(132) Exploring AI-Enabled Electrocardiogram Prediction to Detect Cardiovascular Effects of CAR-T Therapy

Introduction: Clinical trials have reported adverse cardiovascular effects with CAR- T immunotherapy. We aim to explore such effects in patients undergoing CAR-T immunotherapy and the utility of pre-existing AI tool validated to identify atrial fibrillation and low ejection fraction in this context.

Methods: We conducted a retrospective study on patients receiving CAR- T immunotherapy at Mayo Clinic between October 2017 and October 2021 (IRB21-000970). Demographics, baseline laboratory, clinical and outcome variables were collected by retrospective chart review corresponding to the admission for receiving CAR- T therapy. AI-dashboard which is inbuilt in this institution’s EMR, predicts low ejection fraction ( < 35%) and atrial fibrillation probability based on EKG data. All statistical analyses were performed on BlueSky v10.3.1

Results: Our study cohort consisted of 214 patients, of which the majority were males (60.7%) and White (90.1%). Median C reactive peptide (13.75 vs. 76.75 mg/dl, p< 0.001) and ferritin levels (257 vs 693 mcg/L p=0.001) increased significantly after CAR T therapy. ​ Nonsurvivors (n=114) had a higher incidence of heart failure, baseline CRP (19.50 vs. 10.80, p=0.015), baseline ferritin (394 vs. 159, p< 0.001), and post-CART ferritin (1092 vs. 479, p=0.004)​. However, other comorbidities, baseline demographics, pre- and post-CART ventricular size and functions (both LV and RV) did not vary between survivors and non-survivors.
There was a significant increase in the AI-assisted probability of atrial fibrillation (3.14% vs. 6.60%, p=0.021) and low ejection fraction (1.13% vs. 1.89%, p=0.021) after receiving CAR T therapy. ​A linear regression model comparing AI-predicted low ejection fraction probability and echocardiography-derived ejection fraction value revealed a linear relation (p < 0.001). Each 1% increase in AI predicted low EF probability was associated with 0.16% drop in actual ejection fraction during the pre-CART period and a 0.21% drop in the post-CART period.

Conclusions: This retrospective study showed a linear association between AI-predicted low ejection fraction probability and echo-guided ejection fraction value. Despite a difference in such pre-CART and post-CART probabilities, no differences were noted between survivors and non-survivors.