Validation of A Newly Automated Web-Based 24-Hour

Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

Lafrenière et al.

BMC Nutrition (2017) 3:34


DOI 10.1186/s40795-017-0153-3

RESEARCH ARTICLE Open Access

Validation of a newly automated web-


based 24-hour dietary recall using fully
controlled feeding studies
Jacynthe Lafrenière1,2, Benoît Lamarche1,2, Catherine Laramée2, Julie Robitaille1,2 and Simone Lemieux1,2*

Abstract
Background: Assessment of food intake is a cornerstone of nutritional research. However, the use of minimally
validated dietary assessment methods is common and can generate misleading results. Thus, there is a need for
valid, precise and cost-effective dietary assessment tools to be used in large cohort studies.
The objective is to validate a newly developed automated self-administered web-based 24-h dietary recall (R24W),
within a population of adults taking part in fully controlled feeding studies.
Methods: Sixty two adults completed the R24W twice while being fed by our research team. Actual intakes were
precisely known, thereby allowing the analysis of the proportion of adequately self-reported items. Association
between offered and reported portion sizes was assessed with correlation coefficients and agreement with the
kappa score while systematics biases were illustrated with Bland-Altman Plot.
Results: Participants received an average of 16 food items per testing day. They reported 89.3% of the items they
received. The more frequently omitted food categories were vegetables included in recipes (40.0%) as well as side
vegetables (20.0%) and represented less than 5% of the actual daily energy intake. Offered and self-reported portion
sizes were significantly correlated (r = 0.80 P < 0.001) and demonstrated a strong agreement as assessed by the kappa
score of 0.62. Reported portion sizes for individual food items were on average 3.2 g over the offered portion sizes.
Portions of 100 g and above were on average underestimated by 2.4% (r = 0.68 P < 0.01; kappa score = 0.50) while
small portions (less than 100 g) were overestimated by 17.1% (r = 0.46 P < 0.01; kappa score = 0.43). A nonsignificant
underestimation (−13.9 kcal ± 646.3 kcal; P = 0.83) of energy intake was noted.
Conclusion: R24W performed well as participants were able to report the great majority of items they ate and selected
portion size strongly related to the one they received. This suggests that food items are easily to find within the R24W
and images of portion sizes used in this dietary assessment tool are adequate and can provide valid food intake
evaluation.
Keywords: Food assessment, Dietary recall, Portion size estimation, Validation, Feeding studies

Background those errors may be significantly reduced by improving


High quality nutritional research hinges on valid assess- data collection techniques and by selecting tools adapted to
ment of food intake. However, it remains a real challenge the studied population. Furthermore, it is essential to valid-
to adequately measure food intake. Because of wide within ate a new dietary assessment tool before its first use [3].
subject variation, self-reported tools have some degree of Self-reported food assessment tools are often associ-
random errors [1] often associated with incoherent ated with high rates of misreporting leading to under-
research results [2]. Recent researches demonstrated that estimation of energy and nutrient intakes compared
to objective measurements [4]. Misreporting can be
explained in part by undereating and in part by
* Correspondence: [email protected]
1
School of Nutrition, Laval University, Québec, Qc, Canada
under-recording. Indeed, subjects tend to reduce or
2
Institute of Nutrition and Functional Foods, Laval University, Québec, Qc, change their food intake when they have to report it.
Canada

© The Author(s). 2017 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and
reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to
the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver
(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Lafrenière et al. BMC Nutrition (2017) 3:34 Page 2 of 10

This has been referred to as the reactivity bias [5]. In of fully controlled feeding studies. In these projects
addition, studies have shown that subjects frequently intended to evaluate the effects of specific nutritional
fail to remember all the items they ate and they have manipulation, participants receive all their meals for the
trouble estimating the exact portion sizes consumed duration of the study. The specific composition and
[6–8]. These errors could be attributed to memory or weight of each food item consumed is therefore known
social desirability bias [4, 9]. and can be compared to the recall filled by the partici-
Automated self-administered 24-h recalls received atten- pants afterwards.
tion lately and are increasingly used because there are The R24W is a newly developed web-based, self-
convenient and cost-effective [10]. For cohort studies, auto- administered and fully automated 24-h recall [23]. It is
mated 24-h dietary recall is becoming the tool of choice the first French-language web-based automated 24-h
instead of food frequency questionnaire because of its su- recall developed to assess dietary intake in the French-
perior precision and accuracy [11, 12]. Furthermore, it pre- Canadian population. The aim of this study was to val-
sents characteristics that can help reducing the above- idate the R24W in a context of fully controlled feeding
mentioned biases. First, as participants generally filled their studies. More precisely, we wanted to evaluate ad-
recalls on unannounced days, this limits the reactivity bias. equate reporting of food items and portion size evalu-
Second, those recalls are completed by the respondent, out- ation. We hypothesize that the majority of offered food
side of a laboratory setting, thus in a neutral environment items are adequately reported using the R24W. We
[13]. This reduces the social desirability bias as compared also hypothesize that there is a portion size estimation
with a face to face administration of recalls [14, 15]. Third, error of less than 10%. Furthermore, as other authors
the inclusion of memory cues in the recall can attenuate reported that adults tend to underestimate large por-
memory bias which can therefore contribute to reduce tion sizes compared with smaller ones [24, 25] we de-
underreporting. For example, with an approach like the cided to test specifically error rate in small and larger
USDA automated multiple pass method, items are portions and we hypothesize that, because of the large
reviewed to make sure that nothing has been forgotten, distribution of portion sizes illustrated in the R24W,
context of the meal is accentuated because it helps to the difference between them is not significant.
remember details about the food consumed and many
questions are asked about frequently forgotten food items
[16]. In addition, images of portion sizes can improve Methods
estimation accuracy by up to 60% [17, 18]. Finally, presen- Participants
tation of simultaneous different portion size options can This analysis was conducted on 33 men and 29 women
reduce error rates compared to presentation of only one already enrolled in three fully controlled feeding studies
option [7]. currently conducted in our research institute. To be in-
Some studies have shown that, compared with the trad- cluded in these studies, they had to be non-smoking
itional interview, web-based 24-h recall generates equiva- men or women aged between 18 and 75 years old with
lent results with a reduced precision for some nutrients stable weight. Women should not be pregnant or lactat-
[6, 19, 20]. However, the preference for the web-based ing. Each participant had to be free from cardiovascular
version was highlighted by many researchers. Indeed, or endocrine diseases and should not have a food allergy
Thompson et al. [19], showed that using a web-based 24-h or aversion to any food item offered in the feeding
recall reduced attrition rate compared to interview admin- protocol. These studies received the approval of the La-
istered 24-h recall. Furthermore, with the web-based val University Ethics Committee and participants pro-
approach, the completion time is reduced and the coding vided written informed consent prior to taking part in
step is automated thereby saving a significant amount of the study. The analysis performed in this article was dir-
time [21]. The main concern remaining is the accessibil- ectly presented in one of the three consent forms (the
ity as internet connexion is not yet equivalently spread latest study) and was proposed to the participants
causing an underrepresentation of older adults with already included in the two other studies in an adden-
lower incomes (in 2012, 28% of them had internet dum to their initial consent form. Therefore, all partici-
access compared with 95% of younger adults with pants gave their informed consent before completing
higher incomes [22]). their first 24-h recall knowing that it was part of a valid-
It has to be emphasized that when a new dietary as- ation process.
sessment tool is developed, a rigorous validation process These clinical trials were registered at http://www.cli
needs to be performed before it can be used in cohort nicaltrials.gov as NCT02763930, NCT02106208 &
studies. One strategy that can be used when validating NCT02029833. Participants followed the initial research
dietary assessment tools is to compare self-reported food protocol but in addition to this, we asked them to fill
intake to the actual food intake consumed in the context the R24W twice.

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Lafrenière et al. BMC Nutrition (2017) 3:34 Page 3 of 10

The automated web-based automated 24-h recall (R24W) or gained greater than 1 kg and maintained that body
Details on how the R24W was developed have been pub- weight for at least 3 days. In order to standardize the
lished elsewhere [23]. Briefly, R24W was developed in testing between studies, the R24W was completed in the
French language and was inspired by the AMPM of the first two weeks of one of the feeding phases for each
USDA [16]. However, as opposed to AMPM, R24W is project. Menus were composed of three meals and a
using a meal-based approach in the first step. An unlim- snack per day. Except for a few exceptions, lunch (40% of
ited number of meals or snacks per 24-h period can be daily calories) was consumed in the lab facility while din-
added by the respondent. R24W allows automatic calcu- ner, breakfast and snack were packed in a cooler and
lation of different diet quality scores in addition to consumed outside the clinical facility. Items were all la-
energy and nutrient intakes. The application was pro- belled and participants received a checklist to remind
grammed in such a way that the days for which the re- them to eat the entire menu in order to enhance compli-
call has to be completed are randomly generated with ance (see Additional file 1). On this checklist, the general
the possibility of using specific criteria (e.g., proportion name of the meals and side dishes were given but the list
between weekdays and weekend days). For this valid- of ingredients for mixed dishes was not included. This
ation study, the same two days in each study were used gave us the opportunity to assess how well participants
for recalls. The application also allows the automatic managed to find the food items they consumed among
sending of email messages to participants prompting the list of food available in the R24W with a reduced
them to complete their 24-h food recall. The database memory bias. Since no cues were provided on the check
includes 2865 items linked to the Canadian Nutrient File list about portion size of food items consumed, a memory
(2010 version) or the USDA Nutrients Database for the bias could however influence how subjects were choosing
few items that were not available the Canadian Nutrient the portion sizes when filling the R24W.
File. Questions about the context of the meals are asked
to help respondents to recall all the items consumed. Validation strategy
There is also systematic questions about frequently forgot- In each study included in this protocol, the nutrient com-
ten food items. Portion sizes are represented by up to position of the meals were manipulated without the know-
eight food pictures representing predetermined portion ledge of the participants. Therefore, it would not be the
sizes in a fixed neutral set-up. Portion sizes are expressed most appropriate study design to evaluate the accuracy of
in units and/or volume under each picture. Respondents reported nutrient intakes. Instead, we decided to use a
also have the option to select a multiplicative or a fraction validation strategy in which we compared food items re-
of each portion shown. The format has been designed to ported to food items actually offered. This was done by
be intuitive. In a pre-test in a cohort of 29 adults with dif- classifying reported food items as “matches”, “omissions”
ferent levels of computer skills, the R24W was found to and “inclusions” as previously suggested [6]. A “perfect
be easy to understand and to complete [23]. match” corresponded to a situation where subjects selected
the exact appellation of the item they received in the R24W
Research context (e.g., boiled potatoes for boiled potatoes). A “close match”
Studies in which participants were initially involved had described the selection of an item with related characteris-
different feeding protocols. Most of these studies aimed tics (e.g., mashed potatoes instead of boiled potatoes). A
at assessing the metabolic effects of some diets for which “far match” was used to classify an item in the same food
nutrient composition was manipulated. Menus were for- categories but with different characteristics and nutritional
mulated using typical French Canadian food items. composition (e.g., fried potatoes instead of boiled potatoes).
Meals were prepared and provided by the research team An “omission” was used to define a food item that was pro-
according to a 7-day cyclic menu for 4 to 8 weeks vided but not reported. Finally, an “intrusion” corresponded
followed by 4 to 8 weeks of wash-out where subjects to a food item that was reported but not provided. For all
returned to their normal diet. The portion sizes of the items classified as matches (either perfect, close or far),
different food items offered were individualized to offered and reported portion sizes were compared. Finally,
ensure that all participants maintained a stable body only for indicative purposes, energy and macronutrient
weight while the exact diet composition was kept the intakes as reported by participants who filled the R24W
same for each experimental condition in each study. Par- were compared with actual energy and macronutrient
ticipants were not aware of the precise amount of food intakes as provided by the menu offered.
they received. They were instructed to eat all the food
items provided every day and nothing else. Participants’ Statistical analyses
body weight was measured throughout each project to Proportions of matches and omissions were calculated
achieve isoenergetic conditions and energy intake was and the average number of inclusions was reported.
increased or decreased by 250 kcal/day if a subject lost Omissions were then analyzed in-depth to determine

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Lafrenière et al. BMC Nutrition (2017) 3:34 Page 4 of 10

which categories of food items were more frequently Table 1 Characteristics of the participants (n = 62)
omitted by participants. The impact of these omis- N (%)
sions on energy intake assessment was then evaluated. Sex
To do so, we first classified the omitted items by cat- Men 33 (53.2%)
egories and then calculated the mean contribution to
Women 29 (46.8%)
the daily energy intake of all the mentioned items in
this category. More precisely, the energy content of Mean BMI
each item was determined and then a weighted aver- Normal weight (18.5–24.9 kg/m2) 7 (11.3%)
age was calculated to represent the contribution of Overweight (25.0–29.9 kg/m2) 25 (40.3%)
the category to the daily energy intake. As all menus 2
Obese (30 kg/m and above) 30 (48.4%)
were standardized, food items from a given food cat- Mean age (y)
egory had the same relative contribution to the total
< 25 5 (8.1%)
energy intake for all participants.
The difference between each reported and offered 25–50 30 (48.4%)
portion size was assessed (absolute number) and, in the > 50 27 (43.5%)
present study, we refer to this difference as the bias
which provides an indication of the systematic under- Results
estimation (in case of a negative bias) or overestimation Characteristics of the participants are shown in Table 1.
(in case of a positive bias) in portion sizes. The differ- Data from 62 adults aged 21 to 71 years with body mass
ence between reported and offered portion size was index from 21 to 52 kg/m2 are included in the analyses.
also characterized as the estimation error that is the Participants received on average 16 different food
ratio between the bias and the offered portion size (in items per day. The proportion of matches and omissions
percentage). A Student T-Test was used to compare as well as the number of intrusions in participants’
mean portion size reported to mean portion size responses to the R24W are presented in Table 2. A total
offered. Analysis were conducted on all data and also of 89.3% of the offered food items were reported with
separately on the smaller portions (characterized as the R24W and 76.8% were reported using the exact
less than 100 g) and on the larger portions (charac- descriptor in the database. Descriptions of the main
terized as 100 g and above). A Student T-test was omissions are presented in Table 3. Omissions were clas-
performed to determine if a larger estimation error sified according to different food categories and the
occurs in larger portion sizes compared to smaller weighted average contribution of the category to the
portion sizes. To assess accuracy in portion size esti- offered daily energy intake was calculated. Finally, Table 3
mation, we used correlation coefficients, weighted kappa indicates whether the omitted food items were specifically
scores for classification in quartiles and the Bland-Altman named in the checklist that was provided to participants.
plots. The Kappa score describes the agreement between Table 4 presents examples of small and large portions
two measures as poor (<0.00), slight (0.00–0.20), fair offered at breakfast and at lunch/dinner meals. Most of
(0.21–0.40), moderate (0.41–0.60), substantial (0.61–0.80) the main dishes were offered in portions larger than
and almost perfect (0.81–1.00) [26]. In order to investigate
if individual characteristics could influence portion size es- Table 2 Proportion of matches (exact, close and far) and
timation, a stepwise linear regression model was tested omissions related to the amount of food items offered and
with age, sex and BMI as predictive factors for estimation number of intrusions for all subjectsa (n = 62)
error. Proportion or number of items
In the controlled feeding trials, lunch was served Number of items offered/day (n) 16.1 ± 3.1
on a plate and consumed at the research institute
Exact matches (%) 76.8 ± 15.3
while dinner was provided in individual plastic con-
tainers to be taken home. The extent to which meal Close matches (%) 8.2 ± 8.7
presentation affected the accuracy of portion size es- Far matches (%) 4.3 ± 5.2
timation with the R24W was also assessed using T-tests All matches combined (%) 89.3 ± 11.1
comparing estimation bias between meals. Finally, energy Omissions (%) 10.7 ± 11.1
and macronutrient intakes reported were compared to Intrusions (n) 0.2 ± 0.7
values corresponding to the offered food items using a Stu- a
Perfect match: a situation where subjects selected the exact appellation of
dent T-test. Correlation analyses were also performed be- the item they received in the R24W. Close match: an item with related
tween reported and offered energy and nutrient intakes. characteristics. Far match: an item in the same food category but with
different characteristics and nutritional composition. Omission: a food item
Statistical analyses were conducted with the software SAS that was provided but not reported. Intrusion: a food item that was reported
version 9.4 (SAS Institute). but not provided

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Lafrenière et al. BMC Nutrition (2017) 3:34 Page 5 of 10

Table 3 Counts of the items most frequently omitted by participants in relation to offered items
Food items Number of subjects who Items included in Number of omissions Mean contribution to the
received the item the checklist daily energy intakeb
Vegetables in a salad or a mix dish 72 0.7%
Peppers 45 No 24
Celery 32 No 17
Cucumbers 13 No 11
Corn 26 Yes/Noa 10
Onions 13 No 7
Tomatoes 13 Yes 3
Side vegetables 36 3.1%
Sweet potatoes 32 No 21
Potatoes 32 Yes 4
Coleslaw 32 Yes 3
Tomatoes 13 Yes 2
Cucumbers 13 Yes 2
Broccoli 18 Yes 2
Cauliflower 18 Yes 2
Snacks/drinks 30 6.2%
Cheddar cheese 19 Yes 7
Sweet bread/muffin 73 Yes 6
Raspberries 18 Yes 6
Milk 18 Yes 3
Milk shake 26 Yes 3
Yogurt 31 Yes 3
Blueberries 18 Yes 2
Sauces 26 1.8%
Vinaigrette 13 Yes 7
Salsa 13 Yes 7
BBQ sauce 32 Yes 6
Mayonnaise 13 No 6
Ingredients in a salad 16 3.7%
Feta cheese 13 No 10
Cranberries 13 Yes 3
Chicken 13 Yes 3
a
Corn was offered in two different menus, one where it was written as an ingredient on the checklist (3 omissions/13 presentations) and one where it was not
included on the checklist (7 omissions/13 presentations)
b
The energy content of each item was determined and then a weighted average was calculated to represent the energetic contribution of the category

100 g while side vegetables, fruits, sauces and spreads nonsignificant (P = 0.12) systematic bias of 3.2 g (9.3%
were mostly in portions smaller than 100 g. ±66.0%) meaning that R24W tends to slightly overesti-
Table 5 describes the agreement between reported and mate portion sizes. The correlation coefficient of 0.80 and
offered portion sizes (kappa score, correlation coefficient the weighted kappa score of 0.62 suggest a substantial
and estimation bias). Analyses were first conducted with agreement between reported and offered portion sizes. Of-
all portions irrespective of their size and then with small fered portions of less than 100 g (kappa score of 0.43) as
and large portion sizes separately. When analysing all por- well as portions larger than 100 g (kappa score of 0.50)
tions irrespective of their size, we found a small showed a moderate agreement with their corresponding

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Lafrenière et al. BMC Nutrition (2017) 3:34 Page 6 of 10

Table 4 Examples of small and large portions of food items offered portions were known, these were used as the
offereda independent variables. The plots demonstrated accept-
Small portions (g) Large portions (g) able agreement. The linear regression model showed
Breakfast Bread/bagel 27–88 Orange 100–320 that neither age, sex nor BMI were significantly pre-
Peanut butter 16–50 Milk 125–400 dicting the estimation error in this sample. Results
showed that there was no significant difference in the
Ham 12–40 Milkshake 130–416
estimation error between meals offered for lunch
Cereals 30–96 Orange juice 160–512
(2.9 g ± 134.8 g) and meals offered for dinner
Cream cheese 8–26 Apple sauce 125–400 (−11.0 g ± 93.7 g), P = 0.29 (not shown).
Raspberries 25–80 Finally, comparisons between reported and offered
Blueberries 27–88 energy and macronutrients are presented in Table 6. Signifi-
Lunch/Dinner Cranberries 8–25 Vegetable juice 125–400 cant correlations were observed (r = 0.38–0.64 P < 0.01)
and reported energy and protein intakes were not statisti-
Cucumbers 15–48 Rice with shrimps 320–570
cally different from the offered amounts. However,
Tomatoes 25–80 Potatoes 172–307
reported carbohydrate intakes were significantly lower
Vinaigrette/mayo 10–32 Fajitas with beans 200–355 while reported fat intakes were significantly higher than
Cheese 20–64 Meat loaf 165–528 offered values.
Broccoli 22–72 Pesto pasta 207–368
Carrots 22–72 Chili con carne 330–840
Salsa 14–45 Mexican turkey 108–270
Discussion
The objective of this study was to validate in terms of
Parsley salad 165–528
food items reporting and portion size evaluation a new
Ham quiche 102–256 web-based 24-h recall, the R24W, using the context of
Roasted peppers 97–272 fully controlled feeding studies. We observed that the
Carrot soup 165–528 majority of the offered food items – close to 90%, were
Rice 129–292 reported by the participants and that the mean differ-
ence between offered and reported portion sizes was less
Couscous 137–309
a
than 10% (i.e., 9.3%), as we hypothesized. This slight dif-
A range is presented as portions were individualized according to
participants’ energy needs varying from 1750 to 4500 kcal per day ference resulted in a non-significant 13.9 kcal underesti-
mation of energy intake. However, contrary to our initial
reported portion sizes. The estimation error was signifi- assumptions, we observed that portions smaller than
cantly larger for small (17.1% ±78.5%) than for large 100 g were estimated at a greater error rate than those
portions (−2.4% ±55.8%; P < 0.01). of 100 g and above.
Figures 1, 2 and 3 illustrate the Bland-Alman plots Nutritional assessment errors can be attributed to par-
for differences between reported and offered portion ticipants’ recall bias (reactivity, memory or social desirabil-
sizes for all food items, and also for small portions ity bias) or to inherent characteristics of the tool (e.g.,
and large portions separately. Because the exact inadequate strategy for data collection or visual support).
While it is difficult to distinguish them in a validation
study conducted in the context of real life, fully controlled
Table 5 Agreement between reported and offered portion sizes studies allow to minimize some participants’ related bias.
as determined by the Kappa scores for portion size classification However, few studies have used controlled feeding studies
in quartiles, correlation coefficients and estimation bias for all, to validate automated self-administered 24-h dietary
small and large portions
recalls. Indeed, tools are generally compared to another
Kappa Correlation Estimation bias
score coefficient (P) (estimation error)a
self-reported tool or with biomarkers that provide precise
information about only one nutrient at a time [13].
All portions 0.62 0.80* 3.2 g (9.2%)
N = 1373 Kirkpatrick et al. conducted a similar study intended to
Small portions (<100 g) 0.43 0.46* 7.6 g* (17.1%)
validate the ASA24 [6], another web-based dietary recall,
N = 640 in a context where subjects were invited to eat in a re-
Large portions (≥100 g) 0.50 0.68* −0.6 g (−2.4%) search cafeteria setting and asked to report their intakes
N = 733 during a follow-up visit, the day after. They observed that
a
Estimation bias: the average of the difference between reported and offered the proportion of matches between food consumed in the
portion size. Estimation error: the mean ratio between the bias and the offered lab facility and food reported was 80%, a value that is
portion size
*Significant at P < 0.05 (Difference between reported and offered portion was
slightly lower than what we observed in our study (i.e.,
calculated to obtain the estimation bias) 89%). However, memory was more of a confounding

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Lafrenière et al. BMC Nutrition (2017) 3:34 Page 7 of 10

800

Difference between offered and reported


600

400
portion sizes (g)

200

0
0 100 200 300 400 500 600 700 800 900
-200

-400
Offered portion sizes (g)
Fig. 1 Bland Altman Plot of the comparison between offered and reported portion sizes for all portions. Bias = 3.2 g Limits of agreements
(dotted line): −148.3 to 154.7 g R: −0.15 P: <0.001

factor in that study because subjects were exposed to the underreported or omitted. In our study, this bias was
experimental meals for the first time and did not have any limited because participants did not have to take responsi-
cues to help them remember their past intakes. This was bility for their food choices, which were predetermined as
not the case in our study since subjects might have part of the experimental procedures. As suggested by
received the same meals in the previous phase of their others, the web interface may also contribute to redu-
studies and because they had access to a checklist to help cing the social desirability bias compared with a human
them remember to consume all food items. interviewer [14, 15] but our study was not designed to
Omitted items identified in the present study were differ- specifically address this issue.
ent from those reported in previous studies. Underreport- The most frequently omitted items by our participants
ing of fat and carbohydrates with food records and 24-h were vegetables included in recipes (72 omissions) as
recalls have been repeatedly demonstrated [5, 27]. This is well as side vegetables (36 omissions). This is in accord-
supported by studies showing high-fat foods such as cakes, ance with the observations of Kirkpatrick et al. [6]. Some
pastries, cookies and savoury snacks are more often under- of these items were not extensively described in the
reported than other food groups [28, 29]. However, it is im- checklist that was provided to the participant (87 items
portant to note that in the present study, the variety of not described in the checklist/108 omissions in vegeta-
snacks and desserts was limited as participants were im- bles included in meals and side vegetables), so a memory
posed a specific menu. Nevertheless, they received cakes bias could in part explain these omissions. We can also
and potato chips and those items were not among the most suggest that for these omitted food items the checklist
frequently omitted food items. Hebert et al. [9] suggested was misleading in a way, suggesting to the participant
that underreporting is associated with social desirability, that some items were more important than others. For
which may explain that typical unhealthy foods tend to be example the “Mexican tortillas” included peppers and

500
Difference between offered and reported

400
portion sizes (g)

300

200

100

0
0 20 40 60 80 100 120
-100
Offered portion sizes (g)
Fig. 2 Bland Altman Plot of the comparison between offered portions of less than 100 g and reported portion sizes Bias = 7.6 g Limits of
agreements (dotted line): −76.9 to 92.1 g R: −0.02 P = 0.54

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Lafrenière et al. BMC Nutrition (2017) 3:34 Page 8 of 10

800

Difference between offered and reported


600

portion sizes (g)


400

200

0
100 200 300 400 500 600 700 800

-200

-400
Offered portion sizes (g)
Fig. 3 Bland Altman Plot of the comparison between offered and reported portion sizes of 100 g and larger. Bias = −0.6 Limits of agreements
(dotted line): −192.1 to 190.8 g R: −0.18 P < 0.001

onions even if it was not mentioned on the checklist in assessing the portion sizes of each ingredient. How-
(see Additional file 1). We calculated that the energy ever, our results showed that there was no difference in
contribution of vegetables included in recipes and side portion size estimation that was noted between presen-
vegetables was however minimal (i.e., less than 5% of the tation formats. As meals are not always consumed in a
daily energy intake). plate in real life settings, it is of interest to assess portion
When analysing all food items, irrespective of the por- sizes estimation in different contexts.
tion size, we noted a mean estimation bias of 3.2 g, The observed overestimation of small portions de-
which is close to the −3.7 g differences observed by serves attention. In fact, the estimation error was close
Kirkpatrick et al. [6]. Williamson et al. [30] suggested to 20% and the correlation coefficient between offered
that digital photos helped to accurately estimate portion and reported portion sizes was below 0.50 [32]. Al-
sizes of food items. In the present study, participants though it was not a specific objective of the study due to
tended to overestimate small portion sizes and to under- the blind manipulation of some food items, the analysis
estimate larger portions. This observation has also been comparing reported and offered energy seems to bring
highlighted in a previous study by Nelson et al. [31]. confidence that estimation errors are counterbalanced
However, our estimation error in large portion size was by omissions, or are of low importance in the estimation
only −2.4% suggesting that we selected enough pictures of energy intakes. Indeed, the items most frequently
of larger portion sizes [24]. overestimated are the same as those most frequently
Our results suggest that images used to illustrate the omitted (vegetables, sauces and spread) and the overall
portion sizes in the R24W contribute to influencing ad- energy intake as obtained from the R24W filled by par-
equately estimation of real intakes. It is of importance to ticipants is not significantly different from the energy
remember that participants did not receive information content of the diet offered. However, some differences
about the size of the portion they ate. Moreover, they were found between reported and offered amounts of
received their lunch meal in a plate similar to the one carbohydrate and fat. These differences can be ex-
that appears on R24W portion size images while break- plained, at least partially by the fact that participants
fast and dinner were served in plastic containers where were unaware of the dietary manipulation of the meals
items were often mixed, which can increase the difficulty they received. For example, in two out of three studies,

Table 6 Comparison between reported and offered intakes for energy and macronutrients (n = 62)
Reported intake Offered intake Difference between reported Correlation coefficient between
and offered intakes reported and offered intakes
Energy (kcal) 2762.4 ± 781.1 2776.4 ± 603.8 −13.9 ± 646.3 0.59*
Proteins (g) 110.9 ± 39.2 110.0 ± 25.2 2.1 ± 26.3 0.60*
Carbohydrates (g) 340.9 ± 101.6 366.9 ± 75.9 −26.1 ± 79.0* 0.64*
Fat (g) 111.0 ± 32.0 102.8 ± 26.6 9.1 ± 38.0* 0.38*
*Significant at P < 0.05

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Lafrenière et al. BMC Nutrition (2017) 3:34 Page 9 of 10

they received a milkshake supplemented in fat to assess Availability of data and materials
the metabolic effect of different types of oils. Mean- The datasets used and/or analyzed during the current study available from
the corresponding author on reasonable request.
while, the proportion of fat in other recipes, like muf-
fins was reduced for balancing the diet. These Authors’ contributions
manipulations could explain some of the differences JL analyzed and interpreted most of the data and was actively involved in
drafting the manuscript. BL, CL, JR and SL critically revised the manuscript
observed between reported and offered amounts of and were major contributors to the revised version. JL, SL and CL made
carbohydrate and fat. substantial contributions to the conception and design and the acquisition
Another limitation to consider is the burden associated of data. All authors read and approved the final manuscript.

with self-reporting food intake for many days. In the con- Competing interests
text of this study, participants were asked to fill the R24W The authors declare that they have no competing interests.
twice on top of the initial research requirements. They
Consent for publication
were highly motivated and did not seem to significantly Not applicable.
underreport. However, it will be a concern with wider and
more diversified cohorts. Indeed, considering that at least Ethics approval and consent to participate
Data from three studies were included in this article. These studies received
3 non-consecutive days of 24-h recall are needed to repre- the approval of the Laval University Ethics Committee and participants
sent typical intakes [33], the time commitment for partici- provided written informed consent prior to taking part in the study. The
pants could impact their motivation. Nevertheless, the analysis performs in this article was directly presented in one of the three
consent forms (the latest study) and was proposed to the participants
web-based format of the R24W seems to be an important already included in the two other studies in an addendum to their initial
asset. Based on a pre-test conducted during its develop- consent form. Therefore, all participants gave their informed consent before
ment, 59% of responders completed the recall in less than completing their first 24-h recall knowing that it was part of a validation
process (p.5 ln 1–8).
30 min [23]. In comparison, the food frequency question-
naire used recently in our large cohort studies takes on
Publisher’s Note
average 45 min to complete [34]. Moreover, as the coding Springer Nature remains neutral with regard to jurisdictional claims in
is automated, the web-based approaches lead to a consid- published maps and institutional affiliations.
erable time saving for researchers.
Received: 21 December 2016 Accepted: 1 April 2017

Conclusions References
To conclude, when controlling in part for some of the 1. Beaton GH, Burema J, Ritenbaugh C. Errors in the interpretation of dietary
personal bias (memory, reactivity and social desirability), assessments. Am J Clin Nutr. 1997;65(4):1100S–7.
2. Archer E, Hand GA, Blair SN. Validity of U.S. Nutritional surveillance: national
participants reported most of the items they ate with a
health and nutrition examination survey caloric energy intake data, 1971–
good level of accuracy in portion sizes reported. This 2010. PLoS ONE. 2013;8(10):e76632.
data also provides preliminary evidence supporting the 3. Serra-Majem L, Frost Andersen L, Henríque-Sánchez P, Doreste-Alonso J,
Sánchez-Villegas A, Ortiz-Andrelluchi A, et al. Evaluating the quality of
validity of R24W to assess food intakes. Analyses in lar-
dietary intake validation studies. Br J Nutr. 2009;102(Supplement S1):S3–9.
ger cohorts of free living individuals including biomarker 4. Hill RJ, Davies PSW. The validity of self-reported energy intake as
analysis would be completed soon. This will allow to val- determined using the doubly labelled water technique. Br J Nutr. 2001;
85(4):415–30.
idate dietary assessment with a larger variety of con-
5. Goris AH, Westerterp-Plantenga MS, Westerterp KR. Undereating and
sumed food items. In the near future, the R24W could underrecording of habitual food intake in obese men: selective
be used to assess the food intake in large-scale research underreporting of fat intake. Am J Clin Nutr. 2000;71(1):130–4.
6. Kirkpatrick SI, Subar AF, Douglass D, Zimmerman TP, Thompson FE, Kahle
projects and in nutrition practice.
LL, et al. Performance of the Automated Self-Administered 24-hour Recall
relative to a measure of true intakes and to an interviewer-administered
24-h recall. Am J Clin Nutr. 2014;100(1):233–40.
Additional file 7. Subar AF, Crafts J, Zimmerman TP, Wilson M, Mittl B, Islam NG, et al.
Assessment of the accuracy of portion size reports using computer-based
Additional file 1: Word document (.docx) Annexe 1: Checklist offered to food photographs aids in the development of an automated self-
the participant with their meals for each project. (DOCX 13 kb) administered 24-hour recall. J Am Diet Assoc. 2010;110(1):55–64.
8. Di Noia J, Contento IR. Criterion validity and user acceptability of a CD-
ROM-mediated food record for measuring fruit and vegetable consumption
Abbreviations among black adolescents. Public Health Nutr. 2009;12(1):3–11.
BMI: Body mass index; R24W: Newly developed automated self-administered 9. Hebert JR, Ebbeling CB, Matthews CE, Hurley TG, MA Y, Druker S, et al.
web-based 24-h dietary recall Systematic errors in middle-aged Women’s estimates of energy intake:
comparing three self-report measures to total energy expenditure from
doubly labeled water. Ann Epidemiol. 2002;12(8):577–86.
Acknowledgements
10. Illner A-K, Freisling H, Boeing H, Huybrechts I, Crispim SP, Slimani N. Review
Not applicable.
and evaluation of innovative technologies for measuring diet in nutritional
epidemiology. Int J Epidemiol. 2012;41(4):1187–203.
Funding 11. Subar AF, Kipnis V, Troiano RP, Midthune D, Schoeller DA, Bingham S, et al.
The present study was supported by a grant from the Canadian Institutes of Using intake biomarkers to evaluate the extent of dietary misreporting in a
Health Research (CIHR; grant no. FHG 129921). large sample of adults: the OPEN study. Am J Epidemiol. 2003;158(1):1–13.

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Lafrenière et al. BMC Nutrition (2017) 3:34 Page 10 of 10

12. Carroll RJ, Midthune D, Subar AF, Shumakovich M, Freedman LS, Thompson
FE, et al. Taking advantage of the strengths of 2 different dietary
assessment instruments to improve intake estimates for nutritional
epidemiology. Am J Epidemiol. 2012;175(4):340–7.
13. Willett WC. Nutritional Epidemiology [Internet]. Oxford University Press; 1998
[cited 2016 Jan 26]. Available from: http://www.oxfordscholarship.com/view/
10.1093/acprof:oso/9780195122978.001.0001/acprof-9780195122978
14. Tourangeau R, Rips LJ, Rasinski K. The psychology of survey response.
Cambridge: Cambridge University Press; 2000.
15. Kohlmeier L, Mendez M, McDuffie J, Miller M. Computer-assisted self-
interviewing: a multimedia approach to dietary assessment. Am J Clin Nutr.
1997;65(4):1275S–81.
16. Moshfegh AJ, Rhodes DG, Baer DJ, Murayi T, Clemens JC, Rumpler WV, et al.
The US department of agriculture automated multiple-pass method reduces
bias in the collection of energy intakes. Am J Clin Nutr. 2008;88(2):324–32.
17. Byrd-Bredbenner C, Schwartz J. The effect of practical portion size
measurement aids on the accuracy of portion size estimates made by
young adults. J Hum Nutr Diet. 2004;17(4):351–7.
18. Howat PM, Mohan R, Champagne C, Monlezun C, Wozniak P, Bray GA.
Validity and reliability of reported dietary intake data. J Am Diet Assoc.
1994;94(2):169–73.
19. Thompson FE, Dixit-Joshi S, Potischman N, Dodd KW, Kirkpatrick SI, Kushi
LH, et al. Comparison of interviewer-administered and automated self-
administered 24-hour dietary recalls in 3 diverse integrated health systems.
Am J Epidemiol. 2015;181(12):970–8.
20. Diep CS, Hingle M, Chen T-A, Dadabhoy HR, Beltran A, Baranowski J, et al.
The automated self-administered 24-hour dietary recall for children, 2012
version, for youth aged 9 to 11 years: a validation study. J Acad Nutr Diet.
2015;115(10):1591–8.
21. Liu B, Young H, Crowe FL, Benson VS, Spencer EA, Key TJ, et al.
Development and evaluation of the Oxford WebQ, a low-cost, web-based
method for assessment of previous 24 h dietary intakes in large-scale
prospective studies. Public Health Nutr. 2011;14(11):1998–2005.
22. Government of Canada SC. The Daily — Individual Internet use and
e-commerce, 2012 [Internet]. 2013 [cited 2017 Jan 23]. Available from:
http://www.statcan.gc.ca/daily-quotidien/131028/dq131028a-eng.htm
23. Jacques S, Lemieux S, Lamarche B, Laramée C, Corneau L, Lapointe A, et al.
Development of a Web-based 24-h dietary recall for a French-Canadian
population. Nutrients. 2016;8(11):724.
24. Harnack L, Steffen L, Arnett DK, Gao S, Luepker RV. Accuracy of estimation
of large food portions. J Am Diet Assoc. 2004;104(5):804–6.
25. Nelson M, Atkinson M, Darbyshire S. Food Photography I: the perception of
food portion size from photographs. Br J Nutr. 1994;72(5):649–63.
26. Landis JR, Koch GG. The measurement of observer agreement for
categorical data. Biometrics. 1977;33(1):159–74.
27. Heitmann BL, Lissner L, Osler M. Do we eat less fat, or just report so? Int J
Obes Relat Metab Disord. 2000;24(4):435–42.
28. Vereecken CA, Covents M, Matthys C, Maes L. Young adolescents’ nutrition
assessment on computer (YANA-C). Eur J Clin Nutr. 2005;59(5):658–67.
29. Lafay L, Mennen L, Basdevant A, Charles MA, Borys JM, Eschwège E, et al.
Does energy intake underreporting involve all kinds of food or only specific
food items? Results from the Fleurbaix Laventie Ville Santé (FLVS) study. Int
J Obes Relat Metab Disord. 2000;24(11):1500–7.
30. Williamson DA, Allen HR, Martin PD, Alfonso AJ, Gerald B, Hunt A.
Comparison of digital photography to weighed and visual estimation of
portion sizes. J Am Diet Assoc. 2003;103(9):1139–45.
31. Nelson M, Atkinson M, Darbyshire S. Food photography II: use of food
photographs for estimating portion size and the nutrient content of meals. Submit your next manuscript to BioMed Central
Br J Nutr. 1996;76(1):31–49.
32. Masson L, MCNeill G, Tomany J, Simpson J, Peace H, Wei L, et al. Statistical and we will help you at every step:
approaches for assessing the relative validity of a food-frequency
• We accept pre-submission inquiries
questionnaire: use of correlation coefficients and the kappa statistic. Public
Health Nutr. 2003;6(3):313–21. • Our selector tool helps you to find the most relevant journal
33. Stote KS, Radecki SV, Moshfegh AJ, Ingwersen LA, Baer DJ. The number of • We provide round the clock customer support
24 h dietary recalls using the US Department of Agriculture’s automated
• Convenient online submission
multiple-pass method required to estimate nutrient intake in overweight
and obese adults. Public Health Nutr. 2011;14(10):1736–42. • Thorough peer review
34. Labonté M-È, Cyr A, Baril-Gravel L, Royer M-M, Lamarche B. Validity and • Inclusion in PubMed and all major indexing services
reproducibility of a web-based, self-administered food frequency
• Maximum visibility for your research
questionnaire. Eur J Clin Nutr. 2012;66(2):166–73.
Submit your manuscript at
www.biomedcentral.com/submit

Content courtesy of Springer Nature, terms of use apply. Rights reserved.


Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers and authorised users (“Users”), for small-
scale personal, non-commercial use provided that all copyright, trade and service marks and other proprietary notices are maintained. By
accessing, sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of use (“Terms”). For these
purposes, Springer Nature considers academic use (by researchers and students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and conditions, a relevant site licence or a personal
subscription. These Terms will prevail over any conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription
(to the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of the Creative Commons license used will
apply.
We collect and use personal data to provide access to the Springer Nature journal content. We may also use these personal data internally within
ResearchGate and Springer Nature and as agreed share it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not
otherwise disclose your personal data outside the ResearchGate or the Springer Nature group of companies unless we have your permission as
detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial use, it is important to note that Users may
not:

1. use such content for the purpose of providing other users with access on a regular or large scale basis or as a means to circumvent access
control;
2. use such content where to do so would be considered a criminal or statutory offence in any jurisdiction, or gives rise to civil liability, or is
otherwise unlawful;
3. falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association unless explicitly agreed to by Springer Nature in
writing;
4. use bots or other automated methods to access the content or redirect messages
5. override any security feature or exclusionary protocol; or
6. share the content in order to create substitute for Springer Nature products or services or a systematic database of Springer Nature journal
content.
In line with the restriction against commercial use, Springer Nature does not permit the creation of a product or service that creates revenue,
royalties, rent or income from our content or its inclusion as part of a paid for service or for other commercial gain. Springer Nature journal
content cannot be used for inter-library loans and librarians may not upload Springer Nature journal content on a large scale into their, or any
other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not obligated to publish any information or
content on this website and may remove it or features or functionality at our sole discretion, at any time with or without notice. Springer Nature
may revoke this licence to you at any time and remove access to any copies of the Springer Nature journal content which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or guarantees to Users, either express or implied
with respect to the Springer nature journal content and all parties disclaim and waive any implied warranties or warranties imposed by law,
including merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published by Springer Nature that may be licensed
from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a regular basis or in any other manner not
expressly permitted by these Terms, please contact Springer Nature at

[email protected]

You might also like