2.3.1. General Analyses

The sample size for the study was calculated assuming a 30% difference in the mean for one of the main study variables (plasma level of docosahexaenoic acid) between children with ASD and healthy children, an α error = 0.05, a power of 0.90 (β error of 0.1) and 5% of dropout during the follow-up of the study. For these calculations, data from recent studies were used, considering that the percentages expected would be similar to several studies carried out in populations with similar characteristics to those of children with ASD. Data were expressed as mean ± standard deviation or median plus interquartile range, depending on whether each variable's values followed a normal distribution or not. The Shapiro-Wilk normality test was used to determine the normality of variables. The comparison between the groups was carried out using the Student's *t*-test for continuous variables when the distributions were normal and the Mann-Whitney U test when the

distributions did not follow the normality. The differences between the frequencies of the sexes were studied by means of the Chi-square test. For the statistical analysis of the data, the computer program IBM SPSS 25.0 (IBM Corp., Armonk, NY, USA) was used.

#### 2.3.2. Dietary Patterns

A principal component analysis (PCA) was accomplished to identify underlying DPs using each individual's serving average from nine food groups as input variables [31]. These food groups considered in this study were as follows: "milk and dairy products (e.g., milk, yogurt, cheese, and milkshakes)", "cereals and pasta (e.g., bread, cereal bars, and rice)", "fatty meat and derivates", "fats", "snacks, sweets, bakery and pastry (e.g., chocolate, cookies, ice cream, candies, and bakery products)", "fruits and vegetables", "beverages", "fish and shellfish (e.g., white fish, bluefish, and shellfish", and "lean meat and eggs". This mathematical model calculates new variables (principal components) that account for the variability in the food group's data and enables the study of covariances or correlations between variables. We interpreted only components with eigenvalues over one and factor loadings with an absolute value higher than 0.4 (which explains around 16% of variance) as the significance of factor loading depends on the sample size. A high factor score for a given pattern indicated a high consumption of the foods constituting that food factor, and a low score indicated a low intake of those foods.

The Kaiser–Meyer–Olkin (KMO) and Bartlett test of sphericity were applied to assess the sampling adequacy. KMO values >0.50 were considered. Communalities were estimated using the squared multiple correlations of each variable with all others. We retained variables with communalities higher than 0.5. Factors were orthogonally rotated (the Varimax option) to maximize the dispersion of loading within factors, facilitating interpretability. Radar maps were used to display data in the form of a two-dimensional map of nine food groups represented on axes starting from the same point.
