Derivation and Validation of a Total Fruit and Vegetable Intake Prediction Model to Identify Targets for Biomarker Discovery Using the UK National Diet and Nutrition Survey

Owen, Elliot J ORCID: https://orcid.org/0000-0002-1850-2645, Patel, Sumaiya, Flannery, Orla ORCID: https://orcid.org/0000-0002-4669-2781, Dew, Tristan P and O'Connor, Laura M (2021) Derivation and Validation of a Total Fruit and Vegetable Intake Prediction Model to Identify Targets for Biomarker Discovery Using the UK National Diet and Nutrition Survey. The Journal of Nutrition, 151 (4). pp. 962-969. ISSN 0022-3166

Preview

Accepted Version Download (993kB) | Preview

Abstract

Background: Dietary assessment in research and clinical settings is largely reliant on self-reported questionnaires. It is acknowledged that these are subject to measurement error and biases and that objective approaches would be beneficial. Dietary biomarkers have been purported as a complimentary approach to improve accuracy of dietary assessment. Tentative biomarkers have been identified for many individual fruit and vegetables (FV) but an objective total FV intake assessment tool has not been established. Objective: We aimed to derive and validate a prediction model of total FV intake (TFVpred) to inform future biomarker studies. Methods: Data from the National Diet and Nutrition Survey (NDNS) were used for this analysis. A modelling group (MG) consisting of participants aged >11 years from the NDNS years 5-6 was created (n=1746). Intake data for 96 FVs were analysed by stepwise regression to derive a model that satisfied three selection criteria: standard error of the estimate (SEE) ≤80, R2>0.7, and ≤10 predictors. The TFVpred model was validated using comparative data from a validation group (VG) created from the NDNS years 7-8 (n=1865). Pearson’s correlation coefficients were assessed between observed and predicted values in the MG and VG. Bland-Altman plots were used to assess agreement between TFVpred estimates and total FV intake. Results: A TFVpred model, comprised of tomatoes, apples, carrots, bananas, pears, strawberries and onions, satisfied selection criteria (R2=0.761, SEE=78.81). Observed and predicted total FV intake values were positively correlated in the MG (r=0.872, P<0.001, R2=0.761) and the VG (r=0.838, P<0.001, R2=0.702). In the MG and VG, 95.0% and 94.9% of TFVpred model residuals were within the limits of agreement, respectively. Conclusions: Intakes of a concise FV list can be used to predict total FV intakes in a UK population. The individual FVs included in the TFVpred model present targets for biomarker discovery aimed at objectively assessing total FV intake. Keywords: fruit and vegetables, prediction model, dietary assessment, biomarkers, dietary questionnaires.