KMAN Publication Inc. Quality of Life and Health Sciences 1 Serial Number 2 2025 10 01 Machine Learning Prediction of Low Quality of Life in Patients With Chronic Kidney Disease Using XGBoost and Explainable Artificial Intelligence Machine Learning Prediction of Low Quality of Life in Patients With Chronic Kidney Disease Using XGBoost and Explainable Artificial Intelligence 1 14 10.61838/smrh9f43 EN Journal Article 2025 04 07 <table> <tbody> <tr> <td> <p><strong>Objective:</strong> This study aimed to predict low quality of life among patients with chronic kidney disease using the XGBoost machine learning algorithm and to identify the most important predictors through SHAP-based explainable artificial intelligence analysis.</p> <p><strong>Methods and Materials: </strong>This cross-sectional predictive modeling study was conducted among 426 patients with chronic kidney disease in Tehran, Iran. Data were collected using a demographic and clinical information checklist, medical record data, laboratory indicators, and the Kidney Disease Quality of Life questionnaire. Low quality of life was defined as a total quality-of-life score below 50 and was used as the binary outcome variable. After data preprocessing, missing-value management, and encoding of categorical variables, the dataset was divided into training and testing sets. Logistic regression, support vector machine, random forest, and XGBoost models were developed and compared. Five-fold cross-validation and hyperparameter tuning were applied to optimize model performance. The final model was interpreted using SHAP values to determine global and patient-level feature importance.</p> <p><strong>Findings: </strong>Patients with low quality of life had significantly longer disease duration, lower estimated glomerular filtration rate, higher dialysis frequency, longer dialysis duration, more comorbidities, greater medication burden, lower hemoglobin, lower serum albumin, higher serum phosphorus, and higher serum creatinine levels than patients without low quality of life (p &lt; 0.05). Among the evaluated models, XGBoost showed the strongest predictive performance, with accuracy of 0.847, sensitivity of 0.829, specificity of 0.860, precision of 0.806, F1-score of 0.817, and area under the receiver operating characteristic curve of 0.912. SHAP analysis identified burden of kidney disease score, serum albumin, hemoglobin, estimated glomerular filtration rate, number of comorbidities, dialysis status, disease duration, serum phosphorus, cardiovascular disease, and number of prescribed medications as the most influential predictors.</p> <p><strong>Conclusion: </strong>The XGBoost model demonstrated strong and interpretable performance in predicting low quality of life among patients with chronic kidney disease, and SHAP analysis showed that patient-reported burden, renal function, anemia, nutritional status, comorbidity load, and treatment-related factors jointly contributed to risk prediction.</p> </td> </tr> </tbody> </table> https://www.journals.kmanpub.com/index.php/qlhs/article/download/5780/10944