16: Artificial intelligence modelling to predict the risk of cardiotoxicity among renal cell carcinoma patients treated with vascular endothelial growth factor receptors tyrosine kinase inhibitors.
Clinical Fellow Vanderbilt University Medical Center Nashville, TN, United States
Background: Vascular endothelial growth factor receptors tyrosine kinase inhibitors (VEGFRi) are standards of care in renal cell carcinoma (RCC). Despite efficacy and safety, there is a risk for cardiotoxicity with an estimated incidence between 3-8%. Cardiotoxicity can occur months or years after treatment and can be life threatening. Historically, an oncologist refers patients to a cardiologist when cardiotoxicity is suspected/observed. However, general cardiologists may be unfamiliar with VEGFRi cardiotoxicity. Cardio-oncology, an emerging subspecialty, aims to prevent and/or treat cardiovascular complications among cancer patients/survivors. Great progress has been made in utilizing this subspecialty, but standardized referral workflow is limited. Machine learning (ML) is a discipline of artificial intelligence (AI) and computer science that utilizes algorithms to find patterns in data and help create models to predict future events. Using AI may help identify patients who face risk of cardiotoxicity to promote referral to cardio-oncology in a timely manner.
Methods: De-identified data on RCC patients were obtained from Vanderbilt University Medical Center EMR. Random forest (RF) and artificial neural network (ANN) ML models were applied to analyze the cohort. A global team of cardio-oncologists devised cardiotoxicity risk factors used in calculating the risk groups (potential/mild/moderate/major) (see table).
Results: 2,047 RCC records were analyzed. Data was randomly divided into training (80%) and validation (20%) sets. RF and ANN, applied to analyze patient records extracted to specifications outlined in the clinical risk model, performed > 95% for accuracy and at >94% precision. Limited validation showed 58% of the RCC patients treated with VEGFRi with major risk for cardiotoxicity not referred to cardio-oncology.
Conclusions: AI models accurately predict RCC patients with cardiotoxicity risk. Integration of AI models into EMR can assist oncologists with identifying these patients and referring them for proactive cardio-oncology treatment/monitoring.
