Dataset Information

Improving precision of glomerular filtration rate estimating model by ensemble learning.

ABSTRACT: Accurate assessment of kidney function is clinically important, but estimates of glomerular filtration rate (GFR) by regression are imprecise.We hypothesized that ensemble learning could improve precision. A total of 1419 participants were enrolled, with 1002 in the development dataset and 417 in the external validation dataset. GFR was independently estimated from age, sex and serum creatinine using an artificial neural network (ANN), support vector machine (SVM), regression, and ensemble learning. GFR was measured by 99mTc-DTPA renal dynamic imaging calibrated with dual plasma sample 99mTc-DTPA GFR.Mean measured GFRs were 70.0 ml/min/1.73 m2 in the developmental and 53.4 ml/min/1.73 m2 in the external validation cohorts. In the external validation cohort, precision was better in the ensemble model of the ANN, SVM and regression equation (IQR = 13.5 ml/min/1.73 m2) than in the new regression model (IQR = 14.0 ml/min/1.73 m2, P < 0.001). The precision of ensemble learning was the best of the three models, but the models had similar bias and accuracy. The median difference ranged from 2.3 to 3.7 ml/min/1.73 m2, 30% accuracy ranged from 73.1 to 76.0%, and P was > 0.05 for all comparisons of the new regression equation and the other new models.An ensemble learning model including three variables, the average ANN, SVM, and regression equation values, was more precise than the new regression model. A more complex ensemble learning strategy may further improve GFR estimates.

SUBMITTER: Liu X

PROVIDER: S-EPMC5679185 | biostudies-literature | 2017 Nov

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Improving precision of glomerular filtration rate estimating model by ensemble learning.

Liu Xun X Li Ningshan N Lv Linsheng L Fu Yongmei Y Cheng Cailian C Wang Caixia C Ye Yuqiu Y Li Shaomin S Lou Tanqi T

Journal of translational medicine 20171109 1

<h4>Background</h4>Accurate assessment of kidney function is clinically important, but estimates of glomerular filtration rate (GFR) by regression are imprecise.<h4>Methods</h4>We hypothesized that ensemble learning could improve precision. A total of 1419 participants were enrolled, with 1002 in the development dataset and 417 in the external validation dataset. GFR was independently estimated from age, sex and serum creatinine using an artificial neural network (ANN), support vector machine (S ...[more]

PMID: 29121946

Similar Datasets

Project description:BackgroundThe performance of previously published glomerular filtration rate (GFR) estimation equations degrades when directly used in Chinese population. We incorporated more independent variables and using complicated non-linear modeling technology (artificial neural network, ANN) to develop a more accurate GFR estimation model for Chinese population.MethodsThe enrolled participants came from the Third Affiliated Hospital of Sun Yat-sen University, China from Jan 2012 to Jun 2016. Participants with age < 18, unstable kidney function, taking trimethoprim or cimetidine, or receiving dialysis were excluded. Among the finally enrolled 1952 participants, 1075 participants (55.07%) from Jan 2012 to Dec 2014 were assigned as the development data whereas 877 participants (44.93%) from Jan 2015 to Jun 2016 as the internal validation data. We in total developed 3 GFR estimation models: a 4-variable revised CKD-EPI (chronic kidney disease epidemiology collaboration) equation (standardized serum creatinine and cystatin C, age and gender), a 9-variable revised CKD-EPI equation (additional auxiliary variables: body mass index, blood urea nitrogen, albumin, uric acid and hemoglobin), and a 9-variable ANN model.ResultsCompared with the 4-variable equation, the 9-variable equation could not achieve superior performance in the internal validation data (mean of difference: 5.00 [3.82, 6.54] vs 4.67 [3.55, 5.90], P = 0.5; interquartile range (IQR) of difference: 18.91 [17.43, 20.48] vs 20.11 [18.46, 21.80], P = 0.05; P30: 76.6% [73.7%, 79.5%] vs 75.8% [72.9%, 78.6%], P = 0.4), but the 9-variable ANN model significantly improve bias and P30 accuracy (mean of difference: 2.77 [1.82, 4.10], P = 0.007; IQR: 19.33 [17.77, 21.17], P = 0.3; P30: 80.0% [77.4%, 82.7%], P < 0.001).ConclusionsIt is suggested that using complicated non-linear models like ANN could fully utilize the predictive ability of the independent variables, and then finally achieve a superior GFR estimation model.

Project description:Estimates of glomerular filtration rate (GFR) that are based on serum creatinine are routinely used; however, they are imprecise, potentially leading to the overdiagnosis of chronic kidney disease. Cystatin C is an alternative filtration marker for estimating GFR.Using cross-sectional analyses, we developed estimating equations based on cystatin C alone and in combination with creatinine in diverse populations totaling 5352 participants from 13 studies. These equations were then validated in 1119 participants from 5 different studies in which GFR had been measured. Cystatin and creatinine assays were traceable to primary reference materials.Mean measured GFRs were 68 and 70 ml per minute per 1.73 m(2) of body-surface area in the development and validation data sets, respectively. In the validation data set, the creatinine-cystatin C equation performed better than equations that used creatinine or cystatin C alone. Bias was similar among the three equations, with a median difference between measured and estimated GFR of 3.9 ml per minute per 1.73 m(2) with the combined equation, as compared with 3.7 and 3.4 ml per minute per 1.73 m(2) with the creatinine equation and the cystatin C equation (P=0.07 and P=0.05), respectively. Precision was improved with the combined equation (interquartile range of the difference, 13.4 vs. 15.4 and 16.4 ml per minute per 1.73 m(2), respectively [P=0.001 and P<0.001]), and the results were more accurate (percentage of estimates that were >30% of measured GFR, 8.5 vs. 12.8 and 14.1, respectively [P<0.001 for both comparisons]). In participants whose estimated GFR based on creatinine was 45 to 74 ml per minute per 1.73 m(2), the combined equation improved the classification of measured GFR as either less than 60 ml per minute per 1.73 m(2) or greater than or equal to 60 ml per minute per 1.73 m(2) (net reclassification index, 19.4% [P<0.001]) and correctly reclassified 16.9% of those with an estimated GFR of 45 to 59 ml per minute per 1.73 m(2) as having a GFR of 60 ml or higher per minute per 1.73 m(2).The combined creatinine-cystatin C equation performed better than equations based on either of these markers alone and may be useful as a confirmatory test for chronic kidney disease. (Funded by the National Institute of Diabetes and Digestive and Kidney Diseases.).

Project description:Rationale & objectiveEvaluation of glomerular filtration rate (GFR) is challenging in adults undergoing bariatric surgery because creatinine and cystatin C levels are influenced by changes in muscle and fat mass. Additionally, indexing of GFR by body surface area (BSA) may by affected by decreases in BSA.Study designProspective observational study.Setting & participants27 adults with body mass index (BMI) ≥ 35 kg/m2 who underwent measurement of GFR before and after bariatric surgery.OutcomesIndexed and nonindexed GFRs measured (mGFRs) using plasma iohexol clearance, indexed and nonindexed estimated GFR (eGFR) based on levels of creatinine, cystatin C, or both from Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equations.Analytic approachBias and percent of estimates within 20% and 30% of mGFR (P20 and P30) for estimating equations were examined.ResultsMean presurgery BMI was 49.5 (SD, 9.4) kg/m2, BSA was 2.42 (SD, 0.27) m2, nonindexed mGFR was 117.3 (SD, 34.1) mL/min, and indexed mGFR was 84.1 (SD, 22.0) mL/min/1.73 m2. After 6 months, mean BMI changed by -13.8 (95% CI, -15.9 to -11.8) kg/m2, BSA by -0.30 (95% CI, -0.33 to -0.27) m2, and nonindexed mGFR by -9.2 (95% CI, -17.2 to -1.1) mL/min, while indexed mGFR was unchanged at 5.1 (95% CI, -0.1 to 10.4) mL/min/1.73 m2. Nonindexed eGFRcr was unbiased (median bias, 5.0 [95% CI, -4.3 to 11.6] mL/min) before surgery, but overestimated mGFR (8.8 [95% CI, 1.8 to 16.9] mL/min) after surgery. Nonindexed eGFRcys underestimated mGFR before (median bias, -12.1 [95% CI, -21.4 to -1.2] mL/min) and after surgery (-11.2 [95% CI, -21.8 to -7.3] mL/min). Nonindexed eGFRcr-cys was unbiased before (median bias, -6.0 [95% CI, -11.0 to 1.0] mL/min) and after surgery (-2.0 [95% CI, -8.8 to 4.9] mL/min). Findings were similar for indexed eGFR compared with indexed mGFR.LimitationsSmall, mostly white sample.ConclusionsChanges in indexed and nonindexed GFRs may be discordant after bariatric surgery in adults because of decreases in BSA. Indexed and nonindexed eGFRcr-cys may be less biased than indexed or nonindexed eGFRcr or eGFRcys because of opposite biases in estimating mGFR.

Project description:BackgroundWe aimed to investigate the accuracy of different equations in evaluating estimated glomerular filtration rate (eGFR) in a Chinese population with different BMI levels.MethodsA total of 837 Chinese patients were enrolled, and the eGFRs were calculated by three Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equations, three full-age spectrum (FAS) equations and two Modification of Diet in Renal Disease (MDRD) equations. Results of measured GFR (mGFR) by the 99Tcm-diathylenetriamine pentaacetic acid (99Tcm-DTPA) renal dynamic imaging method were the reference standards. According to BMI distribution, the patients were divided into three intervals: below 25th(BMIP25), 25th to 75th(BMIP25-75) and over 75th percentiles (BMIP75).ResultsThe medium BMI of the three BMI intervals were 20.9, 24.8 and 28.9 kg/m2, respectively. All deviations from mGFR (eGFR) were correlated with BMI (p < 0.05). The percentage of cases in which eGFR was within mGFR ±30% (P30) was used to represent the accuracy of each equation. Overall, eGFRFAS_Cr_CysC and eGFREPI_Cr_2009 performed similarly, showing the best agreement with mGFR among the eight equations in Bland-Altman analysis (biases: 4.1 and - 4.2 mL/min/1.73m2, respectively). In BMIP25 interval, eGFRFAS_Cr got - 0.7 of the biases with 74.2% of P30, the kappa value was 0.422 in classification of CKD stages and the AUC60 was 0.928 in predicting renal insufficiency, and eGFREPI_Cr_2009 got 2.3 of the biases with 71.8% of P30, the kappa value was 0.418 in classification of CKD stages and the AUC60 was 0.920 in predicting renal insufficiency. In BMIP25-75 interval, the bias of eGFRFAS_Cr_CysC was 4.0 with 85.0% of P30, the kappa value was 0.501 and the AUC60 was 0.941, and eGFRFAS_Cr_CysC showed balanced recognition ability of each stage of CKD (62.3, 63.7, 68.0, 71.4 and 83.3% respectively). In BMIP75 interval, the bias of eGFREPI_Cr_CysC_2012 was 3.8 with 78.9% of P30, the kappa value was 0.484 the AUC60 was 0.919, and eGFREPI_Cr_CysC_2012 equation showed balanced and accurate recognition ability of each stage (60.5, 60.0, 71.4, 57.1 and 100% respectively). In BMIP75 interval, the bias of eGFRFAS_Cr_CysC was - 1.8 with 78.5% of P30, the kappa value was 0.485, the AUC60 was 0.922. However, the recognition ability of each stage of eGFRFAS_Cr_CysC eq. (71.1, 61.2, 70.0, 42.9 and 50.0% respectively) was not as good as GFREPI_Cr_CysC_2012 equation.ConclusionFor a Chinese population, we tend to recommend choosing eGFRFAS_Cr and eGFREPI_Cr_2009 when BMI was around 20.9, eGFRFAS_Cr_CysC when BMI was near 24.8, and eGFREPI_Cr_CysC_2012 when BMI was about 28.9.

Dataset Information

Improving precision of glomerular filtration rate estimating model by ensemble learning.

Publications

Improving precision of glomerular filtration rate estimating model by ensemble learning.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets