Population-based surveys are essential to guiding nutrition-oriented public health policies (1). When collected regularly, nutrition surveys provide valuable data regarding trends in a population's dietary intakes, nutrient intake adequacy, food insecurity, and impact of diet-related policies, which can then inform the development of dietary recommendations (2). In Canada, the most recent methodology in national nutrition survey employed face-to-face interviewer-administered 24-h recall assisted by computer (hereafter referred to as TRAD) (3). Dietary intake is assessed using a recall instrument because it is considered less biased than other instruments such as food frequency questionnaires (FFQ) or dietary screeners (4–6). In the case of the FFQ, respondents report the frequency of consumption of a finite list of foods and over periods generally ranging from 1 month to 1 year. Accordingly, intakes assessed using a FFQ depend on one's perceived usual intakes and do not reflect foods actually consumed on a particular day (6). While using a similar approach, a screener focuses on shorter list of foods [e.g., vegetables and whole fruits (6)]. Although the use of a 24-h recall is known to produce higher quality intake data, the training of the interviewers, the impromptu visits to conduct interviews and the review of the 24-h recall coding are technical and monetary barriers to frequent nutrition survey data collection using TRAD.

Technological advances have prompted the development and the broader use of web-based dietary assessment instruments (7–9). Collecting dietary intake data using a self-administered web-based 24-h recall is efficient as it greatly reduces burden on the research team. It also standardizes questions and offers the respondent the ability to complete the 24-h recall at the time of the day and place deemed appropriate (10–12). Food and nutrient intake estimates derived from an interviewer-administered 24-h recall and web-based 24-h recall instruments are generally consistent (13–15). However, data suggest that digital pictures of web-based 24-h recall may attenuate misestimation of portion size compared with food booklet pictures of an interviewer-administered 24-h recall (16). It is stressed that meticulous dietary assessment interviews conducted by expert teams in research settings often may not reflect the methodology employed in a large national survey. For example, interviewers involved in the Canadian Community Health Survey 2015 (CCHS 2015) to conduct the dietary assessment using TRAD received 2.5 days of training, but were not required to have a formal degree in nutrition (3). The consistency of dietary intake data obtained in the CCHS 2015 with a more contemporary dietary assessment methodology is not well-documented to date. Thus, comparing data from a web-based 24-h recall with data from an interviewer-administered 24-h recall as used in the national survey is essential to best inform on future nutrition survey methodology.

Our team has developed and validated a web-based 24-h recall instrument, the R24W (9, 17–19), which we used subsequently to assess the diet of an age- and sex-representative sample of 1147 French-speaking adults from five regions in the Province of Quebec, Canada from 2015 to 2017 as part of the PREDISE (PRÉDicteurs Individuels, Sociaux et Environnementaux) study (20, 21). Because CCHS and the PREDISE study were conducted within the same period in similar population-based samples, we have a unique opportunity to compare the consistency of dietary intake data obtained using TRAD and a web-based 24-h recall. The overall aim of the present study was therefore to compare distribution of food and nutrient intake estimates obtained using a TRAD and the R24W in similar population-based samples from the Province of Quebec. The primary outcome of this analysis was the difference between mean food intake estimates (vegetable and fruit, grain products, milk and alternatives, meat and alternatives, and “other foods”) between TRAD and the R24W. As exploratory outcomes, we investigated mean differences for nutrients of public health interest, namely, energy, saturated fats, sodium, sugar, potassium, and fibers. We hypothesized that the dietary assessment method influences dietary intake estimates.

Using population-based sample survey studies, we have compared dietary intake estimates obtained using a self-administered web-based 24-h recall (the R24W) to an interviewer-administered 24-h recall (TRAD). Our comparison revealed that intake estimates of most food groups and nutrients were higher when obtained from the R24W than from TRAD. In particular, data revealed that self-reported intake of “other foods,” a category that includes mostly low nutritive value foods, is higher with the R24W than with TRAD. As a result, total energy intake measured by R24W was less likely to be underreported than energy intake measured by TRAD, even more so among women than among men. These data suggest that the use of a web-based 24-h dietary recall may attenuate the prevalence of energy underreporting in the context of nutrition surveys, a major critic of current dietary assessment methods relying on self-reported intakes.

In the present study, French-speaking adults from Québec reported consuming more food items with R24W than with TRAD, which has contributed to the systematically higher food and nutrient intake estimates with the R24W. Of note, the underestimation of low nutritive value foods on TRAD compared with the R24W accounts for ~50% of the difference in total energy intakes between the two instruments. Thus, the greater prevalence of underreporting on TRAD is partially explained by specific underestimation of low nutritive value foods intake. Social desirability bias is known to influence self-reported dietary intakes (34). Indeed, self-reporting dietary intakes in person may be intimidating when conducted in-person (35) and less so when using a web interface, therefore yielding intake data that are less likely to be biased and therefore more likely to be accurate. Differences observed in food and nutrient intakes between the two instruments may be due, in part, to differences in portion size estimation. Indeed, close to 1,500 digital pictures of serving sizes are available within the R24W (9), while a total of 30 pictures are available in the entire CCHS' food booklet (3). Kirkpatrick et al. (16) compared dietary intake data obtained from a web-based 24-h recall (the ASA24) with data from an interviewer-administered 24-h recall in a non-population-based sample. They provided evidence that portion size estimation may be more accurate using numerous digital pictures. In addition, the standardized interface within web-based applications may limit variability of the 24-h recall interviews attributable to the interviewers per se (36). In sum, assessing dietary intake with a web-based 24-h recall is likely to yield different dietary intake estimates compared with an interviewer-administered 24-h recall, possibly through a lower degree of desirability bias, standardized questions, and more accurate portion size estimation.

Andreeva et al. compared dietary intake data obtained using a web-based 24-h recall (the NutriNet-Santé web dietary record) to data from a nationally representative sample obtained using interviewer-administered 24-h recall (37). Samples were matched for several sociodemographic characteristics, thus allowing comparison of dietary intake data obtained by the two instruments. Mean dietary intake estimates obtained through the web-based and interview-administered 24-h recalls were similar for most foods and macronutrients. However, intake estimates of sweetened non-alcoholic beverages were lower with the web-based 24-h recall than with the interviewer-administered 24-h recall (−48% in men and −44% in women). Similar proportions of respondents (12–15%) were identified as potential underreporters when using either the interviewer-administered or web-based 24-h recalls. These data are at odds with data from the present study, where important differences were noted between the two measurement approaches. In the study by Andreeva et al., respondents knew beforehand when the web-based 24-h recall would have to be completed, thus potentially triggering reactivity bias (38), which was not the case in the present study. Plus, the interview-administered 24-h recalls were conducted by phone in Andreeva et al., while the majority of respondents in CCHS 2015 (TRAD) were interviewed in person. In-person interviews may have increased the risk of selective underreporting due to more impactful social desirability biases compared with a phone interview (35). This, in turn, may explain the more pronounced differences between web-based and interview-based instruments when in-person interviews are conducted in population-based surveys. Studies to date that compared telephone with in-person interviews have reported inconsistent results regarding the potential impact of social desirability biases on total energy intakes (39–41). Additional studies are therefore needed to determine the extent to which the use of web-based dietary assessment tools alleviates some of the social desirability biases seen with interviewer-based methods.

Other studies have employed a more controlled design using repeated measurements and directly compared dietary intakes obtained using interviewer-administered and web-based 24-h recalls. In a study comparing an interviewer-based 24-h recall and the web-based ASA24 relative to known dietary intakes, Kirkpatrick et al. found that while dietary intakes assessed using both instruments were consistent, the interviewer-administered 24-h recall performed somewhat better than the web-based instrument relative to true intakes (14). Thompson et al. (15) found that intake of energy and the majority of foods and nutrient measured using the web-based ASA24 were similar to values obtained with interviewer-administered 24-h recall. In the same study, only fruits (lower intake using the ASA24) and meat (higher intake using the ASAS24) were considered non-equivalent to data from the interviewer-administered 24-h recall. Timon et al. (42) found that energy, saturated fats, sugars, potassium and sodium were all lower using the web-based Foodbook24 instrument, compared with an interviewer-administered 24-h recall. Intake of most food groups were also lower using the FoodBook24 compared with the interviewer-administered 24-h recall (42). Overall, within-individual comparisons of intake data obtained using an interviewer-administered 24-h recall and a web-based 24-h recall revealed either a similar performance of both instruments (15) or a better performance of the interviewer-administered 24-h recall (14, 42). These observations are also inconsistent with findings from the present study. It is stressed that these studies, which are experimental in nature, have been conducted by expert teams, thus potentially yielding data of greater quality overall compared to data obtained by non-expert interviewers in national surveys. It is also possible that each web-based instrument may have intrinsic features that differentially influence the quality of the measurement. In that regard, Timon et al. (43) reviewed numerous web-based 24-h recall instruments, highlighting that these instruments widely differ in terms of the number of foods considered, the portion size estimation methods and the use of prompts. These differences between instruments must be kept in mind when comparing results among different studies. Finally, only the comparison of self-reported dietary intakes against gold standard method such as doubly labeled water can determine which instrument produces the most accurate estimates. Metabolomics may also help improve current dietary intake assessment based on self-reported approaches (44).

This study has several strengths, including the carefully matched samples between CCHS 2015 (TRAD) and PREDISE (R24W), temporal proximity of both study, reasonable sample size, and the use of the same reference food and nutrient tables to assess food group and nutrient intakes. In this context, we believe that differences between dietary intake estimates assessed using a web-based and an interview-based 24-h recall are plausibly related to intrinsic characteristics of each instrument. These characteristics include, among others, the mode of administration (i.e., web vs. interviewer-administered) and the portion size estimation support (i.e., digital pictures vs. food booklet). Limitations must also be discussed. First, this comparison of dietary assessment methodology is opportunistic and, consequently, the comparison of dietary intake estimates may also reflect true differences in intake between samples and different degree of selection bias. In particular, the PREDISE sample (R24W) included slightly fewer smokers and a higher proportion of dietary supplement users, suggesting a tendency toward healthier lifestyle habits. Second, missing data for height and weight led to a greater proportion of participants with “imputed” energy intake reporting status in the CCHS 2015 sample and this may have also biased our assessment of the plausibility of self-reported energy intakes in this sample. In particular, missing data regarding reporting status may be associated with body weight, potentially contributing to bias if the imputation model does not fully capture the relationship between body weight and reporting status based on the data available. The wide 95% confidence intervals in CCHS 2015 are indeed compatible with a low to moderate prevalence of underreporting (12–33%). Third, the lack of an objective measure of physical activity limited our ability to assess the proportion of plausible reporters of energy intake, but this was the case in both samples.

In conclusion, this comparison of dietary intake estimates obtained in matched population-based samples of adults from Québec revealed systematically higher intake of foods and nutrient when using a web-based 24-h recall compared to using a traditional, interview-administered 24-h recall. These differences, which are not trivial, are plausibly related to the dietary assessment methodology considering the similar population examined and the temporal proximity of both dietary assessments. Such differences between instruments need to be considered along with other characteristics such as cost, efficiency of administration, and burden of data management when determining which dietary assessment instruments is best suited for use in surveillance and population-based surveys.

For the PREDISE study, the raw data supporting the conclusions of this article will be made available by the authors, without undue reservation. For the Canadian Community Health Survey 2015—Nutrition (CCHS), only the summary data supporting the conclusions of this article will be made available by the authors, without undue reservation. The raw CCHS data used in this article is not readily available because the confidential microdata files can only be accessed through Research Data Centers in Canada. Instructions to access the raw data can be found at https://www.statcan.gc.ca/eng/microdata/data-centres/access.

The studies involving human participants were reviewed and approved by 1.1) PREDISE STUDY (online study protocol): Laval University Ethics Committee (# 2014–271); 1.2) PREDISE STUDY (clinical assessment study protocol): ethics board of each participating institution (Laval University, #2014–271; Montreal Clinical Research Institute, #2015–02; Université du Québec à Trois-Rivières, #15–209- 07.13; Center Hospitalier Universitaire de Sherbrooke and ECOGENE-21 Biocluster, #MP-31-2015-997); 2) CANADIAN COMMUNITY HEALTH SURVEY (CCHS): The ethical approval for CCHS 2015 Nutrition conducted by Statistics Canada is based on the authority of the Statistics Act of Canada. The patients/participants provided their written informed consent to participate in this study.

DB and BL designed the study. DB performed statistical analyses and wrote the first draft of the manuscript. SL, BL, JR, and CL have developed and validated the R24W. All authors critically reviewed the manuscript, provided final approval of the submitted manuscript, had full access to all of the data in the study, take responsibility for the integrity of the data, the accuracy of the data in the analysis, affirm that the article is an honest, accurate, and transparent account of the study being reported and that no important aspects of the study have been omitted.

DB has received a speaking honorarium from the Dairy Farmers of Canada in 2018. DB is the recipient of a doctoral training award from the Fonds de recherche du Québec–Santé. DB has been a casual employee at the Office of Nutrition Policy and Promotion (Health Canada) in 2019 and 2020. SL, BL, JR, and CL have developed and validated the R24W, which is currently used for research purposes exclusively.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.