Z. Daher Abdi (1, 2), JB. Woillard (1,2), A. Prémaud (1,2), Y. Le Meur (3), P. Marquet (1, 2,4), A. Rousseau (1,2).
(1) Inserm, UMR-S850, Limoges, F-87025, France ; (2) Université Limoges, F-87025, France; (3) CHU Brest, Département de Néphrologie, Brest, France; (4) CHU Limoges, Pharmacologie, Toxicologie et Pharmacovigilance, Limoges, F-87042, France.
Objectives:
The goal of this study was to evaluate the influence of MPA exposure on the occurrence of anemia in a one year longitudinal study using 3 modeling approaches.
Methods:
One hundred thirty renal transplant patients (APOMYGRE trial) treated with MPA were analyzed (1). Both MPA area under the curve (AUC) previously estimated (2) (2.5th-97.5th percentiles: 13-73 mg.h/L) and the presence of anemia (i.e. hemoglobin -Hb- < 10 g/dL) were collected at different visits post graft.
The association between anemia and the previous measurement of MPA AUC carried forward was investigated using: i) a regression logistic with fixed and random effects (GLMM),ii) a regression logistic using the generalized estimating equations (GEE) and iii) a Markov mixed effects model. GLMM and GEE models were fitted in R using lme4 (3) and geepack (4) packages respectively. Markov models were fitted in NONMEM® using the Laplacian method; the probabilities of occurrence anemia and return to normal Hb level were estimated.
Effect of covariates (donor and recipient ages, sex, dosing strategy, baseline Hb level) was investigated. Odds ratios (OR) and 95% confidence interval (CI) of OR were calculated for the 3 models. For the Markov model, the 95% CI of OR were based on the 2.5th and 97.5th percentiles of parameter estimates obtained from 500 bootstrap samples. The Markov model was evaluated by performing posterior predictive checks (PPC) using 200 simulated datasets.
Results:
The Hb baseline and the MPA AUCs were significant predictors of anemia with the 3 approaches. In the GLMM approach, visit, baseline Hb levels and MPA AUC were used as fixed effects; random effects were added on visit and subjects (as correlated random effects). With the GEE approach, an unstructured matrix was used for correlation within subjects. The ORs associated with a one–unit increase of AUC [95%CI] were: 1.027 [1.001-1.053], p=0.044 for the GLMM model; 1.016 [1.002-1.031], p=0.029 for the GEE model and 1.020 [1.003-1.444] for the Markov model. The PPC showed that, for the transition to anemia, the observed number was within the 95% prediction interval.
Conclusions:
The 3 models evidenced a significant MPA exposure-anemia relationship; OR were estimated with good precision. 95% CI OR was smaller using GEE, this advocates for this method when subject-specifics parameter estimates are not of special interest. Markov model could allow to identify predictors to obtain an Hb >10 g/dL (reverse transition).
References:
[1] Le Meur Y, Büchler M, Thierry A, et al. Individualized mycophenolate mofetil dosing based on drug exposure significantly improves patient outcomes after renal transplantation. Am. J. Transplant. 2007;7(11):2496-2503.
[2] Prémaud A, Le Meur Y, Debord J, et al. Maximum a posteriori bayesian estimation of mycophenolic acid pharmacokinetics in renal transplant recipients at different postgrafting periods. Ther Drug Monit. 2005;27(3):354-361.
[3] Douglas Bates, Martin Maechler and Ben Bolker. lme4: Linear mixed-effects models using S4 classes (2012). R package version 2.15.0. http://CRAN.R-project.org/package=lme4
[4] Højsgaard, S., Halekoh, U. & Yan J. The R Package geepack for Generalized Estimating Equations Journal of Statistical Software, 15, 2, pp1-11 (2006).
Reference: PAGE 22 () Abstr 2870 [www.page-meeting.org/?abstract=2870]
Poster: New Modelling Approaches