Reproducibility and Validity of a Chinese Food Frequency Questionnaire

Abstract Objective This study was design to develop a semi-quantitative Chinese Food Frequency Questionnaire (FFQ) and to conduct a validation study for the questionnaire. Methods Based on the survey experience in recent years, a new Chinese food frequency questionnaire (FFQ) with 149 items in 17 food categories was developed. A validation study on this new FFQ was conducted in Jiangsu and Beijing of China between 1999 and 2001. The period of study covered 1 year and the FFQ was validated by comparing with data obtained by a six repeated 24-hour recalls for 3 consecutive days, or a totally 18-day 24-hour recall throughout the year. A total of 271 healthy adult subjects were enrolled in the study. Food and nutrient intakes measured by the 18-day dietary recalls and food frequency questionnaires (FFQs) were computed in the National Institute for Nutrition and Food Safety, China CDC using the existing nutrition database. The average daily intake of foods and nutrients over the 18-day recall was used to compare with FFQ1 and FFQ2, which was conducted at the beginning and the end of the year, respectively. All statistical analyses were carried out using SAS software version 6.12. Results The reproducibility of FFQ in this study was evaluated at three levels between FFQ1 and FFQ2, i.e. comparison of the mean intake of foods and nutrients; correlation analysis of their intake; and cross-classification and agreement on their corresponding intake. The results showed a high degree of reproducibility for both foods and nutrients. Except for wheat flour and fishes, there were no significant differences in the mean intake of all other foods including rice, other cereals, fresh vegetables, salted vegetables, fresh fruits, nuts, pork, poultry, egg, milk, vegetable oil, soy sauce, salt, and liquor; and this is also true for all nutrients except thiamin. The correlation coefficients ranged from 0.43 to 0.90 for foods and 0.23 to 0.73 for nutrients. Relative validity was tested by comparing the results of food consumption and nutrient intake from both FFQ1 and FFQ2 with those from the average of the 18-day 24-hour recall. The relative validity of FFQ1 was performed in the absence of the possible bias due to a learning effect in FFQ2. This was closer to the real situation where subjects were deprived of any previous experience in quantifying their diet. However, the relative validation of FFQ2 covered the same period as the 24-hour recall. By comparing the mean intake of foods and nutrients between FFQ1 and FFQ2 and the 24-hour recall significant differences were revealed in most foods and nutrients. The crude correlation coefficients between FFQ1 and means of the 24-hour recall ranged from 0.12 to 0.87 for foods and from 0.33 to 0.63 for nutrients. The crude correlation coefficients between FFQ2 and the 24-hour recall ranged from 0.33 to 0.85 for foods and from 0.22 to 0.84 for nutrients. The strongest correlations were found for staple food (rice and wheat flour), pork, poultry and fishes, milk, and liquor. The weakest correlations were found for foods which are not consumed regularly such as potatoes, nuts, legume, and products; and also for fresh vegetables. Adjustment for energy and for attenuation improved correlation for nutrients. The correlation coefficients ranged from 0.27 to 0.86 for FFQ1 and the 24-hour recall and ranged from 0.39 to 0.99 for FFQ2 and the 24-hour recall. Discussion Reproducibility: In conducting a reproducibility evaluation, it is unrealistic to administer the questionnaire at a very short interval, such as in a few days or weeks. When a longer interval of time is used, true changes in dietary intake, as well as variation in response, may lead to reduced reproducibility. This study used one year as an interval between the two interviews which was the most desirable one used in other studies. The reproducibility of FFQ in this study was evaluated in three aspects between FFQ1 and FFQ2, i.e. comparison of the mean intake of foods and nutrients; correlation analysis of their intake; and cross-classification and agreement on their intake A high degree of reproducibility was shown for both food consumption and nutrient intake. Validity: Relative validity was tested by comparing the results of food consumption and nutrient intake from both FFQ1 and FFQ2 with those from the average of the 18-day 24-hour recall. The correlation of FFQ2 with the average of the 18 day 24-hour recalls was generally stronger than that of FFQ1. The relative validity of FFQ1 was performed in the absence of the possible bias due to a learning effect in FFQ2. This was closer to the real situation where subjects were deprived of any previous experience in quantifying their diet. However, the relative validation of FFQ2 covered the same period as the 24-hour recall. Among the available and feasible comparison methods of validating a FFQ, diet records are likely to have the least correlation with FFQ which are due to the restrictions imposed by a fixed list of foods, memory, perception of portion size, and interpretation of questions. These sources of error are minimally shared by diet records because diet records are open-ended, do not depend on memory (foods are recorded on a meal-by-meal basis), and allow direct assessment of portion size. The primary alternative for the use of diet records as a standard of evaluating FFQ is the collection of multiple 24-hour recalls. The results of an evaluation of relative validation depend on several factors which include choice of reference method, the degree of homogeneity of intake values within the population, recall period, and the number of the days recorded. The standard method in our study was a six repeated 24-hour recall for three consecutive days, or a totally 18-day 24-hour recall, over one-year period. Our study subjects were a group of adult residents with a fairly fixed lifestyle. These may partly contribute to stronger correlations obtained in our study. A trend that FFQ overestimates the mean intake for most of the food groups and nutrients included in the study has been observed. There have been few studies to reveal information on over- or under-estimates of both food consumption and nutrient intake by FFQ and the 24-hour recall. The overestimates of both food and nutrient intake in our study may possibly be explained by the fact that the 24-hour recall estimates of food and nutrient intake are derived directly from reports of actual diet of 18 days and that in FFQ the intake comes from summaries or averages of foods consumed during the year and the 18 day 24-hour recall may be not long enough to estimate individual one year diet intake since diet variation exists cross the season and day to day. The results of high degree of reproducibility strongly support the assumption that FFQ could reflect the one year dietary information of the individual. Despite some overestimation of both foods and nutrients by FFQs, agreement on cross-classification is comparable to what other studies have shown, and classification in the same quartile in our study shows a mean of over 45% agreement, while classification in the same and next quartile reveals an mean agreement over 75%. Application: FFQ developed by this study has been applied in several other studies including the Chinese National Nutrition and Health Survey in 2002. Conclusion In this study, the reproducibility and validity of FFQ were all satisfactory. The results have shown that FFQ can be used to classify study subjects according to their food consumption or nutrient intake over a one-year period. These findings have also confirmed that FFQ is an appropriate instrument to measure the usual food consumption and nutrient intake, as well as to assess the dietary patterns of adult Chinese. It could be used in studies with different purposes, especially in studying the relationship between diet, nutrition and chronic diseases.