Patients with newly diagnosed chronic lymphocytic leukemia (CLL) have an estimated 1% risk per year of atrial fibrillation (AF) incidence. Now that artificial intelligence electrocardiography (AI-ECG) has emerged as a tool for predicting AF in the general population, researchers hypothesized that it may also be beneficial in identifying those patients with newly diagnosed CLL who are most at risk for AF. New research indicates that an AI-ECG algorithm, in conjunction with the Mayo CLL AF risk score, can predict the risk of AF in patients with newly diagnosed CLL.1 Georgios Christopoulos, MD, of Northwestern Medicine in Chicago, and colleagues published the results of the retrospective single-center study in JACC: CardioOncology.
The study included 754 patients newly diagnosed with CLL (71.4% male; median age = 69 years). The median baseline AI-ECG score was 0.02 (range = 0-0.93), with a value of at least 0.1, indicating high risk. When the investigators calculated the c-statistic (increased probability of having the outcome relative to the general population), they found that the AI-ECG model had a discriminative ability of 0.62, similar to the c-statistic value of the Mayo CLL AF score alone.2 The two combined had a c-statistic of 0.71.
The authors explained in their discussion that because an AI-ECG output of 0.1 reflects a 10% AF probability within 30 days of obtaining the ECG, the AI-ECG outcome may be underestimating AF incidence over the years.1 They also acknowledge that because AF is highly heterogeneous, it is conceivable that “provoked” AF, such as postoperative AF or the AF induced by Bruton tyrosine kinase inhibitors (BTKis) in this study, may not be accurately predicted using an AI model initially trained on the general population. Consistent with this, the researchers found that, unlike the general CLL cohort, a baseline AI-ECG score of 0.1 or greater did not demonstrate predictive ability for developing AF in patients with CLL starting on BTKis. The Mayo Clinic criteria, however, remained predictive in this patient population.
Joerg Herrmann, MD, a cardiologist at Mayo Clinic in Rochester, Minnesota, and senior investigator on the study, said the inability of AI-ECG to accurately predict risk in patients treated with BTKis may reflect the fact that there were not enough patients in the cohort receiving treatment with BTKis to reach statistical analysis. Another possibility, he said, is that BTKis introduce an element of AF risk that the AI algorithm has not been trained on, as it was developed in the general population without such exposure. The absence of such data thus acts like a “blind spot” when pursuing a focused view. Dr. Herrmann said more research is needed to determine the role of AI-ECG in predicting AF risk in patients with CLL treated with BTKIs.
Dr. Herrmann also explained that the paper “shows both the possibilities and limitations of AI.” He pointed to the importance of shared decision-making between patients and their care teams when choosing medications, with the hematologists, as prescribers, being the most essential members of the team. Thus, the intention is for tools such as AI-ECG to aid in this decision-making and help — rather than complicate — clinical practice.
“In this day and age, we are looking for accurate risk prediction, particularly in identifying those individuals at high risk of cardiovascular complications with one type of therapy who then may be better served with another; or at least, the risk and benefit can be discussed in a more informed way,” Dr. Herrmann said.
He proposed that AI-ECG can be useful for that purpose but needs further prospective evaluation for validation and integration into the clinical workflow, which includes its combination with other predictive tools. For instance, in this study, patients not identified as high risk for AF by the AI-ECG algorithm could be risk stratified further using the Mayo Clinic CLL AF score. One could thus see a staggered or “cascade” approach to optimize risk prediction and individualization of care.
Any conflicts of interest declared by the authors can be found in the original article.
References
- Christopoulos G, Attia ZI, Achenbach SJ, et al. Artificial intelligence electrocardiography to predict atrial fibrillation in patients with chronic lymphocytic leukemia. JACC CardioOncol. 2024;6(2):251-263.
- Barrios JP, Seshadri MR, Tison GH. Artificial intelligence to complement, not replace, clinical knowledge: reading between the lines. JACC CardioOncol. 2024;6(2):264-266.