Food Predictors of Plasma Carotenoids

Abstract

Empirical prediction models that weight food frequency questionnaire (FFQ) food items by their relation to nutrient biomarker concentrations may estimate nutrient exposure better than nutrient intakes derived from food composition databases. Carotenoids may especially benefit because contributing foods vary in bioavailability and assessment validity. Our objective was to develop empirical prediction models for the major plasma carotenoids and total carotenoids and evaluate their validity compared with dietary intakes calculated from standard food composition tables. 4180 nonsmoking women in the Nurses’ Health Study (NHS) blood subcohort with previously measured plasma carotenoids were randomly divided into training (n = 2787) and testing (n = 1393) subsets. Empirical prediction models were developed in the training subset by stepwise selection from foods contributing ≥0.5% to intake of the relevant carotenoid. Spearman correlations between predicted and measured plasma concentrations were compared to Spearman correlations between dietary intake and measured plasma concentrations for each carotenoid. Three to 12 foods were selected for the α-carotene, β-carotene, β-cryptoxanthin, lutein/zeaxanthin, lycopene, and total carotenoids prediction models. In the testing subset, Spearman correlations with measured plasma concentrations for the calculated dietary intakes and predicted plasma concentrations, respectively, were 0.31 and 0.37 for α-carotene, 0.29 and 0.31 for β-carotene, 0.36 and 0.41 for β-cryptoxanthin, 0.28 and 0.31 for lutein/zeaxanthin, 0.22 and 0.23 for lycopene, and 0.22 and 0.27 for total carotenoids. Empirical prediction models may modestly improve assessment of some carotenoids, particularly α-carotene and β-cryptoxanthin.

Keywords

• carotenoids
• vitamin A
• α-carotene
• β-carotene
• β-cryptoxanthin
• lutein/zeaxanthin
• lycopene
• food predictors
• biomarkers