{"title":"Application of improved glomerular filtration rate estimation by a neural network model in patients with neurogenic lower urinary tract dysfunction.","authors":"Ying-Chun Ma, Yang Xie, Limin Liao","doi":"10.5414/CN111150","DOIUrl":null,"url":null,"abstract":"BACKGROUND\nPrevious studies have indicated that creatinine (Cr)-based glomerular filtration rate (GFR) estimating equations - including the new Chronic Kidney Disease Epidemiology creatinine (CKD-EPIcr) equation without race and the estimated glomerular filtration rate (eGFR) equation developed for the Chinese population - displayed suboptimal performance in patients with neurogenic lower urinary tract dysfunction (NLUTD), which limited their clinical application for detecting changes in GFR levels in all cohorts.\n\n\nOBJECTIVE\nTo develop a neural network model based on multilayer perceptron (MLP) for evaluating GFR in Chinese NLUTD patients, and compare the diagnostic performance with Cr-based multiple linear regression equations for Chinese and the CKD-EPIcr equation without race.\n\n\nDESIGN\nSingle-center, cross-sectional study of GFR estimation from serum Cr, demographic data, and clinical characteristics in Chinese patients with NLUTD.\n\n\nPATIENTS\nA total of 204 NLUTD patients, from 27 different geographic regions of China, were selected. A random sample of 141 of these subjects was included in the training sample set, and the remaining 63 patients were included in the testing sample set.\n\n\nMETHODS\nThe reference GFR (rGFR) was assessed by the technetium-99m-labeled diethylenetriaminepentaacetic acid (99mTc-DTPA) double plasma sample method. A neural network model based on MLP was developed to evaluate GFR in the training sample set, which was then validated in the testing sample set and compared with Cr-based GFR equations.\n\n\nRESULTS\nThe MLP-based model showed significant performance improvement in evaluating the difference, absolute difference, precision, and accuracy of GFR estimation compared with the Cr-based GFR equations. Additionally, compared with the rGFR, we found that the MLP-based model provided an acceptable level of accuracy (greater than 85%, which was within a 30% deviation from the rGFR).\n\n\nCONCLUSION\nThe MLP-based model offered significant advantages in estimating GFR in Chinese NLUTD patients, and its application could be suggested in clinical practice.","PeriodicalId":10396,"journal":{"name":"Clinical nephrology","volume":" 23","pages":""},"PeriodicalIF":1.1000,"publicationDate":"2024-04-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Clinical nephrology","FirstCategoryId":"3","ListUrlMain":"https://doi.org/10.5414/CN111150","RegionNum":4,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q3","JCRName":"UROLOGY & NEPHROLOGY","Score":null,"Total":0}
引用次数: 0
Abstract
BACKGROUND
Previous studies have indicated that creatinine (Cr)-based glomerular filtration rate (GFR) estimating equations - including the new Chronic Kidney Disease Epidemiology creatinine (CKD-EPIcr) equation without race and the estimated glomerular filtration rate (eGFR) equation developed for the Chinese population - displayed suboptimal performance in patients with neurogenic lower urinary tract dysfunction (NLUTD), which limited their clinical application for detecting changes in GFR levels in all cohorts.
OBJECTIVE
To develop a neural network model based on multilayer perceptron (MLP) for evaluating GFR in Chinese NLUTD patients, and compare the diagnostic performance with Cr-based multiple linear regression equations for Chinese and the CKD-EPIcr equation without race.
DESIGN
Single-center, cross-sectional study of GFR estimation from serum Cr, demographic data, and clinical characteristics in Chinese patients with NLUTD.
PATIENTS
A total of 204 NLUTD patients, from 27 different geographic regions of China, were selected. A random sample of 141 of these subjects was included in the training sample set, and the remaining 63 patients were included in the testing sample set.
METHODS
The reference GFR (rGFR) was assessed by the technetium-99m-labeled diethylenetriaminepentaacetic acid (99mTc-DTPA) double plasma sample method. A neural network model based on MLP was developed to evaluate GFR in the training sample set, which was then validated in the testing sample set and compared with Cr-based GFR equations.
RESULTS
The MLP-based model showed significant performance improvement in evaluating the difference, absolute difference, precision, and accuracy of GFR estimation compared with the Cr-based GFR equations. Additionally, compared with the rGFR, we found that the MLP-based model provided an acceptable level of accuracy (greater than 85%, which was within a 30% deviation from the rGFR).
CONCLUSION
The MLP-based model offered significant advantages in estimating GFR in Chinese NLUTD patients, and its application could be suggested in clinical practice.
期刊介绍:
Clinical Nephrology appears monthly and publishes manuscripts containing original material with emphasis on the following topics: prophylaxis, pathophysiology, immunology, diagnosis, therapy, experimental approaches and dialysis and transplantation.