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

Area of Science:

• Nephrology
• Medical Informatics
• Biostatistics

Background:

• Kidney diseases present a significant global health burden, necessitating effective risk factor identification for prevention and treatment.
• Early detection and management of chronic kidney disease (CKD) progression are vital to reduce morbidity and mortality.

Purpose of the Study:

• To compare the predictive performance of three time-dependent survival models for renal failure in patients with stage 3-5 CKD.
• To identify key clinical and demographic features for optimal prediction of CKD progression.

Main Methods:

• A cohort of 497 stage 3-5 CKD patients in Taiwan was followed for 3 years with 3-month clinical measurements.
• Three survival models—Cox proportional hazard model (Cox PHM), random survival forest (RSF), and artificial neural network (ANN)—were employed.
• Model performance was evaluated using concordance indexes, sensitivity, and specificity, with validation via Kaplan-Meier estimation.

Main Results:

• The random survival forest (RSF) model demonstrated superior predictive performance with a concordance index of 0.89, compared to Cox PHM (0.71) and ANN (0.72).
• RSF achieved a sensitivity of 0.79 and specificity of 0.88 in predicting CKD progression within 3 years.
• Key predictors identified by RSF included creatinine, age, estimated glomerular filtration rate, and urine protein to creatinine ratio.

Conclusions:

• The random survival forest model offers robust and accurate prediction of CKD progression.
• Creatinine, age, eGFR, and UPCR are significant factors for predicting renal failure in CKD patients.
• These findings support the use of RSF for instantaneous risk assessment in routine CKD patient follow-ups.