Page 108 - Fertility in Women with Rheumatoid Arthritis Vruchtbaarheid van vrouwen met reumatoïde artritis

Chapter 7
a conversion factor was used to compare AMH levels in the patient group to those in the controls:25
AMH (picoAMH assay in ng/mL) = (1.45 * AMHGen II) + 0.32
Statistical analysis
Values are presented as mean ± SD for normally distributed variables, as median (IQR) if values were non-normally distributed, and as number (%) for dichotomous variables. The number of missing values are given for each variable.
To approach a normal distribution, AMH levels were log-transformed for analysis. To correct for different distribution of ages between patients and controls at the different time points, AMH levels were compared to controls using analysis of covariance (ANCOVA) on log-transformed AMH levels with adjustment for age.
Sensitivity analyses were performed by excluding women using combined oral contraceptives, and by excluding women using any steroid sex hormones.
To reduce potential bias in the longitudinal analysis due to missing values, multiple imputation was performed. To incorporate the longitudinal outcome into the imputation procedure, a preliminary linear mixed model (LMM) with a random intercept and random slope was tted, using only completely observed covariates, and age as time variable. The random intercept and slope estimated by this model were considered a summary of the outcome, and were added as predictor variables to the imputation models.27
To study the effect of RA-related factors (disease duration, presence of ACPA, presence of RF, presence of erosions, past use of MTX) on the AMH levels, a LMM with random intercept and slope was built, using biological age as time variable. To obtain a more normal distribution of residuals and random effects in the LMM, a square root transformation was applied to the AMH levels. Interactions between ACPA, RF, and erosions, and between age and disease duration, were considered in the model. Pooled results from the 10 multiple imputed datasets are presented.
Since gynaecological age (i.e. years since menarche) may more precisely explain the inter-individual difference than biological age, a second analysis was performed using gynaecological age as time variable.
Statistical analysis was performed using Stata SE v14 (College Station, Texas, USA) and R version 3.3.1 (2016, the R foundation for statistical computing; packages used: mice, lme4).
106