MDR-1 and MRP-1 activity in peripheral blood leukocytes of rheumatoid arthritis patients

MAF measurements were modeled using analysis of covariance (ANCOVA) with residual error stratified in patient groups to allow for unequal variances. Models were fitted using the R statistical system (R Development Core Team 2007, ISBN 3-900051-07-0), with the GLS (generalized least squares) procedure of the NLME package (R Core team, 2007, R-package version 3.1-86). Models were compared by using maximum likelihood (ML) fits, and Wald-type significance tests were performed using restricted maximum likelihood (REML) estimates. Differences were considered statistically significant below the level of p = 0.05. The following predictors were considered during model selection: patient group (5 levels), sample storage duration (continuous), drug doses of methotrexate, delagil, arava, medrol (continuous), administration of the same drugs (yes/no), administration of NSAIDs (yes/no). We have selected the best model using the Akaike Information Criterion (AIC) (Akaike 1974: A new look at the statistical model identification. IEEE Transactions on Automatic Control 19 (6): 716–723.) All subsets regression was not computationally feasible due to the large number of predictors, so forward and backward stepwise variable selection algorithms were applied. When the stepwise algorithm yielded a model where the same drug was included both as a continuous and binary predictor, we checked if the coefficients corresponded to a monotonous dose-effect relationship. In case of non-monotonicity we have considered the two parameters for the same drug as overfitting, dropped the model and the stepwise algorithm was restarted twice, first omitting the binary dose variant, second omitting the continuous one. When the stepwise algorithms led to different models, the one with lowest AIC value was selected.