3 CHAPTER 3 54 and blank samples) were corrected with a regression model generated through the data acquired from regular QC injections.16 Concentration calculations were performed by using the Pearson’s linear correlation within the calibration range for both CSF and plasma samples. The R2 for each calibration line was selected ≥ .95. Based on corrected area ratios, the analytes were defined as not detected if 30% of the mean value of all study samples is equal or less than the mean value of the lowest calibration level. The final concentration values were reported in µmol/L (µM) unit. Statistical analysis Quality control Outlier detection was performed using principal component analysis (PCA) with a 99% confidence interval. Prior to the statistical analyses data were 10log-transformed and values below [mean - 4 * SD] and above [mean + 4 * SD] were replaced with these cut-offs. Single-metabolite plasma/CSF correlations After log10-transformation, metabolite levels were adjusted for age, sex and age*sex and were subsequently inverse normal transformed to obtain robust estimates of the plasma/CSF correlation coefficients. Plasma/CSF correlation coefficients r were compared between groups by normalizing r with Fisher’s transformation and subsequently testing the difference in z-values between groups against the centered normal distribution , with n1 and n2 the sample sizes of groups 1 and 2, respectively. Plasma/CSF metabolite ratio correlations In addition to determining plasma/CSF correlation coefficients for all single metabolites, also correlation coefficients were calculated for the ratios between metabolites. Prior to the correlation analysis the ratios were log10-transformed, adjusted for age, sex, age*sex and inverse normal transformed. Since some ratio correlation coefficients seemed higher than the correlation coefficients of the related single metabolites, we tested for this using the Steiger’s approach for comparing correlations coefficients in overlapping data.17 Specifically for a ratio of metabolites M1 and M2, we tested the difference between the Fisher transformed correlation coefficient of the ratio and the average z of both metabolites one-sided against the centered normal distribution , with si the correction factor for overlapping data of M1/ M2 relative to respectively M1 and M2, as defined in the Steiger’s approach.17 Finally, we
RkJQdWJsaXNoZXIy MTk4NDMw