Utilizing AI-Enabled EKG Algorithms to Identify Atrial Fibrillation Risk in Patients with CLL Receiving BTK Inhibitors

INTRODUCTION

Atrial fibrillation (AFib) is a recognized complication in patients with chronic lymphocytic leukemia (CLL) treated with Bruton tyrosine kinase inhibitors (BTKis). Early detection and management of AFib are critical due to its association with increased morbidity and mortality. Traditional detection methods, such as intermittent EKGs and wearable devices, often capture only brief snapshots of cardiac activity, potentially missing episodes. Recent advances in artificial intelligence (AI) have enabled the development of algorithms that analyze electrocardiograms (EKGs) to predict AFib risk. These algorithms can potentially identify patients with undetected AFib and assess risk at a single point, reducing the need for continuous monitoring. This study evaluates the utility of an AI-enabled EKG algorithm in predicting AFib in patients with CLL treated with ibrutinib and acalabrutinib.

METHODS

Patients with CLL treated with ibrutinib and acalabrutinib were identified from a single-center database. Patients with CLL who received BTKi therapy at Mayo Clinic between 2015 and 2024 with available EKGs in the EMR were included. Patients were excluded if they received BTKi for non-CLL indications or if patients with CLL had documented AFib without accompanying EKGs in their medical records. An AI-enabled EKG algorithm developed by Attia et al. (Lancet, 2019) was used to estimate AFib probability (AFibAI). EKGs were categorized into pre-treatment, during treatment, and post-treatment phases. Mean values of all EKGs within each phase were used for analysis to reduce heterogeneity. Paired t-test was used for continuous variables and ROC analysis was used to evaluate diagnostic performance. A p-value of <0.05 was used for statistical analysis.

RESULTS

The analysis included 383 individual CLL patients (74% male, median age 71; ace/ethnicity not available), of which 79 were treated with acalabrutinib and 304 with ibrutinib, with a total of 5844 EKGs analyzed. The AFibAI was significantly higher in EKGs with documented AFib compared to other EKGs, mostly normal sinus rhythm (0.5710 vs. 0.1857, p <0.001). Pre-treatment AFibAI was higher in patients who ultimately developed AFib after starting BTKi treatment compared to those who did not. Mean AFibAI at baseline was 0.2176 for those who developed AFib and 0.0986 for those who did not (p <0.001; difference between means: -0.1191 +/- 0.02005, 95% CI: 0.-0.1585 – 0.0796). Differences of baseline AFibAI were similar when stratified by BTKi. Trends in AFibAI scores indicated a significant increase during and after BTKi treatment in patients who developed AFib, with mean score of 0.2176 at baseline to 0.4064 post-treatment (p<0.001). In patients who did not develop Afib during BTKi therapy, there was minimal increase of AFibAI scores from 0.1041 at baseline to 0.1620 post-treatment (p = 0.0002).

The ROC analysis showed an area under the curve of 0.7280 (SE = 0.03911, 95% CI: 0.6514 to 0.8047, p <0.001), indicating fair discriminatory ability. The sensitivity at the optimal threshold (0.1202) was 65.45% and the specificity was 74.59%, with a Youden’s Index of 0.4004. To achieve a sensitivity of ≥90% for detecting patients who ultimately develop AFib, a threshold of 0.2368 was established (specificity of 41.82%).

DISCUSSION

Our study demonstrates the efficacy of an AI-enabled EKG algorithm in estimating atrial fibrillation risk among CLL patients treated with BTKis. We demonstrated that baseline AFibAI was notably higher in patients who ultimately developed AFib, suggesting that it could potentially identify patients at elevated risk before the onset of AFib. Early identification would allow for closer monitoring and timely intervention, potentially mitigating complications. Pending validation with an external cohort, the 90% sensitivity threshold could help identify the patients at highest risk, where other treatment options could be considered. The trend across different treatment phases revealed a significant increase in scores during treatment, indicating that AFibAI is responsive to changes induced by BTKi therapy and suggest its utility for patient monitoring. Practical questions remain about its impact on clinical decision-making and management. Future prospective studies should confirm the validity of identifying high-risk patients and its role in both treatment decisions and patient outcomes.