A few months ago I came across a media article warning me that coffee consumption might lead to migraine headaches. The coffee lover in me needed to find out more, so I went ahead and looked up the original study.
What Were They Studying?
Researchers were interested in comparing incidence of migraines in people on days when they reported high coffee consumption versus low coffee consumption (1).
How Did They Study It?
Researchers used a matched crossover study. In this complex study design individuals are used as their own “controls”. So, instead of finding different people to compare when studying risk factors or outcomes, each individual is compared to themselves, but at a different point in time. This type of study design can work really well. In this study, the occurrence of migraines during days of low caffeine consumption was compared to occurrence during a day with high caffeine consumption.
Aside: Scientists always want to control their experiments as well as possible by creating comparisons where everything but the "exposure" they are interested in is identical. This allows them to rule out the potential for an unobserved or unknown difference between the two comparisons (called a confounding variable in epidemiology) actually leading to the effect of interest.
Think about a situation where researchers note that ice cream consumption leads to increased risk of drowning. The actual variable affecting drowning is increased time spent near water in the summer. Summer is associated with both ice cream consumption and drowning, and if it is not considered or controlled for by the researchers, they will end up misinterpreting their results, and think that ice cream is a risk factor for drowning!
Of course, the kind of control typically used by scientists is very difficult to implement when studying humans because humans are different in so many ways. We cannot create identical humans and then expose them to different risk factors. So, in epidemiology, control subjects are usually people that are selected because they are similar in some important way to the individuals that have the disease of interest. This is the closest that epidemiologists can get to methods typical of experimental studies.
What Did They Find?
Researchers found a significant dose-response relationship between caffeine consumption and risk of migraines. A dose-response relationship means that as exposure increases, so does the risk of disease.
Were There Any Issues With Their Analysis or Interpretation?
The most important point regarding interpretation of the results is that the presented estimates are odds ratios, not relative risks. One of my previous posts discusses the importance of differentiating between the two measures of association when reading health news.
The Healthline article where I found this paper incorrectly discussed the likelihood of migraine based on caffeine consumption, when really, the estimates reported odds of migraine. In this particular context the odds will be higher than the risk.
Also, although p-values have their limitations when it comes to interpretation of results, I think it is important to point out that none of the individual effects of caffeine consumption on odds of migraine were statistically significant. Rather a dose-response trend towards increasing odds of migraine based on increased caffeine consumption was significant. This result is somewhat confusing to interpret, but means that there is some significant increasing trend where increased caffeine leads to increased odds of migraine.
A few additional considerations for study interpretation are the fact that both exposure and outcome data were based on self-report. So, it might be possible that some participants’ responses were affected by their own knowledge regarding their previous caffeine consumption. This could have impacted the final results of the study.
Finally, the cohort used for the study may limit the generalizability of the findings. The study was conducted in medical centres using patients that reported 2-15 days with migraine headaches per month, for at least three months, and the majority of them were women. It is possible that this specific group of individuals have genetic or hormonal differences that would mean the results are not applicable to the general public. Some of these points were captured in the Healthline article, but could have been better emphasized.
Final Thoughts
After going back to the original study article, I thought it was interesting read with some convincing results about the effects of high caffeine consumption on the odds of migraine. But, readers should be sure to interpret the associations reported as odds and not likelihood.
Unfortunately, the bigger issue of caffeine consumption and migraines is a tricky one. My quick review of the literature found that some studies suggest that caffeine can help reduce headache symptoms (2), while others say it can make them worse (3,4) What is clear from the literature is that the amount of caffeine consumption is an important factor (5).
I personally won’t be reducing my two-cups-a-day habit based on this paper!
References
1. Mostofsky, E., Mittleman, M. A., Buettner, C., Li, W., & Bertisch, S. M. (2019). Prospective Cohort Study of Caffeinated Beverage Intake as a Potential Trigger of Headaches among Migraineurs. The American journal of medicine, 132(8), 984-991.
2. Lipton, R. B., Diener, H. C., Robbins, M. S., Garas, S. Y., & Patel, K. (2017). Caffeine in the management of patients with headache. The journal of headache and pain, 18(1), 107.
3. Scher, A. I., Stewart, W. F., & Lipton, R. B. (2004). Caffeine as a risk factor for chronic daily headache: a population-based study. Neurology, 63(11), 2022-2027.
4. Lee, M. J., Choi, H. A., Choi, H., & Chung, C. S. (2016). Caffeine discontinuation improves acute migraine treatment: a prospective clinic-based study. The journal of headache and pain, 17(1), 71.
5. Nehlig, A. (2016). Effects of coffee/caffeine on brain health and disease: What should I tell my patients?. Practical neurology, 16(2), 89-95.