Biomarker Panels for Predicting Progression of Kidney Disease in Acute Kidney Injury Survivors.

Abstract

Acute kidney injury (AKI) increases the risk for chronic kidney disease (CKD). We aimed to identify combinations of clinical variables and biomarkers that predict long-term kidney disease risk after AKI. We analyzed data from a prospective cohort of 723 hospitalized patients with AKI in the Assessment, Serial Evaluation, and Subsequent Sequelae of AKI (ASSESS-AKI) Study. Using machine learning, we investigated 75 candidate predictors including biomarkers measured at three-month post-discharge follow-up to predict major adverse kidney events (MAKE) within three years, defined as a decline in eGFR ≥40%, development of end-stage kidney disease (ESKD), or death. The mean age of study participants was 64 ± 13 years, 68% were men, and 79% were of White race. Two hundred and four (28%) patients developed MAKE over 3 years of follow-up. Random forest and LASSO penalized regression models using all 75 predictors yielded area under the receiver-operating characteristic curve (AUC) values of 0.80 (95% CI: 0.69-0.91) and 0.79 (95% CI: 0.68-0.90) respectively. The most consistently selected predictors were albuminuria, soluble tumor necrosis factor receptor 1 (sTNFR1), and diuretic use. A parsimonious model using the top eight predictor variables showed similarly strong discrimination for MAKE (AUC = 0.78; 95% CI: 0.66-0.90). Clinical impact utility analyses demonstrated that the eight-predictor model would have 55% higher efficiency of post-AKI care (number needed to screen/follow-up for a MAKE event decreased from 3.55 to 1.97). For a kidney-specific outcome of eGFR decline or ESKD, a four-predictor model showed strong discrimination (AUC = 0.82; 95% CI: 0.68-0.96). Combining clinical data and biomarkers can accurately identify high-risk AKI patients, enabling personalized post-AKI care and improved outcomes.