Accuracy of different reduced versions of a validated food-frequency questionnaire in italian men and women.

ABSTRACT

This study aimed to explore how possible it would be to reduce the length of a food-frequency questionnaire (FFQ) while maintaining adequate accuracy. This FFQ was validated in women in a previous study by our group, whereas the validation in men was one of the aims of the present study. A 15-item FFQ was administered to 136 men and 211 women (age range 25-75 years) recruited by proportional age-stratified sampling, who agreed to fill in a 14-day diet diary. For the intermediate version of the FFQ we considered only eight items in women and 10 in men. For the very short version we considered only three items. The positive likelihood ratio was markedly lower for the very short version (men = 1.2 and women = 1.5) with respect to the intermediate version (men = 6 and women = 6.3) and the 15-item version (men = 19.6 and women = 19.9). The specificity of the very short version was low (22.2% for men and 35.2% for women). The area under ROC curve of the three-item FFQ was significantly lower with respect to the other two versions. In conclusion, our study suggests that the shortest version of the FFQ could be usefully employed only in those clinical settings where the main objective is the identification of individuals who do not have a low calcium intake. ROC analysis and posttest probability calculated by positive likelihood ratio may represent an optimal method for assessing the accuracy of FFQs.