Cardiovascular toxicities in cancer patients treated with immune checkpoint inhibitors: multicenter study using natural language processing on Belgian hospital data.

Background: Immune checkpoint inhibitor (ICI) use may be associated with diverse cardiovascular (CV) adverse events (AEs), but their baseline prevalence and incidence after ICI initiation are poorly known. We aimed to describe CV events using real-world hospital data from Belgian cancer patients. Materials and methods: Electronic health records (EHRs) from patients receiving at least one ICI between March 2017 and August 2022 at three Belgian hospitals were processed into an Observational Medical Outcomes Partnership Common Data Model warehouse. Structured data were enriched with unstructured data that were processed using a natural language processing (NLP) pipeline. We analyzed CV events from first ICI administration until last follow-up, identifying and validating the first detection of a CV event at the patient level. Results: We included 1571 patients (66% male, median age 67 years); CV events were detected in 196 (12.5%) patients [median (min-max) follow-up: 8 (0-63) months]. The CV AEs detected were heart failure (5.3%), atrial fibrillation (4.6%), myocardial infarction (2.0%), atrioventricular block (1.9%), myocarditis (1.2%), vasculitis (0.8%), pericarditis (0.4%), and Takotsubo cardiomyopathy (<0.3%). Median time (min-max) to onset ranged from 109 days (17-849 days) for myocarditis to 529 days (91-967 days) for Takotsubo cardiomyopathy. Conclusions: To our knowledge, this is the first study using a dataset enriched with NLP-processed EHRs that describes the frequency and onset time of CV events. CV event frequencies were higher than those reported in clinical trials, but similar to other real-world studies. However, we observed a later time to onset. Hence, clinicians should note that CV AEs can present in various ways and at any time during or after treatment.