San Francisco Chapter of the American Statistical Association August Seminar

Speaker: Professor Maria Carmen Pardo, Department of Statistics and Operations Research (I), Complutense University of Madrid (Spain),

mcapardo@mat.ucm.es, www.mat.ucm.es/~mcapardo

Title: Time‐dependent ROC Curves for Assessing a Biomarker in the Prediction of Acute Kidney Injury

Time: August 3^rd , 2015 4:30-6:00pm (4:30-5:30pm presentation; 5:30-6:00pm networking/refreshment)

Location: 7000 Shoreline Court, 1st floor dining hall South San Francisco, CA 94080

Abstract:

Serum creatinine is the most commonly used biomarker in detecting acute kidney injury (AKI). However, after major cardiac surgery, the serum creatinine increase is typically delayed by 1-3 days due to several non-kidney factors. The aim of this study is to determine the utility of an alternative routine postoperative biomarker to predict an AKI event 1 or 2 days in advance, in patients undergoing major cardiac surgery (CS).

In a random sample of patients operated with major CS, we explored the predictive ability of routine renal function biomarkers (urea, estimated glomerular filtration rate (GFR) by the Modification of Diet in Renal Disease four-variable formula (MDRD-4), and their respective changes). Time dependent receiver operating characteristic (ROC) curves were used for this investigation, considering three stages of the AKI events (severe, moderate and mild) and the timing after surgery (on the day of surgery, one day, and two days after surgery). Results (change from baseline) obtained from MDRD4 were compared with those obtained from serum creatinine. MDRD4 appears to be a good alternative to serum creatinine in predicting an AKI event 1 or 2 days in advance, particularly for severe and mild events.