An XGBoost machine learning model using 76 clinical features accurately predicted acquired Long QT Syndrome (AUC 0.94) and severe acquired Long QT Syndrome (AUC 0.92) among hospitalized patients.
Observational (n=96,531)
No
Does an XGBoost machine learning model accurately predict acquired Long QT Syndrome in hospitalized patients?
An XGBoost machine learning model using clinical features demonstrated excellent performance (AUC 0.94) in predicting acquired Long QT Syndrome among hospitalized patients.
Effect estimate: AUC 0.94 for aLQTS; AUC 0.92 for saLQTS
Abstract Background Acquired long QT syndrome (aLQTS) represents a significant clinical concern due to its association with malignant arrhythmias and increased all-cause mortality. Various factors can lead to QT interval prolongation. Objective This study aimed to analyze the clinical characteristics of hospitalized patients with aLQTS and to develop a predictive model for identifying aLQTS in this population. Methods Electronic medical records from January 2017 to October 2021 were reviewed to identify cases of aLQTS (defined as QTc ≥470 ms for males and QTc ≥480 ms for females) and severe aLQTS (saLQTS, defined as QTc ≥500 ms) among hospitalized patients at a single center. A control group with normal QTc intervals (360 ms ≤ QTc ≤ 440 ms) was also identified. Using data from 76 features, including demographic information, comorbidities, laboratory test results, and echocardiographic data, eXtreme Gradient Boosting (XGBoost) models were developed for predicting aLQTS and saLQTS. The dataset was split into training and validation sets at a ratio of 7:3. Optimal hyperparameters were selected using 5-fold cross-validation and Bayesian optimization on the training set. Feature importance was determined by the frequency at which each feature was used as a splitting node across all trees in the XGBoost model. Results Among 96,531 hospitalized patients, 3.2% (n=3,089) had aLQTS, and 0.7% (n=731) had saLQTS. Approximately 29.7% of these cases were managed in the cardiology department. The baseline characteristics are detailed in Figure 1. Adverse events, including sudden cardiac arrest, ventricular tachycardia, ventricular fibrillation, and death, occurred in 2.8% (86/3089) of hospitalized aLQTS patients, with the all-cause mortality rate of 1.6% (50/3089). In the validation set, the model achieved an AUC of 0.94 for predicting aLQTS and 0.92 for predicting saLQTS. Sensitivity and specificity were 0.73 and 0.95 for aLQTS, and 0.73 and 0.90 for saLQTS. When sensitivity was set to 0.80, the specificity was 0.92 for aLQTS and 0.89 for saLQTS. The top 20 predictive features are presented in Figure 2, with drugs, heart rate, NT-proBNP, myocardial infarction, and serum phosphorus being leading features in both models. Conclusion The XGBoost models for predicting aLQTS and saLQTS demonstrate excellent performance, aiding healthcare professionals in identifying high-risk patients prone to QT interval prolongation. This enables proactive management of reversible factors such as infections, electrolyte imbalances, and medications known to pose risks for QT prolongation.Characteristics of Hospitalized Patients XGBoost Models for aLQTS patients
Luo et al. (2025) conducted an observational in Acquired Long QT Syndrome (aLQTS) (n=96,531). eXtreme Gradient Boosting (XGBoost) predictive model vs. Control group with normal QTc intervals (360-440 ms) was evaluated on Prediction of aLQTS and severe aLQTS (saLQTS) (AUC 0.94 for aLQTS; AUC 0.92 for saLQTS). An XGBoost machine learning model using 76 clinical features accurately predicted acquired Long QT Syndrome (AUC 0.94) and severe acquired Long QT Syndrome (AUC 0.92) among hospitalized patients.