Multimodal Fusion Artificial Intelligence Model to Predict Risk for MACE and Myocarditis in Cancer Patients Receiving Immune Checkpoint Inhibitor Therapy.

Background: Immune checkpoint inhibitor (ICI) therapy has dramatically improved the prognosis for some cancers but can be associated with myocarditis, adverse cardiovascular events, and mortality. Objectives: The aim of this study was to develop an artificial intelligence (AI) model to predict the increased likelihood for the development of ICI-related myocarditis and adverse cardiovascular events. Methods: Cancer patients treated with ICI at a tertiary institution from 2011 to 2022 were reviewed. Baseline characteristics, laboratory values, electrocardiograms, and cardiovascular clinical outcomes were extracted. A composite outcome of ICI-related myocarditis and major adverse cardiovascular events (transient ischemic attack/stroke, new diagnosis of heart failure, myocardial infarction, and cardiac death) was used to develop a multimodal joint fusion AI model by combining baseline tabular data with electrocardiogram (ECG) in a single end-to-end model. ECG data were parsed using 1-D convolution and tubular data using multilayer perceptron. Results: Of 2,258 cancer patients who had ICI therapy and troponin measurement (mean age 68.5 ± 11.5 years, 59.7% male), the composite of cardiovascular clinical adverse events, including ICI-related myocarditis and major adverse cardiovascular events, occurred in 264 (11.7%) unique patients, with 428 events overall (including 59 [3%] ICI-related myocarditis events and 59 [3%] cardiac deaths). The proposed joint fusion model outperformed individual ECG and baseline electronic medical record data and laboratory value models with an area under the operating characteristics curve of 0.72 (0.64 true positive rate and 0.98 negative predictive value). Conclusion: A multimodal fusion AI model to predict myocarditis and adverse cardiovascular events in cancer patients starting ICI therapy had good prognostic performance. It may have clinical utility in identifying at-risk patients who may benefit from closer surveillance.