TY  - JOUR
AU  - Liu, Xiao
AU  - Zhang, Zenghui
AU  - Cui, Mingyue
AU  - Shu, Chunjie
AU  - Yu, Weiliang
AU  - Pan, Hong
AU  - Abudukeremu, Ayiguli
AU  - Zhao, Ke
AU  - Zheng, Minglong
AU  - Cao, Zhengyu
AU  - Wang, Jingfeng
AU  - Zhang, Yuling
AU  - Chen, Yangxin
PY  - 2026
TI  - Patient‐Reported Quality of Life‐Based Machine Learning Model Predicting Sudden Cardiac Death in Heart Failure With Preserved Ejection Fraction: Kansas City Cardiomyopathy Questionnaire‐Based Sudden Cardiac Death Score
JO  - Medicine Advances
SN  - 2834-4391
SP  - 352
EP  - 362
VL  - 4
IS  - 3
AB  - BackgroundSudden cardiac death (SCD) is the most frequent cause of mortality in patients with heart failure with preserved ejection fraction (HFpEF). While the Kansas City Cardiomyopathy Questionnaire (KCCQ) assesses disease severity in HFpEF, its ability to predict SCD remains unclear. We aimed to develop a machine learning model to stratify the risk of SCD in HFpEF using patients' reported quality of life scores.MethodsUsing data from the Treatment of Preserved Cardiac Function Heart Failure with an Aldosterone Antagonist trial, we developed six models (convolutional neural network [CNN], logistic regression, LASSO‐regularized logistic regression, random forest, gradient boosting, and K‐nearest neighbors) to predict SCD in HFpEF, using age, sex, and the KCCQ score.ResultsAmong 3445 patients (mean age: 69.1 years, 48.5% men), 111 experienced SCD over a mean follow‐up of 3.37 years. The CNN model outperformed the other models, with a C‐statistic of 0.74 (95% confidence interval [CI]: 0.65–0.83), followed by the logistic regression model (0.66, 95% CI: 0.56–0.76), XGBoost (0.65, 95% CI: 0.56–0.75), Light‐GBM (0.62, 95% CI: 0.51–0.74), random forest (0.54, 95% CI: 0.41–0.66), and K‐nearest neighbors (0.54, 95% CI: 0.43–0.64). The total symptom score, social limitation score, self‐efficacy score, and overall summary score were ranked as the most important variables. The KCCQ‐SCD score was associated with a fivefold higher risk of SCD (hazard ratio: 5.69, 95% CI: 1.92–16.81). An online tool to implement the CNN model is available at https://huggingface.co/spaces/KCCQ/KCCQ_SCD_Predictor.ConclusionsA CNN‐based machine learning model incorporating age, sex, and KCCQ scores provides a simple and accurate tool for stratifying the risk of SCD in patients with HFpEF. External validation in more diverse populations and real‐world clinical settings is essential before the KCCQ‐SCD score is used for broad clinical applications.
UR  - https://doi.org/10.1002/med4.70081
DO  - 10.1002/med4.70081