LAUSR

The study by Belzeaux et al. (2017) reminds researchers that n measuring blood cytokine levels, the measurements made by he Luminex platform could not be totally reproduced by the Vlex platform, and the serum measurements could not be totally eproduced by the plasma measurements. I would like to expand he limitations of this study. First, the authors compared the cytokine levels measured by wo different multiplex immunoassays and examined linear correation between plasma cytokine levels and serum cytokine levels. hey also investigated whether the IL-6 levels measured by muliplex immunoassays were comparable to those measured by nzyme-linked immunosorbent assay. However, the authors used pearman’s rank correlation. The correlation coefficient is not ensitive to random variation and systematic bias, and it only ooks at the relative position of the same subjects on different onditions (Atkinson and Nevill, 1998). However, bias may exist etween measurements made by two different immunoassays, and ntra-individual variability may be different for plasma and serum amples. We recommend to assess 95% limits of agreement and rovide the Bland-Altman plots (Bartlett and Frost, 2008). Readers an estimate how large the disagreements between two different lood samples and two different immunoassays will be on 95% of ccasions. Second, the authors investigated whether the cytokine levels ere stable over time in healthy subjects. By using intraclass corelation coefficient (ICC), the authors examined the ICC over a long eriod (day 1, day 14, day 56, and day 210). The ICC is calculated rom a one-way analysis of variance model, which assume that here are no systematic biases between measurements within subects and that the within-subject standard deviations are equal for ll measurements (Bartlett and Frost, 2008). However, the chance hat the underlying cytokine levels are constant over a 210-day eriod is unlikely. Importantly, the ICC is highly dependent on the heterogeneity f study population (Bartlett and Frost, 2008). The ICC reliability of ultiplex immunoassays in healthy controls could not be generlizable to depressed patients and other psychiatric patients. We ecommend to calculate the repeatability coefficient and provide he Bland-Altman plots (Bartlett and Frost, 2008), helping readers o grasp the absolute difference and quantify the test-retest reliaility. Moreover, to avoid a potential bias between measurements, shorter retest interval is recommended. If the authors still want to xamine test-retest reliability with a 30-week interval, the authors