Moving to self-administered dietary assessment in national food consumption surveys

The ever-growing findings from dietary studies have confirmed the important role of dietary intake in the development of certain non-communicable diseases (NCDs). In order to take a better control of the NCD progression in a population, primary prevention measures such as dietary guidelines and food policies are updated constantly according to the most recent scientific evidence and observed dietary patterns of a specific population. To obtain these dietary patterns, dietary intake at a national level is commonly monitored by governments in the form of food consumption surveys. However, assessing the dietary intake of a large population has been a challenging task throughout the years. With highly diversified food consumption practices and varied individual capabilities of reporting dietary intake, vast inputs (financially, physically) are required for collecting detailed dietary information. Hence, this thesis focuses on improving the methodology of dietary data collections, specifically for the Dutch National Food Consumption Surveys (DNFCS). The investigations proceeded in two parallel paths. Firstly, on how to remove burdensome procedures from current survey collection methods. Secondly, by learning from other studies and tools where dietary assessment techniques were investigated.

The current dietary assessment method in the DNFCS is the interviewer-administered 24-hour recalls (24HRs) guided by the computer software called GloboDiet. Each food item goes through a round of detailed questions extracting relevant information (also called facets). This detail acquiring step has been the most time-consuming part of the interview. Besides, undesirable answers are likely to be obtained due to limited knowledge of the respondents. In order to enhance the interview efficiency while minimising the impact on the survey results, the importance of facets in terms of predicting the nutrient outcome was determined using a prediction model called random forest. As a result, 35% of the total facets were deemed unimportant and could be omitted; this would resulted in a change of 3.7% of the foods linked to the NEVO (Dutch Food Composition Database). The majority (79.4%) of the differences between percentile estimates of the population nutrient intake distributions before and after facet deletion ranged from 0% to 1%. The reduction of facets was estimated to save 637 hours for data collection and 442 hours for the data handling for a survey conducted on 3819 participants.

Another complicated task in the current GloboDiet 24HR interview is the recipe pathway. Typically, mixed dishes are firstly identified with a standard recipe, then the ingredient composition and amounts are adjusted according to the available information on the real dish eaten. A replacement of the burdensome recipe modifications with the unchanged standard recipes has been simulated in this study. Comparing the simulated results and the original dataset, the average of the absolute percentage difference for the population mean intakes was 1.6% across all food groups and 0.6% for nutrients. The resulting small difference was mainly owning to the small proportion of energy intake consumed through mixed meals (10%) among the Dutch population according to the survey. A list with more realistic standard recipes would enable the use of a simple recipe function in a self-administered 24hR or food diary.

With a fast-evolving smartphone industry and an increased awareness of diet-health relationships among the general public, large varieties of dietary recording apps have been developed and were made available for download from app stores. Since most of the apps were designed as self-administered instruments, their functionality might be a useful example for developing self-administered tools for large-scale nutritional monitoring or research. Out of 57 popular food record apps, 12 apps having a recipe function were scored according to pre-defined criteria. The variation in food composition databases (FCDs) underlying each app contributed the most to the differences larger than 5% of Daily Reference Intake (DRI) in 49% of the micronutrients and 20% of the energy and macronutrients. Applying retention factors decreased the nutrient content for specific heat-sensitive vitamins such as B6, B12, and folate up to 45%.

Different from commercial apps, that have been mostly compared with each other, research-based apps have been described more in detail, reporting on their development, validity compared with a reference method, and usability or feasibility of applications in a sample population. A systematic review and meta-analysis on validation studies therefore provides insights into the general applicability and potentially the common flaws of apps. From an online search of literatures from 2013 to 2019, 14 studies were found that have validated food record apps in real-life settings. The pooled mean difference between the apps and the reference methods across studies showed a general underestimation of energy intake (-202 kcal/day) by using the apps. Studies with different FCDs for each method had the largest mean differences. The sources of variation were traced for studies that compared food group/ food item differences. A variation in study designs has been found among studies, which impedes the comparisons across studies, (e.g. use of energy-adjusted/log-transformed values).

In conclusion, this thesis has shown that a simplification of current interviewer-administered 24HR is promising, which implies that the simplified functions might work equally well in a more cost-effective self-administered method. The advanced features and prevalence of use have made smartphones the optimal platform for monitoring dietary intakes at a population level. Still, a larger underestimation of energy intake using self-administered methods is expected compared to interviewer-administered methods, which implies the need for more guidance compared to using commercial apps, and careful interpretation of results. Moving to a self-administered method is a big step for NFCS, which requires careful considerations and large inputs during the development and validation phase. However, the lower costs and efforts required by using self-administered method could highly likely to counterbalance the initial investment, in the meanwhile, providing participants with a more flexible platform for dietary recording.