|
Activity Number:
|
328
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 8, 2006 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Statistics in Epidemiology
|
| Abstract - #307245 |
|
Title:
|
Application of Errors-in-Variables to Model Variation between Studies in Regression Equations for GFR
|
|
Author(s):
|
Tom Greene and Liang Li*+
|
|
Companies:
|
The Cleveland Clinic and The Cleveland Clinic
|
|
Address:
|
Department of Quantitative Health Sciences, Cleveland, OH, 44195,
|
|
Keywords:
|
errors-in-variables ; attenuation ; generalizability ; kidney function
|
|
Abstract:
|
Glomerular filtration rate (GFR) is the optimum measure of kidney function, but hard to measure. Thus, GFR is estimated from serum creatinine (SCR) and demographic factors using one of several equations developed by ordinary least squares regression. Recent reports have questioned the generalizability of these equations. We use errors-in-variables regression with study interactions to jointly relate SCR to GFR and demographic factors in 10 studies. We show that the majority of previously reported heterogeneity is due to predictable differences between studies in attenuation of regression coefficients resulting from differential error in SCR. The study interactions indicate remaining study variation not explained by attenuation. Our approach is applicable to settings in which a regression on easily measured predictors is used to estimate a more rigorous measure across populations.
|
