Performance of a Food Frequency Questionnaire for Estimating Ultraprocessed Food Intake According to the Nova Classification System in the United States NIH-AARP Diet and Health Study

Background How accurately ultraprocessed food (UPF) intake is measured by food frequency questionnaire (FFQ) is unknown. Objectives We evaluated the performance of a 124-item FFQ for estimating UPF intake with Nova. Methods In 1311 NIH-AARP Diet and Health Study participants who completed 2 nonconsecutive 24-hour dietary recalls (24HDR) and 2 FFQ, we used 2 methods to estimate UPF intake. First, 4 experts independently matched FFQ items to food descriptions from another FFQ classified previously according to Nova. Discordant descriptions were discussed, and a consensus reached. Second, data from the 1994-1996 Continuing Survey of Food Intake of Individuals were used to match disaggregated FFQ items to Food and Nutrient Database for Dietary Studies (FNDDS) and standard reference (SR) codes. Nova classification of FNDDS and SR codes was based on database linkage. We evaluated FFQ performance using a measurement error model with 2 24HDR as the reference to estimate validity coefficients (rho(Q,T)) and attenuation factors (lambda(Q)). We applied parametric bootstrapping to construct 95% confidence intervals, accounting for the repeated measures structure in the data. Results For energy-adjusted, kilocalories from UPF (kcal/d), rho(Q,T) were 0.50 (0.39-0.61) and 0.44 (0.30-0.59), for males and females, respectively, for the expert consensus method (ECM) and 0.52 (0.40-0.63) and 0.43 (0.29-0.57) for the food code method (FCM). For energy-adjusted, gram weight from UPF (g/d), rho(Q,T) were 0.65 (0.60-0.71) and 0.66 (0.60-0.72) for the ECM and 0.66 (0.52-0.65) and 0.66 (0.59-0.72) for the FCM; lambda(Q) were >= 0.50 for both methods. UPF variables, defined using the ECM and FCM but expressed in the same unit, were highly correlated (r >= 0.97). Conclusions FCM, performed similarly to ECM but has the potential to standardize UPF exposures across studies because the Nova categorization is assigned at a more granular level through database linkage. UPF intake based on energy-adjusted gram weight outperformed intake based on energy alone.