Dietary assessments from food records (FR) are commonly used to assess children’s food and nutrient intake, and possible relationships between dietary intake and health outcomes [1
]. In infants and young children, parents are asked to record everything that the child has eaten and drunk during a predefined time period [2
]. However, achieving accurate and reliable dietary intake data can be difficult and demanding: for the parents, the process may be tedious and time consuming [3
], and for the clinician or researcher, the generated data may be subject to bias, making interpretation difficult [4
]. Meals with complex content, such as main course meals with several ingredients, are challenging to remember, record, and to determine the amounts of the various ingredients [5
]. In young children, food records tend to overestimate energy intake, e.g., parents may misreport the child’s intake by failing to omit food leftovers and spillage from the FRs [4
]. In order to better understand the complex relationships between diet and health in young children, it is important to develop dietary assessment methods with higher accuracy and precision [7
Mobile phone applications and cameras have been shown to improve self-reported dietary intake, and they have likewise increased participants’ user satisfaction, compared to conventional methods [8
]. Moreover, information obtained from images seems to reduce random and measurement errors for energy intake (EI), especially when it comes from complex and diverse foods. In adults, EI is often underestimated, but this can be corrected with image-assisted dietary assessments [9
]. This was also found among overweight and obese children by using digital camera FRs [10
]. Previous research in pre-school children has shown that EI assessment using images was not significantly different, compared to measuring total energy expenditure (TEE) with doubly labeled water (DLW) [11
]. However, no study to date has investigated methods of active image-assisted FRs in infants [9
The aim of this study was to compare total and main course meal energy and macronutrient intake in 12-month-old, healthy infants using an active image-assisted five-day FR against a regular five-day FR, and to validate the total energy intake measured with the image-assisted food record method against TEE using DLW.
Previous studies in infants have shown the non-random error to be 5–15% when comparing energy expenditure, measured with doubly labeled water to recorded dietary intake [1
]. Such bias increases the risk of type II-error and diminishes the power of the study. Image-assistance, where FRs are complemented with images before and after eating sessions is one possibility to reduce this bias [9
In the present study, comparing total EI in regular and image-assisted FRs, the difference between methods was only 3.9 kJ and not statistically different. The reliability between image-assisted FR and ME, and the golden standard to assess EI, using ICC was excellent. The bias of 366 kJ means that image-assisted FRs overestimated the EI compared to ME by 10%. This bias was higher but the limits of agreement were narrower than reported in similar studies [4
]. Previous studies have shown that image assistance has been successful in reducing underreporting, but to our knowledge no study has used the technique in settings when over reporting is an issue, as was the case in the present study [9
Unaccounted leftovers, i.e., the parts of the meal that are left on the plate or that are lost due to spillage, are possible sources of systematic error if they are not subtracted from the estimated intake. In the present study, 60% of the meals that were recorded with active image assistance showed leftovers. However, in more than half of these eating episodes, the parents correctly modified the FR to include the leftovers, and in the other half, where the research dietician did adjust the recording, it was equally common for the dietician to increase the recording as it was to decrease the estimated EI from that meal. In our earlier pilot study, we found a similar proportion of meals (23/78 meals in the pilot vs 69/210 meals in the present study) had been adjusted, but the regular FR overestimated the energy intake by 169 kJ per day, compared to the image-assisted method. Also, in the pilot, the majority of the adjusted meals (22/23) were overestimations. In the present study, underestimations were equally common to overestimations. A possible explanation to this discrepancy may be that the participants in the present study, being part of a large trial, were more experienced in completing food recordings, this being their third in six months compared to the participants in the pilot. We speculate that greater experience explains some of the lower bias in the regular FRs, compared to the image-assisted method found in our study [28
]. We do not know to what extent the parents in the present study used the images to corroborate their recordings, but we did notice that the parents were skilled at allowing for leftovers in the FRs, which is indicated by the fact that half of the meals where there were leftovers had been already adjusted by the parent. From the present study, we have no information on the amount of leftovers or spillage from the eating episodes where image assistance was not used. We can only speculate that the overestimation of EI compared to ME can be found in unaccounted leftovers, from for example breakfast and snacks. From the FRs, we know that these meals contained large amounts of energy dense foods, such as porridge and milk cereal drinks. Another possible error could have been the preparation of the porridge and milk cereals, where adding too much water could have made the meals more diluted, reducing the actual EI compared to the recorded EI.
The overall energy and macronutrient intake, and its variations, were similar to other studies in the same age group [30
]. Also, the parents adhered well to submitting the FRs and images, with less than 5% of the main meals having missing images. Breast milk intakes were estimated from feeding episodes and not by direct observation, i.e., test weighing. On average, breast milk contributed to <6% of the mean total daily energy intake. The energy and nutrient content of the products used in the present study (porridge, formula, milk cereal drink, baby food in glass jars) were based on high quality analyzed data supplied from Semper AB.
A strength of the study was that all parents used the same reference marker [25
], i.e., the plate estimated the true area of the food portions, and we omitted images of meals, from which other types of plates were used. Also, the same dietician managed all FRs, calculated the dietary intake, and assessed the images, but was blinded as to the outcome, and did not participate in the final analysis, i.e., when the regular FRs was compared to the image-assisted FRs. In the DLW analyses, we used the more accurate FQ [19
], instead of the generic RQ suggested by Schoeller et al. [17
]. The proportion of parents with university education was higher than people of the same age and gender in the general, Swedish population, and higher than that reported in a recent iron supplementation study in the same geographical area [33
]. Despite this, we believe that it is possible to generalize the results to other populations as well.
A limitation in the study was that we did not use the image-assisted method for all meals and snacks. Our hypothesis was to focus on the main meals, i.e., lunch and dinner, which are more complex and diverse in terms of ingredients and nutrient value, and to leave breakfast and snacks, which in this age are less diverse. It is likely that unaccounted leftovers and spillage from energy- and nutrient-dense foods, such as milk cereal drinks or baby porridge, could have been identified through image-assistance. However, we believe that the task of providing before and after photos of every meal and snack, considering the frequency of feedings, including night meals in this age group would have been almost insurmountable for both the parents and researchers.
Variations among different mobile cameras, ambient light conditions, etc., may have contributed to some of the subjectivity of the image analysis [34
]. All images were taken at 90° to the plate, which is the most favorable for capturing which ingredients the meal contained. However, to optimize volume calculations, another photo at 45° would have been preferable [9
]. To improve the quality of the images, and to aid in the volume estimation, some kind of mobile phone application would have been desirable, but to the best of our knowledge no such product validated for use in infants is commercially available [34