Abstract
AbstractBackgroundMachine learning may offer a superior alternative to traditional prediction tools when used to model complicated, nonlinear interactions between variables. While modern machine learning methods are tagged as “black boxes”, the random forest (RF) classifier can be interrogated to understand the contribution of input variables (feature importance), thereby improving the interpretability of its predictions. We hypothesized that a random forest (RF) classifier would have equivalent, if not superior, performance to the 4-variable Kidney Failure Risk Equation (KFRE) in predicting progression to end stage kidney disease (ESKD) in a chronic kidney disease (CKD) population and explored the impact of serum creatinine and primary renal disease on prediction accuracy.MethodsA 2-year risk of ESKD was calculated using the 4-variable KFRE and compared to a RF model using the same four variables (age, gender, eGFR and urine albumin creatinine ratio). Four more RF models were developed using a combination of these as well as serum creatinine and primary renal disease. Performance of the KFRE and RF models was assessed by area under a receiver operating (AUC ROC) curve and feature importance was evaluated for each RF model.ResultsOf 1365 patients with CKD from two renal units included in the analysis, 208 progressed to ESKD in the 2-year follow-up period. The AUC ROC for KFRE was 0.95 (95% confidence interval, 0.93 – 0.96) and for the RF model using the same 4 variables 0.97. The remaining four RF models had similar performance (AUC ROC 0.97 – 0.98). In the RF models, eGFR and serum creatinine had the largest effect on risk prediction while gender had the smallest.ConclusionsOur findings suggest that RF models provide a potential tool to predict CKD progression with competing accuracy and interpretability to the current benchmark equation. They therefore warrant validation in larger and more diverse populations
Publisher
Cold Spring Harbor Laboratory