Chronic kidney disease (CKD) is a common comorbidity in people with diabetes mellitus, and a key risk factor for further life-threatening conditions such as cardiovascular disease. The early prediction of progression of CKD therefore is an important clinical goal, but remains difficult due to the multifaceted nature of the condition. We validated a set of established protein biomarkers for the prediction of trajectories of estimated glomerular filtration rate (eGFR) in people with moderately advanced chronic kidney disease and diabetes mellitus. Our aim was to discern which biomarkers associate with baseline eGFR or are important for the prediction of the future eGFR trajectory.

We used Bayesian linear mixed models with weakly informative and shrinkage priors for clinical predictors (n = 12) and protein biomarkers (n = 19) to model eGFR trajectories in a retrospective cohort study of people with diabetes mellitus (n = 838) from the nationwide German Chronic Kidney Disease study. We used baseline eGFR to update the models’ predictions, thereby assessing the importance of the predictors and improving predictive accuracy computed using repeated cross-validation.

The model combining clinical and protein predictors had higher predictive performance than a clinical only model, with an of 0.44 (95% credible interval 0.37–0.50) before, and 0.59 (95% credible interval 0.51–0.65) after updating by baseline eGFR, respectively. Only few predictors were sufficient to obtain comparable performance to the main model, with markers such as Tumor Necrosis Factor Receptor 1 and Receptor for Advanced Glycation Endproducts being associated with baseline eGFR, while Kidney Injury Molecule 1 and urine albumin-creatinine-ratio were predictive for future eGFR decline.

Protein biomarkers only modestly improve predictive accuracy compared to clinical predictors alone. The different protein markers serve different roles for the prediction of longitudinal eGFR trajectories potentially reflecting their role in the disease pathway.