Abstract 14951: An Interpretable Machine Learning Model for Predicting Long-Term Clinical Outcomes in Recurrent Pericarditis

Abstract

Background: Recurrent pericarditis (RP) is a complex condition associated with high morbidity. A few studies have evaluated which factors are associated with clinical remission and relapse. However currently, no models exist to predict these outcomes. Objective: We developed an explainable machine learning (ML) model that can predict long-term outcomes in patients with RP and enable identification of patients with worse outcomes. Methods: Consecutive patients with RP (n=365) from 2012-2019 included. The primary outcome was CR. The dataset was partitioned into training and evaluation sets with 5-fold cross-validation approach. The eXtreme Gradient Boosting (XGBoost) method was used to identify predictors for CR, calculate the likelihood of CR within 5 years, and stratify patients into high-risk and low-risk. The performance of the model was evaluated using the C-index. The log-rank test was utilized to assess the difference between Kaplan-Meier curves of risk groups. Finally, an explainable ML approach, SHAP, was deployed to generate both global and individual explanations of the model’s decisions. Results: The cohort was 56% female (n=205) with a mean age of 46 ± 15 years. 61% were idiopathic RP (n=223), 21% were post-cardiac injury syndrome RP (n=77), and 18% were autoimmune RP (n=65) patients. CR was achieved in 32% (n=118). XGBoost risk model included steroid dependency, total no. of recurrences, age, HR, LVEF, gender, etiology, and pericardial delayed gadolinium hyperenhancement as important parameters. The model predicted the outcome with C-index of 0.801 (95%CI, 0.74-0.86) on the validation set and exhibited significant ability in stratification of patients into low- and high-risk groups (log-rank P<0.001). Conclusions: We developed a novel ML-based model for predicting CR, and stratifying patients, with high discrimination ability. The use of an explainable ML model can aid physicians in making individualized treatment decision in RP patients.