Integrating Technology and Culture: Smartphone Validation of a Food Frequency Questionnaire for Nutrient Intake Estimates in the Adult Population of Trinidad and Tobago

Abstract

Background

Trinidad and Tobago is home to a multiethnic population, each with distinct dietary traditions. Given this diversity, it is essential to validate a food frequency questionnaire (FFQ) that captures the local food items that contribute to the nation’s unique dietary culture.

Objectives

This study aims to assess the reproducibility and validity of a semiquantitative FFQ to estimate nutrient intake in the adult population of Trinidad and Tobago.

Methods

A 139-item semiquantitative electronic food frequency questionnaire (e-FFQ), developed using Google Forms, captured culture-specific foods commonly consumed in Trinidad and Tobago. The self-administered e-FFQ was distributed to 91 participants aged 18 and older, with 2 administrations 3 mo apart. The first administration of the e-FFQ was validated against the weighted mean of 4 food records with digital images as the reference method. Data were analyzed using SPSS Version 26 to assess validity and reproducibility through paired t-tests, correlations, and cross-classification.

Results

Participants had a mean age of 38 ± 9.6 y, with 22% male and 78% female. Correlations between the e-FFQ and food records ranged from moderate (r = 0.59 for vitamin C) to high (r = 0.83 for carbohydrates). Cross-classification agreements varied from 69% for cholesterol to 89% for fiber and vitamin A. Energy-adjusted correlations averaged r = 0.37, ranging from r = 0.22 for polyunsaturated fatty acids to r = 0.67 for cholesterol. Cross-classification indicated that 61% of e-FFQ estimates were correctly classified within ±1 quintile.

Conclusions

The culture-specific e-FFQ demonstrates strong reproducibility and validity, making it a valuable tool for assessing nutrient intake in Trinidad and Tobago’s adult population.

Introduction

Recent advancements in technology have markedly transformed dietary assessment methodologies, integrating innovative digital tools such as smartphone applications, web-based programs, and automated systems [1]. These technological strides have enhanced the accuracy and efficiency of dietary data collection, processing, and analysis, addressing many limitations associated with traditional methods [2]. The shift toward digital and web-based food frequency questionnaires (FFQs) has revolutionized large-scale epidemiological studies by reducing costs and improving response completeness [1].

Automated FFQs, including those implemented on computer and web platforms, offer significant advantages such as expedited data collection, improved data quality, and reduced incidences of missing data [3]. These digital tools can also utilize visual aids to assist respondents in estimating portion sizes and provide enhanced privacy and confidentiality, which can reduce under-reporting [4]. Moreover, the flexibility of online FFQs enables continuous data collection from any location, with secure electronic storage, helping to minimize risk of data loss [4,5].

In addition to FFQs, technological innovations have facilitated self-monitoring of dietary intake via mobile apps such as MyFitnessPal and My Meal Mate, and automated systems like ASA24 for 24-h recalls and diet records [5]. Researchers are encouraged to consider participants' readiness for these automated methods and to adhere to best practice guidelines to ensure the appropriate use of dietary assessment tools.

For culturally diverse populations, accurate dietary assessment is crucial due to the varying prevalence of noncommunicable diseases (NCDs) and dietary patterns linked to health outcomes. Effective dietary assessment tools are needed to capture culturally specific foods and dietary habits, which can inform public health policies and interventions aimed at reducing NCD risks. Designing a culture-specific FFQ is essential to accurately reflect the dietary practices of a given population [6,7], and it must be validated for reproducibility and relevance to the specific ethnic group [8]. On the basis of specific context and purpose for development, various researchers have successfully used particular approaches for evaluating validity and reliability of FFQs over the years [[9], [10], [11], [12]], whereas others have identified the need to alter or improve on different aspects of the process for specific measurement outcomes [13,14].

Trinidad and Tobago, a multiethnic Caribbean nation and the industrial capital of the region, boasts a 99% adult literacy rate, with 80.1% of the population using the internet and over 900,000 active on social media [15]. Despite these advancements, the country faces a high prevalence of NCDs, which account for 80% of all deaths [16]. Given these health challenges, there is an urgent need for a dietary assessment tool that accurately reflects the diverse nutrient intake of the population. Thus, we developed a semiquantitative e-FFQ tailored to Trinidadian and Tobagonian diets incorporating local and street foods. This e-FFQ was previously evaluated for its reproducibility and validated in assessing habitual food/food groups intake and was found to be a valid tool for assessing and ranking food category intake estimates for the population [17].

Given that a culture-specific FFQ can be a cost-effective tool to examine the role of diet in health and disease, particularly in addressing key NCDs at the national level in Trinidad and Tobago, further evaluation of this e-FFQ is warranted to determine its effectiveness in estimating habitual nutrient intake. This study aims to evaluate the reproducibility and validity of a comprehensive culture-specific semiquantitative e-FFQ developed for adults in Trinidad and Tobago in estimating nutrient intakes. This e-FFQ has the potential to be an effective tool in measuring habitual nutrient intake across the population of Trinidad and Tobago to identify deficiencies and excesses as well as diet quality. When used alongside other data, for example, biochemical markers, anthropometric and/or other health parameters, this e-FFQ may help in determining optimal nutrient intakes associated with NCD risk prevention.

Methods

Study design

A cross-sectional observational study was conducted to evaluate the reproducibility and validity of a culture-specific semiquantitative electronic food frequency questionnaire (e-FFQ) to assess habitual dietary intake over the past 3 mo among adults in Trinidad and Tobago. This study was described elsewhere [17]. Briefly, the e-FFQ was initially sent via email to the participants from March to April 2019 and readministered between June and July 2019. During the interval between the 2 administrations, participants recorded their food intake for 4 separate days using their smartphones. Four 1-d food records (FRs) that covered 2 weekdays and 2 weekend days were required from each participant during the 3-mo duration of the study. The period of FR collection (12 wk) was divided into 4 3-wk periods. During each 3-wk period, one FR was collected from each participant as illustrated in Figure 1.

FIGURE 1.

This diagram shows the timeline for administration of the e-FFQs and multiple 1-d food records. e-FFQ, electronic food frequency questionnaire; FFQ, food frequency questionnaire; wk, week.

Participants were divided into 3 groups of 35 (n = 105) and each group was further divided into 6 subgroups and assigned a day in the week to record their food intake between Saturday and Thursday of each week of the 3-wk periods. A total of ∼ 105 FRs were collected each 3-wk period. The subgroups were alternated to ensure each subgroup recorded 2 weekdays and 2 weekend days over a 12-wk period. Participants had the choice of sending their FRs via WhatsApp or email. A Google calendar linked to an email account (ttfoodfreq@gmail.com) was used as an internal tracking system to monitor when to send out the day/night-before reminders to participants that they had to bring their fiducial marker and charge their smartphones for use on the next day.

Study participants

Participants were recruited through email, social media platforms, and professional associations. These platforms included Facebook groups such as Trinidad and Tobago Association of Teacher Educators, Trinidad and Tobago Police Service Social and Welfare Association, Trinidad and Tobago Registered Nurses Association, Foodie Nation, Caribbean Kitchen, and Trini Food and Recipes. Those interested completed a brief screening questionnaire via Google Forms to determine their eligibility.

Adults aged 18 y and above with access to the internet and who own a smartphone were included.

Those who have a medical condition requiring a therapeutic diet (such as Crohn’s disease, celiac disease, or end-stage renal disease) or being out of the country during the study period were deemed ineligible for the study.

After 125 individuals completed the screening form, 9 were excluded for being out of the country during the study, 3 for not providing complete contact information, and 1 for having Crohn's disease. An information package and informed consent form were emailed to the 112 eligible participants. On returning the signed consent forms, participants received the e-FFQ for self-administration. Participants could ask questions via email or phone. Of the 112 eligible participants, 14 (12%) did not complete the first e-FFQ administration, and 7 (7%) of those who completed it did not finish the second administration, resulting in a final sample size of 91 (10 from Tobago and 81 from Trinidad). All 91 participants provided 3–5 1-d FRs. The study was approved by the Loma Linda University Institutional Review Board (IRB #5180409) and informed consent was obtained from each participant before administering the e-FFQ.

Electronic food frequency questionnaire

The semiquantitative e-FFQ, designed as an electronic survey using Google Forms, included detailed instructions and examples. The e-FFQ was adapted from an existing 146-item FFQ originally developed by Ramdath et al. [7]. Similar foods were grouped together (for example, coconut bread, sweet bread, and coconut drops). The final e-FFQ included 129 items from the original FFQ and 14 popular street foods, totaling 139 items [11].

The e-survey consisted of 2 parts: a 139-item FFQ and a demographic section. The first part contained 139 food items grouped according to the Caribbean’s Six food groups [18] (see Supplemental Table 1). Intake frequency for all food items except water was assessed using 8 categories: never or less than once per month, 1–3 times per month, once per week, 2–4 times per week, 5–6 times per week, once per day, 2–3 times per day, and ≥4 times per day. Consumption frequencies for water intake are reflected per daily intake [17]. Food items of similar nutrient composition were grouped under the following categories: 1) street foods, 2) staple with subcategories of i) breads, cakes and cereals, ii) rice and pasta, iii) ground provisions, 3) vegetables, 4) legumes, 5) fruits, 6) food from animals with subcategories of i) eggs, dairy and dairy substitutes, ii) meat and meat substitutes, 7) fats and oils. Additional categories were created to capture consumption patterns of 8) water, 9) nonalcoholic beverages, 10) alcoholic beverages, and 11) sweets.

The street food category was placed first to minimize duplicate reporting of foods on the e-FFQ because street foods are mainly mixed local dishes which may be comprised of individual food items on the FFQ. For example, bake and shark/fried fish is a common street food but its component food can be eaten individually with other food items. Instructions were provided to avoid reporting the same foods in multiple sections.

The demographic and anthropometric section included items on anthropometric measurements (weight, height), including age, sex (male, female), marital status, religion, race (Afro-Trinidadian, Indo-Trinidadian, Mixed), education (some university or higher/secondary school), employment, physical activity, smoking, dietary supplements, and chronic disease (history one or more NCD, none). BMI was calculated from self-reported weight and height as seen in Supplementary material Trinidad and Tobago Dietary Assessment Questionnaire (Table 1).

TABLE 1.

Demographic characteristics of participants at baseline (n = 91).

	Male	Female	All
n (%)	19 (21)	72 (79)	91
Age (y), mean ± SD	36.9 (9.5)	38.7 (9.6)	38.4 (9.6)
Height (cm), mean ± SD	177.5 (7.8)	165.3 (7.1)	167.9 (8.8)
Weight (kg), mean ± SD	91.8 (18.6)	84.8 (24.5)	86.3 (24.52)
Waist circ. (cm), mean ± SD	93.8 (10.6)	93.2 (16.2)	93.3 (15.2)
BMI (kg/m2), mean ± SD	29.5 (6.1)	30.5 (7.9)	30.5 (7.9)
BMI categories (%)
Normal	26.3	25	25.3
Overweight	36.8	33.3	34.1
Obese	36.9	41.7	40.7
Race (%)
Afro-Trinidadian	84.2	75	76.9
Indo-Trinidadian	5.3	9.7	8.8
Mixed race	10.5	15.3	14.3
Education level (%)
Some university or above	84.3	90.3	89
Secondary school	15.7	9.7	11
Employment (%)
Employed	89.5	84.7	85.7
Unemployed	10.5	15.3	14.3
Physical activity frequency (%)
≤1 per week	21.1	33.3	30.8
1–2 times per week	31.6	29.3	29.7
≥3 times per week	47.4	37.6	39.6
Chronic disease status (%)
One or more NCDs	21.1	26.4	25.3
None	78.9	73.6	74.7

Abbreviation: NCDs, noncommunicable diseases.

FRs with digital photographs

To validate the e-FFQ, multiple 1-d FRs with digital photographs were used as the reference method; this method is described elsewhere [19]. Briefly, the FR entails the use of a smartphone to digitally photograph foods/meals that were eaten and to text the name of the meal, time, and place of intake, and the names of foods and the corresponding amounts eaten. Participants were to take photographs and text the description of every eating occasion. Both the photographs and texts were sent via email or WhatsApp to a designated contact information. Participants are required to include a fiducial marker according to placement instructions before taking pictures of what they were to eat. A fiducial marker specifically designated for each participant was sent a week before the training on fiducial marker placement. The training was necessary to aid the researcher in portion size estimation and to make sure the participants are not sending internet images of foods. Each 1-d FR included text descriptions and photographs of meals/foods with fiducial markers as seen in Supplemental Figure 1.

A registered dietitian at the doctoral level trained in collecting FRs and the use of Nutrition Data System for Research (NDS-R) analysis, reviewed and clarified any incomplete or unclear data before collating the images and texts into a PowerPoint document where each slide includes an imbedded ruler. One PowerPoint document with 3 or more slides that records the meals eaten in a day represents a 1-d FR.

Nutrient intake estimates

Given the lack of a comprehensive food composition database for Caribbean countries, particularly Trinidad and Tobago, a database encompassing most commonly available foods was utilized. Dietary intake data from both the e-FFQ and FRs were analyzed using the NDS-R software version 2018, developed by the Nutrition Coordinating Center, University of Minnesota.

This software features a database of over 18,000 foods and provides values for 174 nutrients, nutrient ratios, and other food components. Local composite dishes and street foods were analyzed by creating recipes based on traditional recipes from a well-established Trinidad and Tobago cookbook, the Naparima Girls’ High School Cookbook within the NDS-R software. Nutrient intake calculations were performed by multiplying the weight in grams of each food item by its consumption frequency, with e-FFQ responses converted to mean daily intake in grams.

Data Analysis

Sample size

A sample size of 84 was required for this study with a medium effect size (r = 0.3), a power of 0.80, and an alpha level of 0.05. To account for a 20% dropout rate, the adjusted sample size was 101.

Reproducibility analysis

To evaluate the reproducibility of the e-FFQ responses, data distributions were assessed, and non-normal distributions were log-transformed. Descriptive statistics, including means and SDs, were calculated, and paired t-tests compared means. Correlation analyses assessed the agreement between nutrient estimates from the 2 administrations of the e-FFQ. Pearson correlations measured the linear association between both administrations of the e-FFQ. Energy adjustment was calculated using the residual method. Residuals were calculated in SPSS with total energy as the independent variable and nutrient intake as the dependent variable.

Cross-classifications of nutrient intakes between e-FFQ1 and e-FFQ2 were analyzed by dividing intakes into quintiles to determine the proportion of participants categorized into the same quintiles or within ±1 adjacent quintile.

Validation analysis

For validation purposes, 11 participants were excluded due to implausible dietary intake values recorded in the FFQ [9,20]. For this study, we used a more conservative threshold that aligns with established practices. Implausible dietary intake values were defined as energy intakes below 900 kcal or above 5000 kcal for males, and below 600 kcal or above 3500 kcal for females.

To validate the e-FFQ, 4 1-d FRs—2 from weekends and 2 from weekdays were collected to reflect habitual intake. The weighted mean nutrient intake from the FRs was calculated using the formula:

$$\text{Weighted}\text{mean}\text{of}\text{nutrient}\text{intake}=\left[\right(5/2\left(\text{WD}1+\text{WD}2\right)+\left(\text{WE}1+\text{WE}2\right)]/7$$

where WD represents weekday intake and WE represents weekend intake. This weighted mean was compared with nutrient estimates from the first e-FFQ administration using paired t-tests.

Crude nutrient intakes were log-transformed to approximate normal distributions, and Pearson correlations measured the linear association between e-FFQ nutrient estimates and FRs, adjusted for energy using the residual method. Residuals were calculated in SPSS with total energy as the independent variable and nutrient intake as the dependent variable. To correct for within-person variation due to repeated measurements in the reference method FR, 2 synthetic wk were created, which consisted of 1 weekday and 1 weekend day. The nutrient intakes for weeks 1 and 2 were then adjusted for total energy intake and compared with the FFQ.

Cross-classifications of nutrient intakes between the e-FFQ and FRs were assessed by dividing intakes into quintiles to determine the proportion of participants correctly categorized into the same or adjacent quintiles. The Bland–Altman plot was used to evaluate the performance of the e-FFQ compared with the reference method (weighted mean of FRs). This analysis plotted the difference in intake between the e-FFQ and FRs on the y-axis against the mean of the e-FFQ and FRs on the x-axis, with limits of agreement (mean difference ± 1.96 SD) displayed to illustrate the degree of agreement between the 2 methods.

Results

The participants in the study were predominantly adults with a mean age of 38 y, indicating that the sample is largely in the middle stages of adulthood. On mean, participants weighed 86.3 kg with a BMI of 30 indicating that the majority were classified as either overweight or obese. Specifically, only 25% of the participants had a BMI within the normal range (18.5–24.9) whereas the remaining 75% exceeded this range, highlighting a significant prevalence of elevated body weight among the group.

Educational attainment among the participants was notably high, with the vast majority (89%) having achieved at least some level of university education which suggests a well-educated group.

Reproducibility of FFQ

A total of 91 participants participated in both e-FFQ administrations. Table 2 shows the mean nutrient intakes in the first and second administrations of the e-FFQ and the mean difference between e-FFQ1 and e-FFQ2. The mean reported intake of nutrients was higher in the first administration (e-FFQ1) than in the second administration.

TABLE 2.

Mean daily nutrient intake estimates and reproducibility using Pearson correlation coefficient and agreement between e-FFQ1¹ and e-FFQ2².

Nutrients (N = 91)	e-FFQ11	e-FFQ22	Mean diff3	P value	Correlations3		%4
	Mean (SD)	Mean (SD)			Crude	Energy adjusted
Energy (kcal)	1951.1 (1211.4)	1643.3 (812.8)	307.8 (724.1)	<0.001	0.81∗		85
Carbohydrates (g)	257.1 (186.2)	212.0 (119.1)	45.0 (109.1)	<0.001	0.83∗	0.66∗	87
Protein (g)	77.2 (47.5)	65.7 (32.8)	11.5 (31.9)	0.001	0.74∗	0.69∗	82
Fats (g)	72.8 (43.2)	62.5 (32.1)	10.2 (30.5)	0.002	0.71∗	0.57∗	87
SFA5 (g)	20.9 (12.9)	18.2 (10.3)	2.7 (9.5)	0.008	0.68∗	0.58∗	86
Omega3 (g)	2.6 (1.5)	2.3 (1.1)	0.3 (1.2)	0.010	0.61∗	0.47∗	73
Trans fat (g)	1.0 (0.8)	0.9 (0.6)	0.11 (0.5)	0.068	0.65∗	0.52∗	82
Cholesterol (mg)	233.1 (257.6)	195.8 (216.8)	37.2 (150.9)	0.021	0.81∗	0.71∗	82
Fiber (g)	32.4 (23.0)	27.1 (15.5)	5.3 (15.4)	0.001	0.75∗	0.61∗	89
Vitamin A (IU)	15,571.6 (11,931.8)	13,166.3 (9665.5)	2405.3 (8811.4)	0.011	0.69∗	0.62∗	89
Vitamin C (mg)	194.9 (175.7)	155.4 (122.6)	39.6 (143.3)	0.010	0.59∗	0.48∗	85
Vitamin D (μg)	5.9 (4.4)	5.1 (4.2)	0.76 (3.3)	0.032	0.69∗	0.66∗	77
Vitamin E (mg)	13.3 (8.4)	10.6 (6.6)	2.7 (6.6)	<0.001	0.62∗	0.51∗	84
Vitamin B12 (mg)	4.1 (2.7)	3.6 (2.6)	0.5 (2.3)	0.038	0.61∗	0.5∗	67
Folate (μg)	561.2 (373.9)	461.3 (241.8)	0.17 (0.3)	< 0.001	0.73∗	0.4∗	85
Choline (mg)	359.0 (265.8)	300.3 (215.4)	58.7 (141.6)	< 0.001	0.85∗	0.67∗	78
Iron (mg)	16.2 (10.0)	13.6 (7.3)	2.5 (7.0)	0.001	0.71∗	0.35∗	84
Calcium (mg)	756.4 (541.8)	679.3 (396.5)	77.0 (3.3)	0.057	0.71∗	0.48∗	84
Phosphorus (mg)	1195.9 (722.7)	1023.1 (517.7)	172.8 (470.1)	0.001	0.76∗	0.55∗	85
Sodium (mg)	3070.9 (1588.6)	2689.9 (1347.7)	381.1 (1200.7)	0.003	0.68∗	0.32∗	80
Potassium (mg)	3437.1 (2553.3)	2787.1 (1548.2)	649.9 (1696.7)	<0.0001	0.76∗∗	0.66∗	88
Zinc (mg)	10.4 (6.5)	8.8 (4.6)	1.6 (4.6)	0.001	0.71∗	0.35∗	80

Abbreviation: e-FFQ, electronic food frequency questionnaire.

∗Correlation is significant at the 0.01 level (2-tailed).

∗∗Correlation is significant at the <0.0001 level (2-tailed).

¹e-FFQ1 is the first administration.

²e-FFQ2 is the second administration.

³Determined with Pearson’s correlations.

⁴Percent agreement between the 2 e-FFQ administrations within the adjacent quintiles (exact ±1 quintile).

⁵SFA Saturated Fatty Acids.

The correlations for nutrients varied from moderate (r = 0.59) for vitamin C to high (r = 0.85) for choline (Table 2). The mean correlation coefficient for nutrients was 0.7. Most nutrient showed strong crude correlations indicating good reproducibility between FFQ1 and FFQ2. The energy-adjusted correlations were slightly lower, reflecting variability independent of energy intake. Nutrients such as choline, cholesterol, and carbohydrates maintained high reproducibility across all adjustments.

Cross-classification into quintiles showed that an mean of 46% of the responses in both administrations of the e-FFQ were in agreement. Agreement within 1 quintile (that is, exact ± 1 adjacent quintile) ranged from 73% for omega-3 to 89% for fiber and vitamin A. Gross misclassification across categories between e-FFQ1 and e-FFQ2 ranged from 0% to 3.2%.

Validation of FFQ

Weighted mean of 4 FRs was used as the reference method to assess the ability of the e-FFQ to estimate the intake of macronutrients and selected micronutrients. The mean intakes and difference in mean between the e-FFQ1 and the FR are shown in Table 3. The mean nutrient intake estimates of the e-FFQ1 were lower than those of the reference method. Crude or energy-unadjusted correlations ranged from 0.11 for total fats to 0.62 for cholesterol. Energy adjustment increased the correlations which ranged from 0.21 for total fats to 0.67 for cholesterol (see Table 3).

TABLE 3.

Mean daily nutrient intake estimates, validity correlations and agreement between e-FFQ1¹ and Food Records² on nutrient intake estimates.

Nutrients (N = 80)	e-FFQ11	FR2	Mean diff3	P value	Correlations3			%4
	Mean ± SD	Mean ± SD			Crude	Energy adjusted	Deattenuated
Energy (kcal)	1951.1 (1211.4)	2191.1 (678.4)	–422.2	< 0.001	0.2
Carbohydrates (g)	257.1 (186.2)	303.9 (110.5)	–72.6	< 0.001	0.29∗∗	0.42∗∗	0.41∗∗	58
Protein (g)	77.2 (47.5)	82.0 (27.5)	–11.6	0.003	0.29∗∗	0.36∗∗	0.35∗∗	64
Fats (g)	72.8 (43.2)	76.7 (28.6)	–10.25	0.017	0.11	0.21	0.17∗∗	54
SFA (g)	20.9 (12.9)	23.2 (8.6)	–4.3	0.001	0.30∗∗	0.42∗∗	0.26∗∗	63
Omega3 (g)	2.6 (1.5)	2.8 (1.6)	–0.3	0.562	0.06	0.18	0.18∗∗	60
Trans fat (g)	1.0 (0.8)	1.6 (0.9)	–0.7	0.000	0.34∗∗	0.46∗∗	0.52∗∗	75
Cholesterol (mg)	233.1 (257.6)	232.5 (110.5)	–31.2	0.002	0.62∗∗	0.67∗∗	0.63∗∗	76
Fiber (g)	32.4 (23.0)	28.1 (11.9)	1.4	0.562	0.36∗∗	0.48∗∗	0.51∗∗	73
Vitamin A (IU)	15,571.6 (11,931.8)	11,160.6 (8081.9)	4075.7	0.001	0.26∗	0.37∗∗	0.40∗∗	59
Vitamin C (mg)	194.9 (175.7)	155.8 (113.1)	20.6	0.034	0.28∗	0.31∗∗	0.37∗∗	63
Vitamin D (μg)	5.9 (4.4)	5.2 (4.0)	0.3	0.127	0.43∗∗	0.45∗∗	0.49∗∗	63
Vitamin E (mg)	13.3 (8.4)	10.8 (4.5)	1.6	0.057	0.19	0.29∗∗	0.19∗∗	63
Vitamin B12 (mg)	4.1 (2.7)	3.8 (2.0)	0.1	0.649	0.29∗∗	0.35∗∗	0.42∗∗	65
Folate (μg)	561.2 (373.9)	562.0 (238.3)	–35.8	0.287	0.23∗	0.26∗	0.27∗∗	65
Choline (mg)	359.0 (265.8)	330.8 (110.6)	–9.0	0.238	0.37∗∗	0.46∗∗	0.50∗∗	69
Iron (mg)	16.2 (10.0)	15.9 (5.3)	–0.8	0.057	0.24∗∗	0.25∗∗	0.28∗∗	56
Calcium (mg)	756.4 (541.8)	870.5 (342.2)	–187.6	0.001	0.13	0.33∗∗	0.31∗∗	60
Phosphorus (mg)	1195.9 (722.7)	1226.8 (354.1)	–133.4	0.017	0.22∗	0.42∗∗	0.4∗∗	58
Sodium (mg)	3070.9 (1588.6)	3407.5 (1181.8)	–503.7	0.003	0.15	–0.1	–0.03	48
Potassium (mg)	3437.1 (2553.3)	2921.2 (1055.3)	180.2	0.289	0.31∗∗	0.45∗∗	0.46∗∗	69
Zinc (mg)	10.4 (6.5)	9.8 (3.4)	–0.3	0.391	0.16	0.27∗	0.32∗∗	64

Abbreviations: e-FFQ, electronic food frequency questionnaire; FR, food record.

¹e-FFQ1 is the first administration.

²FR is the weighted mean of the 4 1-d food records.

³Determined with Pearson’s correlations.

⁴Percent agreement between the eFFQ1 and FR within the adjacent quintiles (exact ±1 quintile).

^∗Correlation is significant at <0.05 level (2-tailed).

^∗∗Correlation is significant at <0.01 level (2-tailed).

Correlations were moderately strong (r $\ge$ 0.5) for cholesterol, fiber, trans-fat, choline, and potassium, and moderate (0.3 r < 0.5) for carbohydrates, protein, SFAs, and vitamins A, C, and E. Deattenuation of the correlation coefficients to correct for within-person variations in intake due to weighted mean of fFRs only improved some micronutrients such as (zinc, trans-fat, vitamin C, and vitamin B12).

For the cross-classification analysis, the mean exact agreement ±1 adjacent quintile increased from 60% to 63% after adjusting for the total energy intake. Cross-classification showed that an mean of 28% was correctly classified into exact agreement. Cross-classification for exact agreement ±1 adjacent quintile between the FFQ and FR ranged from 48% for sodium to 76% for cholesterol after adjusting for total energy intake.

Figure 2 shows the Bland–Altman plots for energy, carbohydrates, proteins, and fats. For all macronutrients and energy, most points were within the 95% CI limits, with few observations outside the CI limits.

FIGURE 2.

Bland–Altman plots comparing dietary intake estimates from FFQ1 and the reference method (multiple 1-d food records) for macronutrients: (A) energy (kcal), (B) protein (g), (C) carbohydrates (g), and (D) fat (g). The difference between FFQ1 and food record values (FFQ1-FR) is plotted on the y-axis against the corresponding food record values on the x-axis. The red line indicates the mean difference (bias) whereas the blue dashed lines represent the 95% limits of agreement (LoA) calculated as the mean difference ±1.96 SD of the differences. These plots assess the extent of agreement and potential systemic bias across the intake range. FFQ1, food frequency questionnaire 1.

Discussion

This study assessed the validity and reproducibility of a culture-specific e-FFQ within the adult population of Trinidad and Tobago. The e-FFQ featured foods commonly consumed by participants, as corroborated by their weighted mean of FRs. Compared with paper-based FFQs, which involve manual data entry and high labor costs, and online FFQs, which incur maintenance costs and have limited accessibility, the electronic FFQ offers advantages in distribution, administration, and response collation in a confidential manner [9]. This approach was well-suited to the Trinidadian and Tobagonian adults, who are highly educated and have internet and smartphone access [15].

Results of the study showed that the e-FFQ has a good validity when compared with the weighted mean of FRs. The mean estimated intakes of the evaluated macro and micronutrients did not differ significantly between the 2 methods. Mean intakes were lower in the e-FFQ indicating that there was underestimation of energy and nutrient intake by participants when completing the e-FFQ. In contrast, other culture-specific validation studies often find overestimation of intake on FFQs [9,[21], [22], [23]]. This discrepancy might be due to the anthropometric profile of our study population, where a high proportion of participants were overweight or obese. Previous research suggests that individuals with higher BMIs may under-report energy intake [24,25].

Correlation coefficients in our study ranged from 0.18 to 0.67, which aligns with validation studies in other Caribbean countries, where correlations ranged from 0.02 to 0.66 for Barbados and 0.17 to 0.86 for Jamaica [26,27]. Willett's criteria suggest that acceptable correlation coefficients for FFQ validation are between 0.4 and 0.7 [9]. However, some coefficients, especially for total fat, omega-3 fatty acids, and sodium, were not significant even after adjusting for energy intake. This might be due to subject recall error or variations in intake.

The validity correlation coefficients for macronutrients are generally higher than those for micronutrients. However, in this study, the validation correlation coefficients for some macronutrients were lower (r < 0.3), whereas those for certain micronutrients, such as vitamin D and choline, were higher (r = 0.4–0.5). After adjusting for energy intake, the correlation coefficients for carbohydrates and proteins showed a slight increase, from ∼0.3 to 0.4, whereas cholesterol’s correlation coefficient rose to 0.67. However, the correlation coefficients for fat remained insignificant. This may be attributed to energy adjustments reducing between-subject variation, which could result in lower or smaller correlations [21].

Cross-classification was used to assess the strength of agreement between the 2 methods by ranking participants according to their intake. Although most studies use quartiles, this study ranked participants in quintiles. We found that 28% had exact agreement, 61% had exact or ±1 adjacent agreement, and only 3% were grossly misclassified.

In reviewing the literature, most reproducibility studies administered the second FFQ between 1 mo and 1 y after the first. It has been suggested that the second FFQ should be administered within a time frame that prevents participants from recalling their previous answers, with a maximum of 1 y to avoid attenuating reproducibility [9]. For this study, participants were asked to complete the second e-FFQ after 3 mo. The correlation coefficients for reproducibility for all nutrients ranged from 0.6 to 0.8, which falls within the acceptable range of 0.5–0.7, indicating good reproducibility of the e-FFQ [28,29].

Cross-classification into quintiles showed that the e-FFQ demonstrated high reproducibility, with all nutrients having exact ±1 adjacent agreement ranging from 67% to 89%, and gross misclassification was minimal at just 1%. This strong reproducibility is likely due to the short interval between the first and second e-FFQ administrations, as shorter time frames typically result in higher reproducibility, which tends to decrease over longer periods [9,28]. Participants reported higher nutrient intake estimates on the first e-FFQ, which is consistent with findings from other reproducibility studies [9,28,29]. These higher estimates may be attributed to learning effects or participant fatigue after completing the first FFQ and 4 24-h recalls [9,29].

Limitations

The sample was not fully representative, with only 19% male and 9% Indo-Trinbagonian participants. Although 91 participants completed both e-FFQ administrations and FRs, 11 were excluded due to energy intake outliers, reducing the sample size to 80 and potentially affecting the validity of the e-FFQ. The a priori power analysis indicated that 84 participants were needed, so this reduced sample size impacted the ability to detect between-person variability.

Future research

The dietary habits and nutritional patterns of individuals living in Trinidad & Tobago remain underexplored, with limited research available on their specific dietary intakes. Existing studies utilizing FFQs often lack cultural specificity or proper validation for the local population [[30], [31], [32]]. The current study’s use of e-FFQ and FRs presents an opportunity for further analysis, particularly in examining associations between diet and NCDs. Alongside dietary data, demographic information—including age, ethnicity, education, employment, physical activity, and self-reported NCDs—was collected, providing valuable context for understanding dietary influences on health.

The study recommends several areas for deeper investigation, including:

• 1.The relationship between dietary trends and NCDs.
• 2.Macronutrient consumption and dietary patterns among the 2 major ethnic groups.
• 3.Socioeconomic disparities in diet.
• 4.Links between food groups and cancer risks in Afro and Indo Trinidadians.
• 5.Cultural and religious impacts on dietary habits.
• 6.Nutritional contributions of street foods.
• 7.Associations between street food consumption and cardiovascular/metabolic diseases.
• 8.Physical activity, dietary habits, and awareness of heart disease risks.
• 9.Impacts of nutrition policies on public health and NCD prevention.

To enhance future research, it is suggested to expand sample sizes and improve representation of males and various ethnic groups, whereas also extending studies to include children, adolescents, and pregnant women. Additionally, refining the e-FFQ by removing low-intake items and including some more commonly consumed sweetened beverages could improve its relevance and accuracy.

In conclusion, the culture-specific e-FFQ demonstrated good reproducibility and validity for assessing nutrient intake in Trinidad and Tobago, comparable with other Caribbean FFQs. It is an effective tool for evaluating dietary intake and its association with NCDs. The study also found moderate correlations for several micronutrients linked to NCDs and maternal health, highlighting the e-FFQ's potential for use in future research to determine dietary and health associations.

Author contributions

The authors’ contributions were as follows – LAF-N, GS: designed research; LAF-N: conducted research; DS, LAF-N: analyzed data; LAF-N, PAD, CH, GS: wrote the paper; LAF-N: had primary responsibility for final content; and all authors: read and approved the final manuscript.

Data availability

Data described in the manuscript, code book, and analytic code will be made available on request pending application.

Funding

The authors reported no funding received for this study.

Conflict of interest

The authors report no conflicts of interest.

Appendix A. Supplementary data

The following are the Supplementary data to this article: