2.6. Statistical Analysis

A one-sample Kolmogorov–Smirnov test was used to examine the normality of energy, macronutrients, and fiber. These data were log transformed to improve normalization of the skewed distribution before analysis. For validation, differences between the energy and macronutrient intake values derived from the FFQ or the three 24-HDRs were determined using paired t test. The Pearson correlation was used to determine correlations between energy and macronutrient intake derived from the above two dietary assessment tools. Partial correlations were calculated with data adjusted for energy intake according to residual methods [28]. For agreement analysis, we further examined the quartile distribution of energy-adjusted nutrient intakes between the two methods in order to classify whether they fell into the same, adjacent, one-quartile-apart, or far apart quartiles. The Bland–Altman method [29] was also used to assess agreement between nutrient intake values obtained using two dietary assessment tools.

Simple linear regression was employed to test the trends of plasma carotenoids concentrations across tertile intake levels of different food groups. Independent associations between tertile intake levels of light-green vegetables, dark-green vegetables, or fruits obtained from the FFQ and carotenoids were tested using multiple linear regression of backward elimination adjusting for potential confounding factors, including age, gender, energy intake based on the FFQ, diabetes duration, education, hemoglobin A1c (<6.5%, 6.5–9% and >9%), and use of vitamin supplements (multivitamins, vitamin A or vitamin C).

All statistical operations were performed using SPSS Version 22.0 (SPSS Inc., Chicago, IL, USA). A p < 0.05 was considered significant.