Adaptive Bayesian approach to clinical trial renal impairment biomarker signal from urea and creatinine.

A major concern with the identification of renal toxicity using the traditional biomarkers, urea and creatinine, is that toxicity signal definitions are not sensitive to medically important changes in these biomarkers. Traditional renal signal definitions for urea and creatinine have not adequately identified drugs that have generated important medical issues later in development. Here, two clinical trial databases with a posteriori known drug induced renal impairment were analyzed for the presence of a renal impairment biomarker signal from urea (590 patients; age 26-92, median 65) and creatinine (532 patients; age 26-97, median 65). Data was analyzed retrospectively using multiple definitions for the biomarker signal to include values outside stratified reference intervals, values exceeding twofold increases from baseline, values classified by the 2009 NIAID renal toxicity table, change from baseline represented as a Z-score based on intra-individual biological variations, and an adaptive Bayesian methodology that generalizes population- with individual-based methods for evaluating a biomarker signal. The data demonstrated that the adaptive Bayesian methodology generated a prominent drug induced signal for renal impairment at the first visit after drug administration. The signal was directly related to dose and time of drug administration. All other data analysis methods produced none or significantly weaker signals than the adaptive Bayesian approach. Interestingly, serum creatinine and urea are able to detect early kidney dysfunction when the biomarker signal is personalized.