AAN 2025 | AI-ECG models to predict cardiovascular events in patients with migraine

At our institution at Mayo Clinic we do have AI ECG models that can predict the probability of undiagnosed paroxysmal atrial fibrillation. That is one algorithm and we also have another algorithm that can estimate the patient’s physiological age just based on a single ECG that shows normal sinus rhythm. As far as the AI ECG estimated age, there is also something called the delta age, meaning that you use the ECG estimated age minus the actual age of the patient, and then you compute the difference between those two ages. The delta age in previous studies have been shown to correlate or is a marker of endothelial dysfunction and associate with adverse vascular outcomes. So previously, in our preliminary studies or pilot studies, we demonstrated that patients with migraine with aura have a higher AI-ECG prediction model output for atrial fibrillation compared to migraine without aura. In a separate study, we also showed that patients with migraine with aura, especially those with older age, have a higher delta age compared to patients with migraine without aura. So in this particular study, we wanted to move these pilot studies forward and see whether we can use the AI-ECG model as a tool to screen patients with migraine who are at risk or at a higher risk to develop vascular events. So we found that based on ECGs, one single ECG that showed normal sinus rhythm in patients with migraine, we found that those who have a higher AI-ECG atrial fibrillation prediction model output and also who have a higher delta age as computed by the age estimation algorithm, those two markers are predictors for patients to develop future cardiovascular events in patients with migraine. And this is based on a retrospective EHR study for everyone at Mayo Clinic who had at least one migraine diagnosis code and also at least one normal sinus rhythm ECG conducted at Mayo Clinic. We also identified through this study the adequate cutoff for both AI ECG models. So for example, for our atrial fibrillation model, we identified the cutoff to be 1% probability of atrial fibrillation, and for delta age, we found the adequate cutoff value to be minus 1. So, based on this cutoff value, we can separate patients into high-risk and low-risk groups for the future risk of cardiovascular events. Another important study, another important finding we identified through this study is that we also identified that those and confirmed that those who had a higher baseline atrial fibrillation prediction model output associate with higher future risk of actually developing atrial fibrillation throughout a 10-year follow-up period. So I think overall the study is showing that for patients with migraine based on a single normal sinus rhythm ECG when they have the AI ECG prediction model output for the risk of atrial fibrillation and for the delta age we can use those values to separate patients into high-risk and lower-risk groups and we also based on the cutoff values that we identify and we also confirmed that those who have a higher baseline AI ECG atrial fibrillation prediction model output correlate or associate with a higher future risk of atrial fibrillation and then for those who were confirmed to have atrial fibrillation that increase the risk of adverse vascular events 2.5 times higher than those who do not and the vascular events that we defined in this study include acute ischemic stroke, myocardial infarction, DVT/PE, and also mortality.

This transcript is AI-generated. While we strive for accuracy, please verify this copy with the video.