Scholarlink

Objective <p>The development of an Electronic Medical Record (EMR) for Diabetic Kidney Disease (DKD) to digitalise the storage process of patients’ records is needed since it is an integral part of the health care system. Deploying machine learning approaches, the occurrence of DKD in type 2 diabetes patients is predicted, and identified the risk parameters that will assist in prediction procedure.</p> Methods <p>The medical records of 297 patients are collected from the hospital over seven months. The 297 patients’ raw data is pre-processed. Seven different machine learning approaches are applied on the data set to train different models for prediction. The models are evaluated using several evaluation metrics, and compared their performances to find an appropriate model for DKD prediction. The degree of significance of the risk factors that influence DKD is determined through the SHapley Additive exPlanations (SHAP) technique. Calibration curves and decision curves are plotted to define the clinical utility of the model.</p> Results <p>Among all models, the Random Forest has given better performance with 81.11% accuracy and 0.86 Area Under Curve of the Receiver Operating Characteristic curve (ROC-AUC). Applying the SHAP technique, high urine protein, Urine Albumin Creatinine Ratio (UACR), Post Prandial Blood Sugar (PPBS), sodium, potassium, and uric acid are identified as the attributes influencing DKD. This model showed a good calibration and positive net benefit across clinically meaningful thresholds.</p> Conclusion <p>We have concluded that machine learning models like Random Forest can predict DKD with different levels of accuracy, and using SHAP techniques, the risk factors for DKD can be identified.</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

Development of an electronic medical record (EMR) registry and machine learning (ML) model for prediction of diabetic kidney disease (DKD) using clinical data

  • 期刊论文

  • 发表日期：

    2025年9月29日

查看全文
  • Subhashree Palaur,
  • Nikunj Kishore Rout,
  • Rekha Sahu,
  • Satya Ranjan Dash,
  • Rajani Kanta Mahapatra

摘要

Objective

The development of an Electronic Medical Record (EMR) for Diabetic Kidney Disease (DKD) to digitalise the storage process of patients’ records is needed since it is an integral part of the health care system. Deploying machine learning approaches, the occurrence of DKD in type 2 diabetes patients is predicted, and identified the risk parameters that will assist in prediction procedure.

Methods

The medical records of 297 patients are collected from the hospital over seven months. The 297 patients’ raw data is pre-processed. Seven different machine learning approaches are applied on the data set to train different models for prediction. The models are evaluated using several evaluation metrics, and compared their performances to find an appropriate model for DKD prediction. The degree of significance of the risk factors that influence DKD is determined through the SHapley Additive exPlanations (SHAP) technique. Calibration curves and decision curves are plotted to define the clinical utility of the model.

Results

Among all models, the Random Forest has given better performance with 81.11% accuracy and 0.86 Area Under Curve of the Receiver Operating Characteristic curve (ROC-AUC). Applying the SHAP technique, high urine protein, Urine Albumin Creatinine Ratio (UACR), Post Prandial Blood Sugar (PPBS), sodium, potassium, and uric acid are identified as the attributes influencing DKD. This model showed a good calibration and positive net benefit across clinically meaningful thresholds.

Conclusion

We have concluded that machine learning models like Random Forest can predict DKD with different levels of accuracy, and using SHAP techniques, the risk factors for DKD can be identified.