Scholarlink

<p>Control of metabolic syndrome (MS) is still a major challenge for patients suffering from Non-communicable diseases (NCDs) in Bangladesh. In this study, the status of MS control was determined, and Bayesian-optimized machine-learning models were developed and tested to classify MS control status concurrently with NCDs in patients attending government hospitals. The study was a cross-sectional hospital-based study in 14 government hospitals of Bangladesh from June to July 2025. A total of 472 adult NCD patients aged 40 years and above were included. The clinical, behavioral, demographic, and health-system-related variables were examined. Boruta feature selection was used, and six classifiers (Random Forest, Multilayer Perceptron, CatBoost, XGBoost, LightGBM, and AdaBoost) were evaluated by means of 10-fold stratified cross-validation, bootstrap 95% confidence intervals, calibration analysis, seed-variance analysis, McNemar’s test, and interpretability based on SHAP values. Overall, 51.1% of participants achieved control of MS. Better control was related to secondary level hospitals, physical activity, medicine adherence, and community participation, follow-up care, and health behavior monitoring, while poorer control was related to polypharmacy, higher burden of comorbidities, and abnormal BMI. Among six Bayesian-optimized classifiers, Random Forest demonstrated the best overall performance (CV AUC: 0.789 ± 0.062; accuracy: 72.9%; F1-score: 0.735; bootstrap 95% CI: 0.630–0.832) and exhibited the fewest false negatives with optimal calibration. Comorbidity, polypharmacy, physical activity, and BMI were the most important factors associated with MS control as identified by SHAP analysis. Bayesian-optimized Random Forest is a viable and meaningful solution for an interpretable, concurrent classification of the control status of MS in the context of limited resources within a hospital environment. The results back the integrated management of comorbidities, medication review, promotion of physical activity, weight management, and structured NCD follow-up.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Bayesian-optimized machine learning models for classifying metabolic syndrome control among NCD patients in Bangladesh hospitals

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月25日

查看全文
  • Md Soumike Hassan,
  • Jakia Sultana Pingky,
  • Amartay Kumar Dhar,
  • Muhammad Al Amin Dewan,
  • Nahid Salma,
  • Nasrin Khatun

摘要

Control of metabolic syndrome (MS) is still a major challenge for patients suffering from Non-communicable diseases (NCDs) in Bangladesh. In this study, the status of MS control was determined, and Bayesian-optimized machine-learning models were developed and tested to classify MS control status concurrently with NCDs in patients attending government hospitals. The study was a cross-sectional hospital-based study in 14 government hospitals of Bangladesh from June to July 2025. A total of 472 adult NCD patients aged 40 years and above were included. The clinical, behavioral, demographic, and health-system-related variables were examined. Boruta feature selection was used, and six classifiers (Random Forest, Multilayer Perceptron, CatBoost, XGBoost, LightGBM, and AdaBoost) were evaluated by means of 10-fold stratified cross-validation, bootstrap 95% confidence intervals, calibration analysis, seed-variance analysis, McNemar’s test, and interpretability based on SHAP values. Overall, 51.1% of participants achieved control of MS. Better control was related to secondary level hospitals, physical activity, medicine adherence, and community participation, follow-up care, and health behavior monitoring, while poorer control was related to polypharmacy, higher burden of comorbidities, and abnormal BMI. Among six Bayesian-optimized classifiers, Random Forest demonstrated the best overall performance (CV AUC: 0.789 ± 0.062; accuracy: 72.9%; F1-score: 0.735; bootstrap 95% CI: 0.630–0.832) and exhibited the fewest false negatives with optimal calibration. Comorbidity, polypharmacy, physical activity, and BMI were the most important factors associated with MS control as identified by SHAP analysis. Bayesian-optimized Random Forest is a viable and meaningful solution for an interpretable, concurrent classification of the control status of MS in the context of limited resources within a hospital environment. The results back the integrated management of comorbidities, medication review, promotion of physical activity, weight management, and structured NCD follow-up.