Statistical analyses

Arithmetic means, standard deviations, medians, intra- and inter-assay CVs were estimated using routine descriptive statistical procedures. The DL (P < 0.01) was estimated as the mean + 3 SD of SAA determination of blank samples. Linearity under dilution was investigated by linear regression analysis. Runs test was performed to determine whether data deviated significantly from the applied model. The effect of different concentrations of haemoglobin or lipid on measured SAA concentrations were examined by calculating the bias percentage as described by Kjelgaard-Hansen and Jensen [27] (Additional file 1). Method comparison of the two assays was performed by Deming regression analysis and visualized in Bland-Altman plots. Calculation of RI and 90% confidence interval was performed with a dedicated software (Reference Value Advisor, http://www.biostat.envt.fr/reference-value-advisor/) using nonparametric methods due to the skewed distribution of native data (determined by the D’Agostino-Pearson omnibus test) as described by Geffre et al. [28]. Overlap performance was assessed by comparing groups (healthy, non-inflammatory disease, inflammatory disease) using the Kruskal–Wallis test; when significant results were obtained, Dunn’s multiple comparison test was performed. Differences between groups with different inflammatory status (healthy, non-inflammatory, local inflammation, systemic inflammation) were analysed with one-way ANOVA and the post hoc test Tukey’s multiple comparisons test. Changes in SAA levels in castrated horses and horses with LPS-induced inflammation were evaluated using the repeated measures ANOVA and Tukey’s multiple comparisons post hoc test. A level of significance of 0.05 was used unless otherwise stated. Statistics were performed using a commercial package (Graph PadPrism 8.00 for windows, GraphPad Software, La Jolla, California, USA).