Modern Technology Applied to a Traditional Approach: A New Way to Assess Atrial Fibrillation Risk in Migraine Patients
Migraine Science Collaborative
Make Connections. Stimulate Discovery. Help us coordinate scientific research, funding, and treatment for migraine.
A large study of 40,000 people with migraine shows that an artificial intelligence-enabled electrocardiogram predicts an increased risk of atrial fibrillation in patients with migraine with aura.
By Neil Andrews
People who have migraine with aura (MwA) are at increased risk of ischemic stroke, with almost two times the risk of those without migraine. Considering that atrial fibrillation (AF) is a risk factor for stroke, assessing the risk of AF could be of immense benefit to MwA patients, hopefully preventing them from a devastating complication of their headache disorder. A new study now says that marrying new technology to an old-school approach may make accurate prediction possible.
The research, led by Chia-Chun Chiang, MD, at Mayo Clinic, Rochester, Minnesota, US, reports that an artificial intelligence algorithm applied to simple and inexpensive electrocardiograms (AI-ECG) predicted an increased risk of AF in patients with MwA, compared to those with migraine without aura (MwoA). Of note, the difference in risk between the two migraine subgroups often remained significant when parsing the results by sex and age, most notably in younger patients under 55, a population for whom it is difficult to accurately assess AF risk.
“This is a very well-done, interesting and innovative study that raises a lot of questions that I think are answerable by continuing on with the same methodology and maybe adding to it,” said Gretchen Tietjen, MD, a headache medicine and vascular neurology expert at the University of Toledo, US.
“This really opens up the door to so much more research, which these researchers and others can do in this population, that could benefit that subset of migraine patients at increased risk of developing stroke,” according to Tietjen, who was not involved with the current investigation.
The paper, for which Chiang received the 2022 Early Career Award from the American Headache Society, appeared June 8, 2022 in Headache.
A new tool discovered in an electronic medical record system
Chiang, a headache and stroke specialist, told Migraine Science Collaborative that the origin of the new study traces back to when she was taking care of stroke patients in the hospital.
At that time, she became aware that the Mayo Clinic's electronic medical record system had an AI-ECG “dashboard” showing the probability that a patient who had received an ECG at Mayo Clinic had paroxysmal AF [an irregular heart rhythm that occurs occasionally and usually lasts only for a few minutes or hours].
“I thought that it was a very cool tool, and when I went back to the headache clinic, I kept thinking about how we could utilize it to study patients with migraine,” said Chiang.
It seemed a particularly promising approach considering the obstacles to detecting AF.
“Paroxysmal atrial fibrillation is very difficult to detect because a patient might be in atrial fibrillation at one time, but they might go back to sinus [normal] rhythm after just a couple of minutes,” said Chiang.
“Typically, the gold standard to diagnose paroxysmal atrial fibrillation is prolonged cardiac monitoring. I thought that this AI-ECG would be a very useful tool to study patients with migraine because very few patients with migraine undergo such monitoring – usually we don’t just give patients a cardiac monitor to wear for 30 or 60 days unless there is a high suspicion of abnormal heart rhythms, because that can be quite expensive,” Chiang continued.
So, the researchers designed a study to test the usefulness of AI-ECG for assessing the risk of AF in people with migraine, specifically the risk of concurrent paroxysmal or impending AF within 30 days of the original ECG.
The AI algorithm was trained based on more than 600,000 ECGs (interpreted as having normal sinus rhythm) that had previously been administered over the past 20 years by cardiologists at Mayo Clinic. After this training, the researchers could use the algorithm to assess the risk of AF in MwA versus MwoA patients, based solely on feeding additional ECGs into the algorithm.
Patients included in the study needed to have a MwA or MwoA diagnosis and at least one ECG. Patients who had a confirmed diagnosis of AF were excluded from the research. Ultimately, this retrospective cross-sectional study included 40,002 people with migraine, (17,840 with MwA and 22,162 with MwoA).
Turning in a solid performance
The team first observed that more MwA patients had a confirmed ECG diagnosis of AF, compared to MwoA. Demographically, the majority of study participants were White and non-Hispanic, and in their mid to upper forties when they had originally received their ECG. More MwA than MwoA study participants had all the vascular risk factors the researchers recorded.
Putting the algorithm to the test, the researchers saw a higher risk of AF in the MwA group, compared to the MwoA group. Specifically, the algorithm indicated that MwA patients had a 7.3% chance of having subclinical AF, compared to 5.6% for MwoA, for a difference of 1.7%.
Notably, a significant difference in AF risk between the two subgroups of migraine patients remained in three of the four age subgroups in the study (18 to under 35 years of age, 35 to under 55 years, and 55 to under 75 years), with no difference in those 75 years and older. After adjusting for six vascular risk factors, the difference between groups remained significant for the overall age group, those 18 to under 35 years of age, and those 35 to under 55 years.
“I am particularly impressed that they found differences in a relatively young age group,” said Tietjen, referring to patients under 55 years of age, and even under 35 years of age.
The difference in AF risk also remained significant when parsing the data by sex. For the women-only group with MwA, there was a greater probability of AF compared to women with MwoA (a 6.4% chance versus a 5% chance, respectively). After adjusting for vascular risk factors, the difference remained significant for women of all ages and for women 35 to under 55 years of age. In the men-only group, the difference between MwA and MwoA remained significant for men of all ages, and for men 18 to under 35 years of age.
How good was the algorithm? The researchers reported an accuracy, sensitivity and specificity of over 80% for each measure. This, the authors say, compares favorably to what's reported in the literature for other medical screening tests. One limitation Chiang noted was that the original ECGs used to train the algorithm came from patients who were older (an average age of about 60 years) than the cohort in her study (an average age of 45).
领英推荐
“Going forward we want to further examine the accuracy specifically in our migraine population,” she said.
Some big questions remain
Chiang said a couple of results from the study particularly stood out to her.
“The most important finding is that migraine with aura is an independent risk factor for subclinical or undiagnosed atrial fibrillation. We found that was true, even after we adjusted for age, sex, and vascular risk factors,” Chiang said.
Chiang also highlighted the finding that the difference in AF risk between MwA and MwoA “is most notable in younger patients under 55 years old. That is very important information because previously it’s been quite challenging to study the association of migraine and atrial fibrillation, especially in the younger population,” she said.
As for Tietjen, while excited by the results, she said that one key question is the nature of the association between AF and MwA.
“What exactly is the relationship? Is migraine with aura causally related to atrial fibrillation, or is atrial fibrillation causally related to migraine with aura? It may be both,” she said, pointing to a possible bidirectional relationship.
Tietjen also noted a lack of information in the study about aura frequency, age of onset of aura, and the duration of migraine, which are limitations that the authors acknowledge.
Another limitation that Tietjen mentioned, and again one that the authors acknowledge, is that it is unclear as of yet if the algorithm can predict the future risk of AF, since the study only looked at concurrent paroxysmal or impending AF. However, Chiang is encouraged in this regard.
“There have been subsequent studies done at Mayo Clinic verifying that there is actually a direct association between the prediction model output that we see on ECGs and the future risk of atrial fibrillation.”
Into the clinic?
What would implementing the findings in the current study look like clinically?
“If we see patients who have a high frequency of migraine aura, or they have new development of migraine aura, and then I look at their ECG and it shows a high probability of atrial fibrillation, then I would be more inclined to arrange a workup – for example, prolonged cardiac monitoring to see whether the patient actually has atrial fibrillation that is associated with the increased frequency of aura,” Chiang said.
Chiang also said she could envision ordering an MRI of the brain to see if there were other signs that a patient may have had paroxysmal AF. Further testing, such as new ECGs, could also be in order.
Tietjen said particular attention could be paid to younger migraine subgroups who may have a significant risk of stroke.
“There's highly likely a group of people with migraine with aura who could be identified before they have a stroke. Treating them and preventing that – that's where I'd love to see this go.”
But before the use of AI-ECG becomes a reality in the migraine field, more research is in order. For instance, what threshold for the probability of having AF should be used to identify a migraine population that merits additional attention? Chiang is planning follow-up research to answer that and other questions.
In the end, the research shows that a new approach like AI can come together with an older one – in this case, the simple, inexpensive and humble ECG – to benefit patients.
“This study is an example of how we can apply novel computer technology to a very traditional and noninvasive test, like an ECG, to study the association between different disorders,” Chiang said.
Neil Andrews is a science journalist and executive editor of the Migraine Science Collaborative.
This article originally appeared on the Migraine Science Collaborative.
Reference