Relative contributions of polyphenolic constituents to the antioxidant status of wines: Development of a predictive model

The concentrations of 17 phenolic constituents in red wine were analyzed by a number of multiple regression models for their contribution to total antioxidant status (TAS). The best model discovered involved a stepwise selection process starting by calculating Mallow's C(p) for each of the 17 subsequent individual models resulting from dropping a single term at a time. The model with the lowest Mallow's C(p) and in which no single step reduced Mallow's C(p) further was chosen as the new model. This model gave excellent correlation between observed and predicted values of TAS. Further analysis of this model with the addition of a nonlinear term, using the 8 predictors discovered from the stepwise selection model, increased the correlation coefficient to 0.961 from 0.930. The equation of this model can predict almost 100% of the TAS by the content of vanillic acid, trans-polydatin, catechin, m-coumaric acid, epicatechin, quercetin, cis-polydatin, and trans-resveratrol with a very high degree of confidence. Although syringic and gallic acids were significantly correlated with TAS in a univariate analysis, they do not contribute to a statistical description of this parameter to the extent of the eight constituents already identified. The remaining constituents do not seem to contribute significantly to TAS, nor do they show significant correlation with TAS on univariate analysis.