Machine Learning Models for the Prediction of Renal Failure in Chronic Kidney Disease: A Retrospective Cohort Study

Abstract

This study assessed the feasibility of five separate machine learning (ML) classifiers for predicting disease progression in patients with pre-dialysis chronic kidney disease (CKD). The study enrolled 858 patients with CKD treated at a veteran’s hospital in Taiwan. After classification into early and advanced stages, patient demographics and laboratory data were processed and used to predict progression to renal failure and important features for optimal prediction were identified. The random forest (RF) classifier with synthetic minority over-sampling technique (SMOTE) had the best predictive performances among patients with early-stage CKD who progressed within 3 and 5 years and among patients with advanced-stage CKD who progressed within 1 and 3 years. Important features identified for predicting progression from early- and advanced-stage CKD were urine creatinine and serum creatinine levels, respectively. The RF classifier demonstrated the optimal performance, with an area under the receiver operating characteristic curve values of 0.96 for predicting progression within 5 years in patients with early-stage CKD and 0.97 for predicting progression within 1 year in patients with advanced-stage CKD. The proposed method resulted in the optimal prediction of CKD progression, especially within 1 year of advanced-stage CKD. These results will be useful for predicting prognosis among patients with CKD.

Improved Survival Analyses Based on Characterized Time-Dependent Covariates to Predict Individual Chronic Kidney Disease Progression

“…Recently, Mondol achieved high accuracy in early CKD prediction using convolutional neural network, ANN, and long short-term memory models [ 25 ]. In our recent study, high performance was obtained for predicting CKD progression using random forest methods, with C-indexes of 0.96 within 5 years in the early stage and 0.97 within 1 year in the advanced stage [ 14 ]. Although the models proposed by previous studies show high performance, the quality and accuracy of the estimates may vary over time [ 24 , 25 ].…”

“…In our recent study, using the Shapley additive explanation value method, the urine creatinine and eGFR were the most and second-most important predictive features in patients diagnosed with advanced-stage CKD within 3 and 5 years. In addition, serum creatinine was the most important predictive feature in patients diagnosed with advanced-stage CKD within 1–3 years [ 14 ]. Ultimately, the key predictive features may help determine the optimal predictive models for the progression from CKD to ESRD.…”

“…Machine learning (ML) methods have been used to predict the risk of CKD applications in recent years [ 15 , 16 , 17 , 18 ]. In addition, several risk prediction models have been proposed for CKD applications [ 14 , 19 , 20 , 21 , 22 ]. The published models applied to renal diseases include the Cox proportional hazard model (Cox PHM) [ 19 , 23 ], random survival forest (RSF) [ 24 ], and artificial neural network (ANN) models [ 22 , 25 ].…”

“…In our recent study, using baseline data, five available classification models (i.e., Gaussian naïve Bayes, linear regression, random forest (RF), support vector machine, and extreme gradient boosting) were developed for predicting the risk of progression among patients with CKD. The results showed that the RF model demonstrated the highest performance compared to the other models [ 14 ]. RF can be adapted to handle complex survival data, including nonlinear effects and complex interactions between features, which may be inappropriate for conventional statistical models.…”

Predicting hypoproteinemia among patients undergoing maintenance hemodialysis: A development and validation study based on machine learning algorithms

“…A study enrolled 858 patients with CKD and predicted progression to renal failure and important features, which revealed that a random forest (RF) classi er with a synthetic minority oversampling technique (SMOTE) had the best predictive performance among patients with early-stage CKD who progressed within 3 and 5 years and among patients with advanced-stage CKD who progressed within 1 and 3 years [23]. The RF algorithm also has many applications in other disease prediction models.…”