Biostatistics Research Today is a free monthly online journal that collates and summarizes the latest research about Biostatistics, including details on statistics, uncertainty, probability, modeling.

Varying coefficients model with measurement error.

Li L, Greene T

Department of Quantitative Health Sciences, Cleveland Clinic Foundation, 9500 Euclid Avenue, Wb4, Cleveland, Ohio 44195, USA. lil2@ccf.org

We propose a semiparametric partially varying coefficient model to study the relationship between serum creatinine concentration and the glomerular filtration rate (GFR) among kidney donors and patients with chronic kidney disease. A regression model is used to relate serum creatinine to GFR and demographic factors in which coefficient of GFR is expressed as a function of age to allow its effect to be age dependent. GFR measurements obtained from the clearance of a radioactively labeled isotope are assumed to be a surrogate for the true GFR, with the relationship between measured and true GFR expressed using an additive error model. We use locally corrected score equations to estimate parameters and coefficient functions, and propose an expected generalized cross-validation (EGCV) method to select the kernel bandwidth. The performance of the proposed methods, which avoid distributional assumptions on the true GFR and residuals, is investigated by simulation. Accounting for measurement error using the proposed model reduced apparent inconsistencies in the relationship between serum creatinine and GFR among different clinical data sets derived from kidney donor and chronic kidney disease source populations.

Published 16 May 2008 in Biometrics, 64(2): 519-26.
Full-text of this article is available online (may require subscription).

© 2005-2008 Biostatistics Research Today. All Rights Reserved.