(OR22-04-24) Performance of a Food Frequency Questionnaire for Estimating Ultra-Processed Food Intake According to the Nova Classification System in the US NIH-AARP Diet and Health Study

Objectives: We evaluated the performance of a food frequency questionnaire (FFQ) for estimating ultra-processed food (UPF) intake according to the Nova classification using two alternative methods.

Methods: In a subgroup (n=1907) of the NIH-AARP Diet and Health Study, which enrolled more than half a million US adults from 1995 to 1996, we used two methods to classify 204 FFQ items into the four Nova groups, ranging from unprocessed/minimally processed foods (group 1) to ultra-processed foods (group 4). First, four experts independently linked FFQ items to a Nova group based on a similar food list from the Nurses’ Health Studies I and II and the Health Professionals Follow-up Study that was previously classified. Discordant items were discussed, and a final consensus classification reached. Second, we used data from the USDA’s 1994–1996 CSFII to disaggregate FFQ items into 3,513 FNDDS food codes, which were further broken down into ingredient codes; Nova classification of food and ingredient codes was based on prior work in the National Health and Nutrition Examination Survey (NHANES). We estimated mean daily UPF intake from two non-consecutive 24-hour dietary recalls, also based on NHANES, and used this as the reference to evaluate FFQ performance.

Results: Pearson correlations for energy from UPF (kcal/day) estimated from 24-hour dietary recalls and FFQ were 0.27 and 0.18, for men and women, respectively, for the expert consensus method and 0.28 and 0.19 for the food code method. For energy-adjusted, gram weight from UPF (g/day), correlations for men and women were 0.48 and 0.51 and 0.49 and 0.51, for expert consensus and food code methods, respectively. Correlations between FFQ-based UPF intake defined using the two methods were high for both energy and energy-adjusted gram (r≥0.97) variables.

Conclusions: The food code-based approach, which involved disaggregating FFQ items and database linkage, performed similarly to the expert consensus approach but has the potential to improve reproducibility since it does not rely on de novo Nova classification. FFQ estimated UPF intake based on energy-adjusted gram weight outperformed intake based on energy alone.

Funding Sources: NIH Intramural Research Program