Published:

Will the real Mr. Average please stand up? (part 2)

By Alec Mian, PhD, CEO Curelator Headache and Anne MacGregor, MD, Specialist in Headache and Women’s Health, Barts Health NHS Trust, London, UK, co-author of the BASH Headache Management Guidelines, now in 3rd edition on January, 2018

Migraine, one of the leading causes of disability worldwide ¹ is a model condition if we want to study variation between individuals and the therapeutic implications of these differences. The hallmark of migraine is episodic, debilitating attacks that are easily diagnosed and monitored. In addition, many people with migraine have several attacks per month, so profiling risk factors both positive and negative (e.g. therapies, protectors, potential triggers etc.) associated with making an individual patient better or worse can be done relatively rapidly.

More importantly, patient susceptibility and response to a wide range of potential factors represent an important spectrum of real-world markers for studying individual variation in genetic, physiological, psychological and ultimately biochemical domains. In addition, many migraine population studies have already been done, generating aggregate data about the average migraine patient, which we can use to benchmark against variation in the individual patient.

A first step toward understanding the level and basis of individual variation in a chronic disease such as migraine versus an “average profile” was recently accomplished in a study done in collaboration with the Department of Neurology, Medical University of Vienna and the Biostatistics Unit, Faculty of Medicine, Universitat Autonoma de Barcelona and a healthcare startup called Curelator Inc.

This study ² , published in Cephalalgia, the journal of the International Headache Society, examined risk profiles of more than 300 individual patients. A key aspect of this research was that it examined a previously analyzed database from a landmark study called the PAMINA study ³ where individuals recorded their daily exposure (or lack thereof) to a list of commonly believed “risk factors” (e.g. commonly called “triggers” but also includes non-causal risk factors such as symptoms that might precede attacks or be part of an attack, e.g. neck pain) associated with migraine: weather, dietary, emotional, physical etc. The original PAMINA study looked at the aggregate population, which yielded the most common trigger associations in that population, namely the “average trigger profile” of the average migraineur. In contrast, the new Cephalalgia study reanalyzed the PAMINA database but did so in each of the individual patients. This individualized approach revealed two unexpected findings.

First, virtually all of the patients in the study where a trigger profile was generated showed unique profiles. How many shared an average profile of four potential triggers the most common being menstruation, neck pain, tiredness and bright lights? Not even one patient.

Second, the data ⁴ revealed that trigger factors in some people were protective factors in others and vice-versa. To be clear: trigger factors are associated with increasing the risk of migraine while protective factors are associated with decreasing risk of migraine. Why is this an alarming result? Because it is one thing to say: there are factors, possibly including therapeutics, that work in some but don’t work in others. It is entirely another thing to say: this works in some but possibly causes harm in others.

The appearance of “protectors” in individual patients is a significant observation. In chronic diseases such as migraine, protectors - factors associated with decreased risk of an attack - have been observed but never measured before. Why not? One explanation is that the aggregation of patient data is subject to a phenomena called, “Simpson’s paradox” ⁵ , which concerns the loss of individual signal after population data aggregation occurs, especially in disease populations with high individual diversity. As an example of Simpson’s paradox, if 10 individuals are each sensitive to 10 different protectors and furthermore those protectors are triggers in other individuals, then the protector signal will likely be lost after data aggregation.

Plot showing mean cycle day of minimum estrogen level in the cycle was day 2 (striped line) but variation in day of minimum estrogen level (range and median day) is shown for each women over three cycles (solid purple line) showing that few women had cycles in which the lowest estrogen day was the same as the calculated mean.

The variation between individual migraine risk factor profiles certainly alerts us that everybody is different on a risk susceptibility level, but will we ever be able to understand the mechanisms driving these differences and relate them to individual therapeutic response?

One of the best studied examples of risk factors for migraine is menstruation and fortunately, data on therapeutic response are also available. As a primary mechanism is believed to be the perimenstrual estrogen ‘withdrawal’ an effective treatment is application of estrogen gel to ‘bridge’ this deficiency. In a study ⁶ of 27 women the authors chose to apply gel perimenstrually for 7 days ending on day 2 of the cycle, which was median day for lowest estrogen level (nadir) recorded in pre-treatment cycles. However, such was the inter- and intra-individual variation among the women that treating until the median day was inadequate - estrogen was consistently lowest at day 2 in only 5 out of 27 women. While estrogen treatment was indeed effective in those 5 women, women with a later estrogen nadir experienced a delayed estrogen withdrawal migraine, that may not have occurred had the timing of treatment been individualized.

Therefore, it seems that an important next step would be to acknowledge the need to understand both the degree of and basis for individual variation in chronic disease. If we proceed in this manner, the journey started last century to discover individual variation may finally acquire a sense of urgency. And if the average approach is limiting, or possibly even causing harm in some individuals – something which might not be detected through aggregate population analysis - then optimizing individual therapeutic pathways and outcomes may be the most effective way forward for patients with chronic disease not adequately addressed by the aggregate approach in modern medicine.

Go back to articles

References:

Vos et al. Years lived with disability (YLDs) for 1160 sequelae of 289 diseases and injuries 1990-2010: a systematic analysis for the Global Burden of Disease Study 2010. Lancet 2012; 380: 2163–96
Peris et al. Towards improved migraine management: Determining potential trigger factors in individual patients. Cephalagia 2016; DOI: 10.1177/0333102416649761
Wöber et al. Prospective analysis of factors related to migraine attacks: the PAMINA study. Cephalalgia, 2007; 304-14.
Donoghue et al. Identification of individual “protective factors” associated with reduced risk of migraine attacks, 2017; In preparation
Simpson EH. The interpretation of interaction in contingency tables. J R Stat Soc Series B Stat Methodol 1951; 13: 238–241
MacGregor. Current Pain and Headache Reports 2008; 12: 468