We assessed the reproducibility and validity of a semi-quantitative 156-item FFQ, which was designed to evaluate the dietary intake of the lactating mothers in China. In our study, the first FFQ was performed after 60 to 65 days postpartum. The FFQ is used to survey a person's dietary behaviour over a period of time. The interval of retest reproducibility ranged from a week to several years in different studies. Compared to the general population, the lactating mothers spend a lot of time on infant care, resulting in a lack of adequate sleep. If the interval between the two FFQs was relatively long, serious recall bias may be warranted. Therefore, the second FFQ was completed 5 weeks later during follow-ups, in which each FFQ recorded the participant’s diet for the previous month.
In the reproducibility study, the average Spearman rank correlation coefficients were 0.46 for food, and 0.43 for energy and nutrients, whereas the average intra-class correlation coefficients were 0.61 for food, and 0.44 for energy and nutrients. Similar to the reports of a previous study, most of the correlation coefficients in this study were above 0.4, thereby showing a good correlation . Among them, the correlation coefficients of eggs and poultry were higher than those of the others. Moreover, this finding was also confirmed in FFQ studies from pregnant women and children aged 7–9 years [40, 41]. A possible reason for the higher correlation coefficients of eggs and poultry could be that these special populations had special physiologic needs, resulting in an increase in the demand for high-quality protein; eggs and poultry are good sources of high-quality protein. Chinese dietary culture entails increasing the intake of eggs and poultry for the lactating mothers. However, as compared to other people with special physiologic needs, the diet of the lactating women has its particularity. For example, as compared to the correlation coefficient of fruits in the FFQ from pregnant women in previous studies [42, 43], the results in our study was relatively low. A reason for such may be due to the dietary taboo wherein the lactating mothers should eat less raw or cold food, such as fruits. Moreover, as compared to the FFQ studies from some coastal countries [44, 45], the correlation coefficients of some fatty acids in our study were low. A reason for this may be that the investigated region was an inland city of China, where consumption of marine food was less. Moreover, the correlation coefficients of vitamin A, vitamin C, iron, and selenium were also low in our study. A possible reason could be that the food sources with these nutrients were relatively limited. For example, animal liver, dark-green leafy vegetables, and fruits are rich in vitamin A. If these kinds of food were not eaten at the time of investigation, the intake levels of vitamin A of the lactating mothers would be greatly different.
As for the validity evaluation of the FFQ, previous studies mostly used 24 h- dietary recalls as a reference method [43, 46, 47]. In this study, instant photography was used as the reference method to record the 3-d dietary intake of the lactating mothers. Our previous study  found that when compared with the conventional 24 h-dietary recalls, instant photography could obtain the data of both food consumption and nutrient intake closer to the data obtained by the weighing method. Therefore, results obtained using instant photographs to assess portions were likely to have a higher degree of accuracy, improving the strength of the validation analysis.
Through statistical and validity analysis, we found that the average unadjusted spearmen rank correlation coefficients were 0.45 for food, and 0.49 for energy and nutrients. The correlation coefficients of most food and nutrients ranged from 0.4 to 0.7, showing moderate agreement. Similar to a previous FFQ study , correlation coefficients of most food and nutrients in this study decreased after energy adjustment. Specifically, the average correlation coefficient of food and nutrients decreased from 0.45 to 0.43 and from 0.49 to 0.32, respectively. This may be due to the large differences in energy intake among individuals. Similarly, variability was associated with an overestimation or an underestimation of systematic errors. In this study, due to the correction for changes in the daily intakes, the de-attenuated correlation coefficients increased, which was similar to the FFQ reliability studies from other special populations in China [23, 43]. In addition to the correlation coefficients, we used quartile cross-classification agreements and the Bland-Altman plots to evaluate the validity of the two methods. The percentages of agreement (the same or adjacent quartile) of food and energy, and nutrient intakes obtained from the FFQs and the 3DR, were 69.0% and 81.1%, respectively. Furthermore, the average percentages of the opposite quartile between the two methods for the intakes food and energy and nutrients were 9.2 and 4.3, respectively. These results were similar to those of previous studies on the FFQs from pregnant women [49, 50]. According to Masson’s study , when more than 10% of the participants were placed in the extreme quartile, the results were unsatisfactory. The misclassification rates of most food and nutrients in our study were less than 10%, which indicated that this study had good quartile agreement. On the Bland–Altman plots, overestimated or underestimated data could be clearly reflected. The closer the mean difference between the two methods was to zero, the narrower was the consistency interval. This indicated better agreement between the two methods . The Bland-Altman plots showed that the agreements of most foods and nutrients between the FFQs and 3DR were satisfactory.
Our study had some limitations. First, as compared to previous FFQs with fewer items, the FFQs used in this study contained 156 food items, which covered a wide variety of foods and seasonal variations in food consumption. As a result, this study was more specific and more accurate in dietary evaluation. However, the increase in food items meant that the lactating mothers needed more time to complete the FFQs, which would have increased the pressure of field investigation. Second, during the FFQ survey, we used face-to-face interviews and a food atlas to help the lactating mothers recall the types of food they ate and to improve the accuracy of food weight estimation. Therefore, whether it could be used in self-administered settings needs further verification. Third, we also took quality control measures in the process of design, implementation, and data collection to minimize bias. However, the dietary survey itself was complex and was coupled with a special group of lactating mothers who needed postnatal recovery and infant care. These conditions could have caused inevitable bias in data reporting. Finally, technological requirements of instant photography were higher than that of the 24 h-dietary recalls. The lactating mothers needed to have mobile phones, cameras, and other photo taking tools, as well as internet connection to transmit photos, which meant that it was not suitable for the lactating mothers in remote rural areas of China.