Recurrent pericarditis (RP) is a complex disease with high morbidity

Previous studies have assessed which variables are associated with clinical remission

However, there are currently no established risk stratification models to predict outcomes in these patients

Developed a risk stratification model that predicts long-term outcomes in patients with RP and is able to identify patients with characteristics predictive of adverse outcomes

A total of 365 consecutive RP patients from 2012 to 2019 were retrospectively studied

The primary outcome was clinical response (CR), defined as discontinuation of all anti-inflammatory treatments and complete resolution of symptoms

Five machine learning survival models were used to calculate the likelihood of CR within 5 years and classify patients into high-risk, intermediate-risk, and low-risk groups

In this cohort, the mean age was 46±15 years, and 205 (56%) were female. 118 patients (32%) achieved CR

The final model included steroid dependence, total number of relapses, late pericardial gadolinium enhancement, age, etiology, sex, ejection fraction and heart rate as the most important parameters

The model predicted outcome with a C-index of 0.800 on the test set and showed significant ability to stratify patients into low-, intermediate-, and high-risk groups (log-rank test; P < 0.0001)

A new risk stratification model was developed to predict CR in RP.

Our model can also help stratify patients with high discriminatory power

Using interpretable machine learning models can help physicians make personalized treatment decisions for RP patients

Baseline characteristics of the overall cohort and outcome groups

Compare the predictive performance of different models

Research flow chart describing research workflow

Global interpretation and feature importance of Xgb model

Global interpretability of XGB models, and the 20 most influential features

Draw SHAP feature dependency graph

The impact of certain continuous variables on the predictive model output is shown in a SHAP dependency plot

Risk models and model evaluation

Data segmentation strategy

Kaplan-Meier curves for specific parameters

Assess the model's ability to independently risk stratify within cohort subgroups through subgroup analyzes based on each variable

The model demonstrates robust performance in risk stratification of cohort subgroups, showing clinical response rates decrease proportionally with increasing risk score

Kaplan-Meier curves of training set and test set score groups

The model demonstrates robust performance in risk stratification of cohort subgroups

Calibration plot of the final model in the test cohort

Baseline characteristics of the training set and test set

The hyperparameter space of the model

Exported scores from the final Cox proportional hazards model

Baseline characteristics of risk groups

Annual cumulative incidence of clinical remission by risk stratification over 5 years