Longitudinal Models for Kidney Function Decline

• Activity Number: 174 - Biomarkers and Endpoint Validation
• Type: Contributed
• Date/Time: Monday, July 30, 2018 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Jing Zhang* and Loki Natarajan and Kumar Sharma and Tina Costacou and Janet Snell-Bergeon and Rachel Miller and Trevor Orchard
• Companies: Moores UCSD Cancer Center and UCSD and Division of Nephrology,University of Texas Health San Antonio and University of Pittsburgh and University of Colorado Anschutz Medical Campus, School of Medicine and University of Pittsburgh and University of Pittsburgh
• Keywords: Type I diabetes; eGFR decline; mixed model

Abstract

Background: Few studies have investigated long-term progression of kidney disease in patients with Type I diabetes. Methods: Using data from 1159 Type I diabetic patients with 10+years follow up, we used mixed models to compare linear and relative eGFR (log-eGFR) trajectory. Random intercept and slopes were included. We also tested if eGFR slopes varied by eGFR level at study entry, by including interaction terms of time and baseline eGFR level. Results: The sample was 48% male, with mean (SD) age 32 (9.8) years; 21 (8.5)years of diabetes, and 103.4 (29) ml/min/1.73m2 baseline eGFR. In the mixed models, eGFR slopes varied by initial eGFR level (likelihood ratio p < 0.001), and models adjusted for age and gender had better fit (p < 0.001). Correlations between observed and fitted values were 0.88-0.89 for the linear and loglinear models. Residual vs fitted and qq-plots indicated slightly better fit for the linear vs loglinear models. Conclusions: Linear and relative decline models gave similar predictions, but linear model had better fit. Inclusion of time*baseline eGFR interaction significantly improved fit, implying that rate of decline varies by starting eGFR-level.