A MACHINE learning model can accurately predict in-hospital mortality in patients with Takotsubo cardiomyopathy (broken heart syndrome), identifying cardiogenic shock and acute kidney injury as key risk factors, according to a National Inpatient Sample analysis of 38,662 cases.
Takotsubo cardiomyopathy (TC), a stress-induced heart failure syndrome, carries significant in-hospital mortality (6.5%) and complications, yet tools for risk stratification remain limited. This study developed a novel machine learning (ML) risk-prediction model using data from 38,662 TC patients (mean age 67 years, 83% female) to address this gap. The model’s performance highlights its potential to guide clinical decision-making for this high-risk population.
The ML model analysed 38,662 Takotsubo patients (83% female, mean age 67), identifying eight mortality predictors: age, race, Elixhauser comorbidity index, hypertension, arrhythmia, cardiac arrest, cardiogenic shock, and acute kidney injury. The model achieved robust performance across datasets (AUC 0.809–0.838), with cardiogenic shock showing the strongest association (OR=12.7, p<0.001). Acute kidney injury (OR=3.2) and cardiac arrest (OR=4.8) were also critical predictors. Mortality risk scores (0–127) stratified patients into tiers, ranging from 2% (low-risk) to >35% (high-risk) mortality probability.
These findings suggest ML-based risk stratification could optimize TC management by identifying high-risk patients for intensive monitoring. Clinicians should prioritise early intervention for patients scoring above 50 points, particularly those with shock or renal impairment. Future research must validate this model in diverse populations and integrate real-time electronic health record data for dynamic risk assessment. Health systems should explore embedding such tools into clinical workflows to reduce TC’s 6.5% mortality rate, which has remained stagnant since 2016.
Reference
Agrawal A et al. Machine learning risk-prediction model for in-hospital mortality in Takotsubo cardiomyopathy. Int J Cardiol. 2025;133181.
