Predictors of Change in Glomerular Filtration Rate Among Nondiabetic Patients with Chronic Kidney Disease

• Activity Number: 465
• Type: Topic Contributed
• Date/Time: Thursday, August 7, 2003 : 10:30 AM to 12:20 PM
• Sponsor: Section on Bayesian Stat. Sciences
• Author(s): Paul C. Stark*+ and Christopher H. Schmid and Andrew Levey
• Companies: Tufts - New England Medical Center and New England Medical Center and Lifespan
• Keywords: multilevel model ; Bayesian inference ; growth curves ; random effects ; kidney disease ; GFR

Abstract

Using data on 1860 patients with nondiabetic kidney disease from 11 randomized controlled trials evaluating the efficacy of angiotensin-converting enzyme (ACE) inhibitors, we showed that ACE inhibitors slow disease progression and this effect is greater with high levels of urinary protein excretion. No other effect modifiers were discovered from survival analysis, perhaps due to a lack of power. Because glomerular filtration rate (GFR) is continuous and serves as a marker for disease progression, its use as an outcome may increase power and uncover additional interactions. Often GFR effects are measured by their change as expressed by a slope calculated by regression of an individual's GFR on time. Using a least squares estimate of slope in a regression weighted by its variance does not incorporate the variability due to the uncertainty of the slope estimate. A better technique uses a multilevel model with the individual slopes treated as random effects. This gives better estimates of the parameters modifying the slope as well as of the slopes themselves. In this analysis, we found additional treatment interactions with age, systolic blood pressure, and serum creatinine.