Neural network algorithm to determine the optimal range of PTH level for patients with end-stage chronic kidney disease to maintain bone metabolism

All over the world, chronic kidney disease (CKD) is a serious public health condition that needs to be detected in advance so that costly end-stage treatments like dialysis, kidney transplantations can be avoided. Neural network model and 10-fold cross-validation methodology under a single platform in proposed as well as implemented in order to classify patients with CKD. This will assist medical care fields so that counter measures can be suggested. The performance of proposed classifier is justified against other baseline classifiers such as Support Vector Machine, K-Nearest Neighbours, Decision tree and Gradient Boost classifier. Experimental results conclude that the performance of neural network with 10-fold cross-validation method reaches promising accuracy of 98.25%, f1-score of 0.98, and kappa score of 0.96 and MSE of 0.0175.

Download Full-text

Determination of the parathyroid hormone reference interval in patients with different stages of the chronic kidney disease

Proceedings of the National Academy of Sciences of Belarus Medical series ◽

10.29235/1814-6023-2021-18-2-186-195 ◽

2021 ◽

Vol 18 (2) ◽

pp. 186-195

Author(s):

N. V. Karlovich ◽

O. S. Spiridonova ◽

E. G. Sazonova ◽

T. V. Mokhort

Keyword(s):

Chronic Kidney Disease ◽

Kidney Disease ◽

General Population ◽

Bone Metabolism ◽

Reference Interval ◽

Control Group ◽

Mineral Density ◽

Metastatic Calcification ◽

Extreme Risk ◽

Pth Level

Secondary hyperparathyroidism (SHPT) is one of the most clinically significant complications of chronic kidney disease (CKD) due to associated mineral, bone disorders, and metastatic calcification. The indicators of mineral and bone metabolism of 635 patients with different CKD stages and 50 persons of the control group were analyzed using a neural network algorithm and the mathematical technology BootStrаp, which allowed determining the target PTH intervals for each stage in patients with CKD, corresponding to the optimal indicators of mineral density and metabolism bone tissue, in order to improve the survival of this category of patients. It was found that the upper limit of the reference interval of the PTH level in patients with CKD and GFR > 35 ml/min coincides with the general population, in patients with CKD and GFR 15‒35 ml/min it is 185 pg/ml, which is 3 times higher than in the general population, and in patients with CKD and GFR < 15 ml/min it is 500 pg/ml (7.5 times higher than in the general population). In dialysis patients with the PTH level of 500‒1500 pg/ml, it is possible to maintain satisfactory parameters of bone metabolism, and the PTH level of >1500 pg/ml determines the extreme risk of developing severe SHPT complications.

Download Full-text

Chronic Kidney Disease Prediction Using Neural Approach

10.1101/2020.06.28.20142034 ◽

2020 ◽

Author(s):

Shawni Dutta ◽

Samir Kumar Bandyopadhyay

Keyword(s):

Neural Network ◽

Chronic Kidney Disease ◽

Kidney Disease ◽

Cross Validation ◽

Health Condition ◽

Support Vector ◽

Kappa Score ◽

Kidney Transplantations ◽

End Stage ◽

Fold Cross Validation

AbstractAll over the world, chronic kidney disease (CKD) is a serious public health condition that needs to be detected in advance so that costly end-stage treatments like dialysis, kidney transplantations can be avoided. Neural network model and 10-fold cross-validation methodology under a single platform in proposed as well as implemented in order to classify patients with CKD. This will assist medical care fields so that counter measures can be suggested. The performance of proposed classifier is justified against other baseline classifiers such as Support Vector Machine, K-Nearest Neighbours, Decision tree and Gradient Boost classifier. Experimental results conclude that the performance of neural network with 10-fold cross-validation method reaches promising accuracy of 98.25%, f1-score of 0.98, and kappa score of 0.96 and MSE of 0.0175.

Download Full-text