Potential error sources in nutrient estimation with the FFQ include inaccurate or biased recall and overestimation or underestimation of intake due to too many or too few items on the FFQ, respectively. Here we report the refinement of an FFQ that overestimated nutrient intake and its validation against multiple 24 h recalls.

Data on 2527 participants in south India (Trivandrum) were available for the original FFQ (OFFQ) that overestimated nutrient intake (132 food items). After excluding participants with implausible energy intake estimates (<2·72 MJ/d (<650 kcal/d), >15·69 MJ/d (>3750 kcal/d)) we ran stepwise regression analyses with selected nutrients as the outcomes and food intake (servings/d) as predictor variables (n 1867). From these results and expert consultation we refined the FFQ (RFFQ), and validated it by comparing intakes obtained with it and the mean of two 24 h recalls among 100 participants.

The OFFQ overestimated usual daily nutrient intake before and after exclusions [for energy: 13·39 (sd 5·46) MJ (3201 (sd 1305) kcal) and 10·96 (sd 2·65) MJ (2619 (sd 634) kcal), respectively]. In stepwise analyses, fifty-seven food items explained 90 % of the variance in nutrients; we retained thirteen food items because participants consumed them at least twice monthly and twelve food items that local nutritionists recommended. Mean energy intake estimated from the RFFQ (eighty-two food items) was 7·94 (sd 2·05) MJ (1897 (sd 489) kcal). The de-attenuated correlations between mean 24 h recall and RFFQ intakes ranged from 0·25 (vitamin A) to 0·82 (fat).

We refined an FFQ that overestimated nutrient intake by shortening and redesigning, and validated it by comparisons with 24 h dietary recall data.