2.4. Statistical analysis

Box and whisker plots were constructed (GraphPad Prism 9.0.0; GraphPad Software, La Jolla, California) for each toxin in part 1 (short‐term) and part 3 (medium‐term) to qualitatively describe variance between individuals (CV_G) and variance within individuals (CV_I). Statistical analysis was performed to determine biological variation estimates according to previously reported guidelines using a specialized statistical software (SAS, Version 9.4; SAS Institute, Cary, North Carolina). 12

Residual diagnostic plots were used to evaluate model assumptions of normality and equal variance. Because of skew and unequal variance, a (natural) log transformation was used for each toxin in each study part to better satisfy model assumptions. Log transformed data were assessed for outliers using a 3‐step process. Data were assessed for outliers by evaluating values falling outside 3 times the interquartile range (a) across all cats, (b) within each cat, and (c) using cat level averages (where a single average was calculated for each toxin, study part and cat). For part 1, no outliers were identified. For pCS in part 2, a single time point for a single cat (both duplicates) was identified as an outlier. For pCS and TMAO in part 3, a single time point for a single cat (both duplicates) was identified as an outlier. All outliers were identified when looking “within cat” but not at the other levels. All observations were retained in the analysis because the outlying values were similar between duplicates and thus were unlikely to be because of analytical error.

A random effects model was fit using restricted maximum likelihood (REML) with SAS Proc Mixed. Cat and time point (nested within cat) were included in the model as random effects. Hence, the variance was partitioned into 3 components for each toxin: (a) CV_G, (b) CV_I, and (c) variation between duplicates (CV_A).

Because the analysis was done on the log transformed scale, the (back transformed) coefficients of variation (CV) were calculated, as previously described, 19 , 20 using the equation:

The II was calculated from the CVs using the formula 12 :

Because the analysis was done on the log transformed scale, RCVs were calculated using the lognormal approach, as previously described. 19 , 20 Specifically:

where Z = 1.96 for 2‐sided interpretation and σ ² represents a log scale variance component.

To determine if recent feeding should be considered a variable when assessing serum concentrations collected at the same time of day, serum IS, pCS, and TMAO concentrations (from only cats that completed both part 1 and part 2 of the study; n = 9) were compared between the fed state and nonfed state at hours 2, 4, 6, 8, 10, and 12 using a paired t test or Wilcoxon matched‐paired signed rank test. To determine whether the time of day should be considered a variable when assessing serum concentrations, a repeated measures 1‐way ANOVA with the Geisser‐Greenhouse correction followed by a Tukey's post hoc analysis or a Friedman test with Dunn's post hoc analysis was used to compare serum IS, pCS, and TMAO concentrations between different time points over a 12‐hour period in the nonfed state (hours 0, 2, 4, 6, 8, 10, 12) and over a 10‐hour period in the fed state (hours 2, 4, 6, 8, 10, 12). The average of the duplicates taken at each time point for each cat was used in the analysis. Data were log‐transformed to base‐10 to meet the assumption of normality. If normality was not met then a nonparametric test was performed on the raw values. Statistical analysis was performed using a statistical software program (GraphPad Prism 9.0.0; GraphPad Software). A P‐value <.05 was considered significant.