens.
An overall summary measure of individual burden is unlikely to be comprehensive, but the concept is attractive for its simplicity . One such measure is willingness-to-pay (WTP) (HARDSHIP questions 67–74). Its reliability as a burden measure remains unclear: its hypothetical nature allows a potential disconnection between what respondents say they will pay and what they actually will pay when confronted by the reality, and of course WTP is constrained by ability to pay. Nevertheless, this form of enquiry has been used to assess sustainability of health-care initiatives in resource-poor countries .
Burden on others, unaffected by headache themselves, is addressed by HARDSHIP questions 75–86. Subjective interpretations are unavoidable. A full account necessitates enquiries among the others, which in practical terms may be possible only among close family members.
Health-care resource consumption (HARDSHIP questions 45–50) is relatively easy to enquire into, but subject to recall bias. It should also be easy to establish who pays for it (the patient, employer, insurer or society via the State). It is less easy to attach accurate costs to individual items of health care, and this may necessitate separate research into health-care costs in the country or region in question .
By far the greater part of the financial cost of headache is the indirect cost of absenteeism and reduced effectiveness at work [16, 26] (HARDSHIP questions 58 and 59). This cost may be borne by individuals, but commonly falls upon employers and/or insurers, and is a cost to national economies (societal economic burden).
Enquiry into comorbidity includes body mass index (HARDSHIP questions 87–89), since obesity may be an important and potentially remediable risk factor for frequent headache . Other (for example, psychiatric) comorbidities can be included by bringing in other instruments (eg, the Hospital Anxiety and Depression Scale , or the Shona Symptom Questionnaire , which may be better suited to some cultures).
The strengths of HARDSHIP are several. First, a very broad base of expert opinion contributed to its evolution. Second, it has undergone testing in many cultures and settings: so far in 19 countries and 18 languages. In some countries within the Eurolight study , rather than being administered by interviewer as envisaged in its development, it was mailed or handed out for self-completion. Third, its successful employment in multiple studies has built a collection of studies conducted with similar methodology in different world regions, facilitating inter-regional comparisons. Likewise, its use in future studies will enable further comparisons. Fourth, its modular design renders it highly amenable to adaptation to suit purpose, resource availability (especially time) and cultural sensitivities.
The one known limitation, discovered empirically, is that the diagnostic question set is relatively insensitive to TTH in all languages and cultures in which it has been tested. The problem is attributable partly to these questions being necessarily tied to ICHD, which makes it difficult to resolve because there is very limited scope for change. More particularly, though, it is due to the nature of TTH itself. Being usually a mild-to-moderate headache, TTH is more likely than migraine to go unreported. Beyond this, because it lacks specific features, it is diagnosed through absence of the features characteristic of migraine, and therefore effectively by default. Reduced sensitivity is inevitable in such circumstances. In this context it is worth noting that, uniquely among the so far reported studies using HARDSHIP, the population survey in China was conducted by physicians. For migraine, the diagnostic question set performed best in China (Table 1) but, for TTH, sensitivity remained low.
In the studies conducted using HARDSHIP, migraine prevalence has been high, although not in China (1-year prevalence 9.3% ), where headache generally appears less prevalent. In Russia the reported 1-year prevalence is 20.8% ; in India (Karnataka State) it is over 25% [unpublished]; in other countries not yet reported, levels of this order have been found. There are two possible explanations of this. First is that some cases of TTH are incorrectly diagnosed as migraine, a possibility suggested by the low sensitivity to TTH just referred to. Second is that these findings reflect the truth, and that migraine is more common than has been thought (global 1-year prevalence about 11% ). We suggest that careful enquiry observes a higher prevalence of migraine by capturing milder cases (ie, by better case ascertainment), and that, although both of these explanations may contribute in part, migraine is indeed more prevalent than past estimates have suggested. Alstadhaug et al.  reported the prevalence of migraine in Norwegian neurologists, among whom diagnoses should be correct and case ascertainment very high. The age-adjusted 1-year prevalence of migraine headache (ie, excluding cases of aura only) was 26.3% (95% CI: 18.5-34.2%), 15.9% in males and 36.7% in females. It is unlikely that Norwegian neurologists are biologically unique.