In which situations should diet be assessed in clinical research? Which considerations should drive the selection of a dietary assessment tool (DAT)? This paper addresses these two questions, emphasizing that one should consider not only the research objectives and study setting, but also the opportunities and limitations of available methods for acquiring and analyzing dietary data.

In epidemiological and clinical observational studies, dietary parameters are adequately considered when information on usual dietary intake is needed at a population level [1] or when diet-disease relationships are being explored [2]. In these settings, evaluating dietary intake over time has proved efficient for investigating diet-disease associations at the population level, especially in chronic conditions [2]. Tracking participant lifestyle was needed to understand the role of unhealthy food and diet in the development of overweight, obesity, type 2 diabetes [3], cardiovascular diseases [3,4], and cancer [5]. However, despite the availability of scientifically validated methods to assess dietary intake and usual intake, their use in establishing diet-disease relationships remains controversial [6,7].

In interventional settings, dietary parameters need to be considered for efficacy assessments or to control confounding effects, irrespective of the nature (dietary, drug, or other) of the intervention [8]. However, background diet is, to our knowledge, not systematically assessed in drug development studies despite evidence of the presence of food-drug interactions.

Food-drug interaction is also an important factor to consider during clinical development, as some chemical compounds in foods can potentially affect the pharmacokinetic, pharmacodynamic, or metabolic pathways of some drugs [9]. For example, it is known that leafy green vegetables such as kale and spinach, which contain high levels of vitamin K, can reduce the effectiveness of some oral anticoagulants, such as warfarin [10]. In parallel, we must not neglect how the side effects of medicinal products can influence appetite [11].

Selecting a DAT adapted to a research question is usually a matter of compromise. Current methods to assess diet are available in various formats (paper or electronic), to obtain data on dietary intake (prospectively or retrospectively) or usual intake [12]. In a comprehensive resource guide issued by the Food and Agriculture Organization of the United Nations [13] and as explained by other authors [14], various available DATs can be distinguished based on technology (including data acquisition):

Collecting dietary data is, however, not always needed for primary study analyses, as one may only be interested in monitoring the background diet or compliance with dietary restrictions while focusing on many other clinical and nonclinical parameters (Figure 1).

Because collecting dietary information usually requires substantial involvement from clinical study participants, selecting an appropriate DAT represents a trade-off between participant burden and the volume or accuracy of the collected information. Compromises may include collecting dietary intake data at specific time points rather than tracking each eating occasion, and choosing a reasonable frequency for assessing food intake throughout the research studies.

However, limiting the dietary assessment to habitual food intake (evaluated using food frequency questionnaires) is not relevant for every study, especially when associations between dietary and biological parameters are investigated, as outlined in Figure 2.

Instead of evaluating habitual food intake, daily tracking of each eating occasion may be needed. One example is the daily variation of gut microbiota composition, known to be related to food choice [15], requiring, when investigated, exhaustive and daily tracking of a participant’s dietary intake. Handling temporality, therefore, represents a major challenge when selecting a relevant DAT, in case there is important intrasubject variability of parameters over time. We believe that a focus is needed for identifying or developing solutions that can effectively track participants’ background diet throughout their participation in research studies (observational or interventional), irrespective of the intervention (drug, food, or dietary) while limiting the associated participant burden. DATs based on new technologies offer opportunities to facilitate traditional food intake measures, and mobile apps have demonstrated being adequate for both participants and researchers. For participants, mobile apps represent an acquisition method that is acceptable, accessible, ubiquitous, and one that can be used by participants alone [16-18]. For the study sponsors or researchers, these apps offer other opportunities in terms of cost-effectiveness, reducing time between the collection and reporting of dietary data, and improving data quality with an accurate, comprehensive, and relevant food composition data set [14].

Mobile apps, therefore, represent a good compromise for researchers (investigators and sponsors) when the collection of dietary data is needed.

When navigating the astronomical number of mobile diet apps (approximately 30,000 diet and physical activity mobile apps are currently available on Google Play and the App Store), selecting a DAT that fits with research purposes remains a key challenge. This selection should be based on multiple parameters, including the suitability of user experience and interface, the reliability of the food composition database (FCD), the relevance of the data acquisition method (identification and quantification of foods), and the quality of nutritional outputs.

To compare these mobile diet apps in terms of reliability, quality, and effectiveness, we cannot consider the number of downloads as a reliable evaluation parameter, as the most frequently downloaded apps are not necessarily the most usable or efficient in medical research [19]. Neither can we assess these tools based on user ratings (ie, stars), which are known to be biased [20]. Currently, no official standards exist for evaluating health-related mobile apps, making the evaluation of the strengths and weaknesses of these apps complex for researchers when choosing a relevant tool [21].

High-quality reviews and meta-analyses are few, and there is limited comparability between the evaluation methods used. Some of the reviews evaluate functionality, information quality, esthetics, and engagement [22]; others evaluate the accountability, usability, scientific coverage, and technology-enhanced features based on the opinions of nutrition experts who had tested different apps for 5 days [23]. In parallel, other experts have developed their own scales for a qualitative assessment of mobile health apps, including the following:

To our knowledge, only 3 studies have compared the nutritional values provided by different mobile diet apps based on quantitative parameters. One research team evaluated the accuracy of 7 diet apps in providing nutrient values. In their study, the nutrient labels of 100 food products available in the Netherlands were compared to the nutrient values provided in the nutrition facts in each app. The researchers concluded that the accuracy of nutrient values varied enormously between these apps, and energy was the most reliable value [26]. Another research team compared the nutrient intake calculations of 5 consumer-grade mobile diet apps to the Nutrition Data System for Research dietary analysis program, concluding that most of the nutritional values (except energy and some macronutrients) provided in these apps are generally underestimated, and that some apps are more accurate than others [27].

More recently, another team evaluated energy, macronutrient, and available micronutrient values provided by 5 popular nutrition apps against a UK reference method (Dietplan6 [28]). The authors concluded that the values of energy, carbohydrates, saturated fats, total fat, and fiber reported by these apps (except for 1 app) were rather reliable, while the values of protein, sodium, and micronutrients were inconsistent and less reliable [29].

In summary, we can surmise that researchers will usually favor using academic mobile apps, while end users may be more attracted to the features offered by consumer-grade mobile apps. Based on our experience and the cited opinions of experts in the field of nutrition, we understand that currently, there is no one-size-fits-all solution for tracking dietary intake that can be used in every type of medical research, as each available tool has its limitations [35,50] due to language availability, validation in specific populations, and the reliability of FCDs. The ideal option would be to develop a mobile app for tracking the diet that has the strengths of both types of apps: an efficient, user-friendly interface and experience, coupled with an FCD that is as rich as it is reliable. In the absence of such a tool, or in developing one soon, we must select the most appropriate one for each type of study according to its objectives. The landmark DIET@NET (DIETary Assessment Tool NETwork) suggests considering 5 steps before choosing the appropriate type of DAT, and the first step is, of course, to define the research question well [50].

If the daily use of consumer apps by millions of people worldwide can generate big data on food consumption, this information can be used by researchers (while complying with data privacy requirements) to study health, diet, and dietary behavior parameters. This can be done by considering cultural aspects of populations at different levels (country, household, or individual) while integrating potential key cofactors such as physical activity level, sleep, weight, or location data, which can often be collected simultaneously. It has also been highlighted that the most frequently downloaded consumer-grade apps lack features relating to emotions, even though emotions are known to be associated with diet [40]. Using these diet apps as a reliable food-tracking tool in clinical trials or research, in general, is a major challenge, considering their limitations; despite the popularity of these apps, the lack of evidence and scientific validation of their use remains the major issue [51]. Little information is available on the quality of these mobile apps, other than what can be gathered from users’ ratings published on mobile app stores. The lack of reliability of the data these apps provide, especially in the FCD, is a real barrier for researchers who want to utilize the user data. It should also be noted that the legal and ethical concerns related to user data for consumer-grade mobile health apps are well documented in the literature [52-54]. A recent review showed that most mobile diet apps do not provide terms of use or privacy policy documents [42].

The use of consumer diet apps in studies should be done with caution: given the nature and suboptimal quality of their FCDs, their use should be limited to the tracking of some macroelements such as energy intake. Energy intake is considered the most consistently and accurately value reported by these apps compared to other macro- and micronutrients [23,26], which are generally underestimated [27,29,55]. Recent studies have also shown that different consumer apps, such as Samsung Health, MyFitnessPal, and FatSecret, provide an acceptable estimate of energy, carbohydrate, and fat intake [29], and that MyFitnessPal offers a good relative validity for energy and fiber tracking [55]. Many dieticians (in the United Kingdom, New Zealand, and Australia) use nutrition apps in their practice [56]. However, for many experts in the field of nutrition, mobile apps designed for consumer awareness are an unreliable source of nutritional values due to the crowdsourcing nature of the FCD. This issue is leading to a lack of interest in using these new tools for nutritional and clinical studies.

Our opinion is that the potential of mobile apps designed for consumer use has not been sufficiently evaluated, especially in clinical settings. Despite the limitations that have been outlined above, we believe that the data collected by these apps represent a source of information we cannot ignore, as it may be sufficient in some epidemiological or clinical studies, when the assessment of diet is not the primary objective of a research study. Potential uses of these tools include segmenting participant populations into subgroups according to usual dietary intake or associating specific meal patterns with biological parameters during a short time frame. However, in our view, further work is needed to assess this potential. We should also consider the potential of other features that are now included in several mobile apps, such as automated, image-based recognition. This feature, while promising, is only in an early stage of development, and there is still room for improvement until it can be used regularly by clinical study participants. However, when relevant, analytical methods offered by artificial intelligence may soon be used to obtain both qualitative (ie, composition of meals) and quantitative (ie, portion sizes) outputs. We believe that such features may one day be the bridge that accommodates participants and researchers, filling the gap between academic and consumer-grade apps once they are available in both types of solutions.