Among eight machine learning models developed to predict major adverse cardiovascular events in dialysis patients, AdaBoost demonstrated superior performance, with NT-proBNP, eGFR, global longitudinal strain, and age identified as the four most important predictive features.
The incidence of MACE among dialysis patients in this cohort was 38.92% over an average follow-up period of 18 months.
LASSO regression identified eight feature variables from general information, laboratory tests, and echocardiographic parameters including global longitudinal strain (GLS).
The AdaBoost model demonstrated superior performance among the eight machine learning models evaluated.
SHAP analysis identified NT-proBNP, eGFR, GLS, and age as the four most important features for predicting MACE in dialysis patients.
Elevated NT-proBNP, advanced age, reduced eGFR, and impaired GLS were independently associated with an increased risk of MACE in patients undergoing dialysis.
This research suggests that machine learning models can effectively predict serious heart events in patients undergoing dialysis (kidney replacement therapy). The study followed 203 dialysis patients for an average of 18 months and found that nearly 4 in 10 experienced a major adverse cardiovascular event (MACE), which included heart attacks, unstable chest pain, heart failure, or cardiovascular death. Eight different machine learning approaches were tested, and the AdaBoost model performed best, correctly identifying patients at risk roughly 80–88% of the time depending on the dataset used. The study also identified which patient characteristics were most useful for making these predictions. The four most important factors were: NT-proBNP (a blood marker of heart stress), estimated kidney filtration rate (eGFR), global longitudinal strain (GLS, a measure of how well the heart muscle squeezes derived from ultrasound imaging), and age. Patients with higher NT-proBNP levels, older age, worse kidney function, and poorer heart muscle function had a greater risk of experiencing a serious cardiovascular event. This research suggests that combining routine blood tests, kidney function measures, and specialized heart ultrasound data into machine learning models could help clinicians better identify dialysis patients who are at high risk for cardiovascular complications. The inclusion of GLS — a relatively detailed echocardiographic measurement — as an important predictor highlights the potential value of cardiac imaging in the routine monitoring of this high-risk patient population.
Jin M, Lin Z, Ma L, Li B, Huang X, Chen M. (2026). Multiple machine learning models for predicting major adverse cardiovascular events in dialysis with clinical and echocardiographic parameters: a retrospective cohort study.. Annals of medicine. https://doi.org/10.1080/07853890.2026.2698905