MATHEMATICAL MODELS TO PREDICT POLYPHENOLS CONTENT IN MENUS - A PILOT STUDY

The regular ingestion of phenolic compounds (PC) from in natura foods can contribute to reducing the risk of chronic diseases. Among the extrinsic factors that alter the PC content in foods is we can quote the degree of processing, considering the NOVA classification. A way to estimate the associations between those is through multiple linear models. This study evaluated the fit of regression models for estimating total polyphenol (Ply) and flavonoid (Flv) contents in Brazilian school menus. The prediction accuracy of the fitted models was evaluated by R², and absolute and relative (%) root mean square error (RMSE). There was a strong correlation between availability of fresh minimally processed foods (FMPF) and Ply and Flv contents. The fitted models presented R² of 0.55 for Ply and 0.74 for Flv. The data of FMPF and ultra-processed foods were the explanatory variables for the model with best fit for Ply; and the data of FMPF, processed culinary ingredients and processed foods were the ones used for the model with best fit for Flv. In the validation step, R² and RMSE were 0.51 and 115.4 mg (52%), and 0.71 and 2.8 mg (50%) for the models used for Ply and Flv, respectively. In addition, we mapped the distribution of polyphenols in the school menus municipalities analyzed. This study aimed to provide an alternative for the use of databases such as Phenol Explorer, especially when there are no conditions to carry out a large amount of chemical analysis.