The development and validation of a digital food-frequency questionnaire to assess habitual diet: The PREDICT Food...

In the context of the rapidly evolving modern food environment, there is a need for dietary assessment tools to accurately assess current food intake. The objective of this research was to develop and validate a new food frequency questionnaire (FFQ). A 265 item web-based, self-administered FFQ (PREDICT-FFQ-v1) was developed using information from two FFQs and US and UK cohort level data. Over a 1-month period, 68 participants completed the PREDICT-FFQ-v1 and six 24-h recalls (reference tool; Intake24). Relative validity was assessed using Bland-Altman plots, cross-classification food group analysis, and Spearman's rank correlation for food groups and nutrients. From the Bland-Altman plots, the median bias between the PREDICT-FFQ-v1 and Intake24 were energy (414 kcal (95%CI 289 to 609) (1734 kj (95% CI 1208 to 2549)), carbohydrate (2.3 %EI, 95%CI 0.08 to 3.68), fat (2.5 %EI, 95%CI 0.23 to 3.52), protein (-1.14 %EI, 95% CI -1.64 to -0.23), and fibre (1.86 g/1000 kcal, 95% CI 1.05 to 2.65). Cross-classification by quartiles of diet quality (plant-based diet index) categorised >76% of participants into the same or adjacent quartiles. Correlation coefficients for food groups assessed using both tools ranged from 0.38 (legumes) to 0.78 (meat), and for energy-adjusted nutrients ranged from 0.28 (total fat) to 0.82 (alcohol). The PREDICT-FFQ-v1 captures intakes of foods that are commonly consumed in modern diets, with greater resolution at the food level than previous tools. The PREDICT-FFQ-v1 has acceptable relative validity in comparison to 24-h dietary recall and provides a valuable alternative for assessing dietary intake in nutritional epidemiology.