Complexity of prevalence and incidence measures in headache research

Defining disease status in population-based studies is often challenging, particularly when disease status changes between active and non-active stages. Migraine is a typical example of a fluctuating disease as its frequency and clinical features can vary considerably during one’s lifespan (1). In general, the data show that the vast majority of patients have an onset of their migraine in childhood or early adulthood and only 2% report a first migraine above the age of 50 (1). We are also accustomed to observing data which demonstrate that the ‘prevalence’ of migraine decreases after the age of 40. But what do we mean by prevalence?

Prevalence is defined as the number of cases of a specific condition in a given population over a defined duration of time. Often, however, the duration of time is not well specified in population-based headache studies. If we think of lifetime history of migraine, then the substantial decrease in migraine prevalence could only be explained if patients with migraine would have a dramatically increased risk of death. While there is some evidence that migraine is associated with a modest increased risk of mortality (2, 3), particularly cardiovascular mortality, the effect would by far not explain the dramatic prevalence decrease. Thus, when we refer to a decline of migraine prevalence in midlife, we likely mean the prevalence of active migraine (such as a 1-year prevalence of migraine). However, such definition conflicts with the current International Classification of Headache Disorders, 2nd edition (ICHD-2) (4). A person is classified to have a migraine after at least five attacks with typical migraine characteristics. As there is no distinction between active and non-active phases in the ICHD-2, a person is a prevalent migraine case for the rest of their life after five typical migraine attacks.

What about incidence? Incidence is the new occurrence of a specific disease in a given population of individuals without this disease in a defined follow-up time. Incidence can only be studied among individuals who are ‘at risk’ of getting the disease, which is referred to as the ‘at-risk population’. With regard to migraine, an incident event is a new onset of migraine in a person who never had a migraine before (based on the ICHD-2 criteria, this means no or less than five migraine-specific attacks).

In this issue of Cephalalgia, Khil et al. (5) evaluate the role of headache incidence when different populations at risk are considered. Incidences were assessed in 1122 men and women who were randomly drawn from the general population in Germany and based on the ICHD-2 criteria using standardized headache questions. They defined three populations at risk:

people without any headache;

people without a specific headache (i.e. without migraine or without tension-type headache);

people without migraine and without tension-type headache (but other headaches were allowed).

All prevalent headache definitions are based on information in the year prior to baseline assessment. Then the authors calculated the incidence of any headache, migraine, and tension-type headache in the various populations at risk. Not surprisingly, the populations at risk differed in size, age, and gender. However, the authors also observed substantial differences in the incidence estimates. The incidence of migraine ranged from 0 to 3% and of tension-type headache from 5 to 9%.

The study by Khil et al. (5) is important as it points out that the development of a specific headache disorder may interact with other existing headaches (for example, existing tension-type headache when studying migraine). As information about the various populations at risk are only based on a 1-year interval prior to baseline, ‘incident’ is a mixture of true incident events and recurrent events among those who had a history of headache in the past.

Which population at risk is the right one to use when studying incident headache disorders? It depends on the specific scientific question that should be addressed. If we are interested in how many people in a specific region will develop migraine during the next 5 years, no one with any headache disorder except migraine should be excluded from the population at risk, as this would underestimate the true number. If, on the other hand, we believe that the development of migraine in a person who is completely free of any headache differs from the development of migraine in a person who has another headache disorder, such as tension-type headache, then a distinction between the populations at risk should be considered.

Thinking about an example from the cardiovascular field, we would not exclude people who have a myocardial infarction when we are interested in the number of new stroke cases that will be added to the prevalent cases in a specific region. On the other hand, we might be interested in risk factors for the first occurrence of stroke among those free of any prior vascular disease events, as this group of individuals may have a different risk factor set.

Despite being considered simple measures, estimating prevalence and incidence may be challenging, specifically if diseases fluctuate over time but definitions of the disease do not technically allow such fluctuations. The results from the Khil et al. (5) study show that the incidence of overall and specific headache disorder matter when different populations at risk are considered. Thus, a researcher should be clear about which population at risk best corresponds to the specific scientific question that is being addressed in a study when studying headache incidence.