Integrating Imaging Markers for Clinical Risk Stratification of Takotsubo Syndrome

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Informatics

Background:

• Takotsubo syndrome (TTS) is a form of acute heart dysfunction.
• Predicting adverse cardiovascular events in TTS patients is crucial for management.
• Integrating diverse patient data can improve predictive accuracy.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting 3-month major adverse cardiovascular events (MACE) in TTS patients.
• To integrate clinical and echocardiographic markers for enhanced prognostic value.
• To assess the interpretability and clinical utility of the developed ML models.

Main Methods:

• Retrospective development of ML models using two datasets (n=450) with clinical and echocardiographic data.
• Random Forest algorithm employed for feature selection and prediction of 3-month MACE.
• Model performance evaluated using AUC, calibration, sensitivity, specificity, and SHAP for interpretability.

Main Results:

• ML models integrating clinical and echocardiographic variables showed good discrimination for 3-month MACE.
• The MLCE-4 model (including E wave deceleration time, SBP, RVFAC, HR) achieved AUCs of 0.80 (internal) and 0.76 (external).
• SHAP analysis identified key predictors, and thresholds for elevated MACE risk were established.

Conclusions:

• ML-based prediction models integrating clinical and echocardiographic data demonstrate prognostic value.
• These models can aid in early risk stratification for hospitalized TTS patients.
• The findings support the use of ML in managing Takotsubo syndrome.