Development of the FFQ

The FFQ was developed using food items recorded in a single 24-hour diet recall of adults from the Malaysian Adult Nutrition Survey (MANS) 2014⁽¹⁴⁾ and adolescents from the Adolescent Nutrition Survey (ANS) 2017.⁽¹⁵⁾ Both were cross-sectional, nationwide surveys utilizing cluster sampling and proportionate allocation to ensure population representativeness. These dietary recalls provided granular data on brand names, meal occasions, and preparation methods. Respondents estimated portion sizes using a food album and standardized household measures, including cups, spoons, and bowls. Energy and nutrient intakes were calculated primarily using the Malaysian Food Composition Database. We extracted a total of 55,000 food items and categorized them based on their primary ingredients, nutrient profiles, and conceptual similarities. To ensure the inclusion of major nutrient contributors across Malaysia’s diverse ethnicity, we ranked food items separately for Malay, Chinese, Indian, and the indigenous groups of Sabah and Sarawak. We calculated the percentage of micronutrient consumption for each food group using Block equations. Ultimately, food items contributing at least 90% to the total intake of energy and selected nutrients were selected for the final FFQ food list.^{(Reference Block, Hartman, Dresser, Carroll, Gannon and Gardner17–Reference Block, Dresser, Hartman and Carroll19)}

Content validity was established through an expert panel’s evaluation of the FFQ layout and food list. The expert panel included researchers with extensive experience in FFQ development, portion-size estimation, and dietary assessment within national surveys and cohort studies. To account for Malaysia’s diverse culinary landscape, we included sub-type items to capture variations in cooking methods, gravies, and sweeteners. Furthermore, we explicitly listed items of public health significance, such as whole grains, legumes, and sugar-sweetened beverages (including packaged, premixed, and artisanal versions) as separate entries. At the request of stakeholders, common fruits and vegetables were also itemized.

The draft instrument underwent pretesting to identify usability issues and assess respondent burden. The finalized FFQ comprised of 147 items categorized into 13 groups: (a) cereals and products, (b) meat and products, (c) fish and seafood, (d) eggs, (e) legumes and products, (f) milk and dairy, (g) vegetables, (h) fruits, (i) confectionery, (j) beverages, (k) soup or gravies, (l) spreads, and (m) seasonings and flavourings (Appendix). To facilitate visual portion-size estimation, we developed a corresponding food photo album. Furthermore, a comprehensive nutrient database was compiled using data from the Malaysian and Singaporean Food Composition Databases, supplemented by the Atlas of Food Exchanges and Portion Sizes. The integrated workflow for the FFQ development is illustrated in Figure 1.

Figure 1.

Development of the semi-quantitative food frequency questionnaire.

Data collection for validation study

This FFQ validation study employed a cross-sectional design. We recruited approximately 160 government community nutritionists from 13 states and three federal territories to serve as interviewers. To ensure high data quality and inter-rater reliability, we conducted decentralized training sessions by geographical zone. All interviewers were provided with a comprehensive standardized manual containing the study protocol, structured questionnaires, food coding systems, and detailed food descriptions. Training sessions included standardized briefings, mock interview exercises, and competency assessments to ensure uniform administration of the FFQ and 3DRR. We enrolled a total of 400 participants, comprising 200 adolescents and 200 adults. Respondents were selected using a purposive sampling strategy with proportionate allocation based on age and ethnicity to ensure a representative cross-section of Malaysia’s major ethnic groups (Malay, Chinese, Indian, and the indigenous populations of Sabah and Sarawak). Purposive sampling was employed over random sampling due to its feasibility in reaching specific ethnic and age group quotas. Interviews were conducted in-person at the respondents’ homes or workplaces to ensure a comfortable and private environment. Eligible participants included healthy Malaysian citizens aged 10–59 years. We excluded individuals who were pregnant, illiterate, or currently following a medically prescribed diet. Furthermore, those with serious chronic illnesses or physical conditions that could significantly alter habitual dietary patterns were also excluded from the study.

Consistent with established dietary assessment protocols for paediatric populations, interviews for children aged 10–14 years were parent-assisted. Research suggests that while children begin to self-report reliably around age 10, parental input remains essential for quantifying portion sizes and identifying ingredients in complex mixed dishes.^{(Reference Baranowski and Domel20)} Furthermore, the interviewer-administered format allowed trained nutritionists to guide the younger respondents through the survey, providing clarifications and using visual aids to reduce respondent burden and maintain engagement.^{(Reference Murtas, Krogh and Intemann21)} Although the FFQ food list were collected from school-based surveys of adolescents aged 13–17 years, the age range was expanded to 10–17 years to align with the World Health Organization’s (WHO) official classification of adolescence. This ensures the tool’s utility across the full developmental spectrum of the adolescent population in Malaysia.

Data collection proceeded in two distinct phases between September 1 and October 1, 2021. First, nutritionists administered the 147-item FFQ to capture habitual intake frequency, portion sizes, and cooking methods over the past month. To facilitate accurate portion-size estimation, interviewers utilized standardized household measures (cups, bowls, and spoons) alongside a dedicated food photo album. Two days after the FFQ administration, the second phase began with the initiation of the three-day dietary record and recall (3DRR), which served as the reference method. Respondents recorded their dietary intake over three non-consecutive days (two weekdays and one weekend day), providing detailed descriptions of food items, brands, and preparation methods. These records were subsequently reviewed by nutritionists during a follow-up recall session to ensure accuracy and resolve any ambiguities on portion sizes and food items. Additionally, anthropometric measurements were performed by nutritionists at the clinics. All completed records were audited by regional supervisors for completeness and coding accuracy before being centralized for nutrient analysis (Figure 2).

Figure 2.

Food frequency questionnaire (FFQ) validation process flowchart.

Analysis of nutrients

Nutrient intake for each participant from the 3DRR was analysed using Nutritionist Pro™. For the FFQ, daily intake for each food item was derived by converting the reported consumption frequency into a daily equivalent using standardized conversion factors. We then calculated the nutrient content for each item by multiplying its daily intake (in grams) by the nutrient values obtained from the Malaysian Food Composition Database and the Singapore Food Composition Database. For food items absent from these databases, nutritional labels and standardized recipes served as supplementary references. We excluded nutrients from dietary supplements to ensure the analysis reflected only habitual food intake. Finally, total daily energy and nutrient intakes were determined by summing the values across all 147 food items. This process ensured a comprehensive and consistent comparison between the FFQ and the 3DRR reference method. The formulas used were:

For each nutrient of each respondent,

Data preprocessing began with a rigorous cleaning phase to ensure the integrity of the dietary records. To minimize the underreporting and data missingness inherent in self-recorded diaries of the preceding day, each dietary record was reviewed via a face-to-face 24-hour recall interview to clarify portion sizes and identify omitted food items. As such, incomplete information on socio-demography and dietary intake were kept to minimal. Records with missing sociodemographic information were excluded from the final analysis to ensure a high-quality dataset. Maintaining complete profiles for all respondents was essential to establish a rigorous standard for comparison and to ensure the accuracy of the validation results. Any non-recorded intake was assumed as ‘zero-intake’. To ensure the physiological plausibility of the reported intakes, we identified outliers using the Epic-Norfolk method.^{(Reference Welch, Luben, Khaw and Bingham22)} The ratio of energy intake to basal metabolic rate (EI:BMR) below the 0.5th percentile or above the 99.5th percentile were classified as implausible and subsequently excluded. The BMR for each respondent was calculated using validated, gender-specific formulas for the Malaysian population.^{(Reference Ismail, Ng, Chee, Roslee and Zawiah23)} The comprehensive workflow of the validation study, from interviewer training to the final sample size (n = 361) available for analysis, is depicted in Figure 2.

Statistical analysis

Sociodemographic characteristics and body mass index (BMI) were described using frequencies and percentages for categorical variables and means with standard deviations (SDs) for continuous variables. Due to skewed distributions, energy and nutrient intakes from the FFQ and 3DRR were presented as medians and interquartile ranges (IQR; Q1, Q3). We utilized the Wilcoxon signed-rank test to assess differences in nutrient distribution between the two methods. Spearman’s correlation coefficients were calculated to estimate the relative validity of the FFQ against the 3DRR, with thresholds defined as poor (<0.2), acceptable (0.2–0.5), and good (≥0.5).^{(Reference Lombard, Steyn, Charlton and Senekal24)} To account for variations in total caloric intake, nutrient intakes were energy-adjusted using the residual method. In this approach, nutrient intake was regressed on total energy intake; the resulting residuals represent energy-adjusted values independent of total caloric consumption. An increase in the correlation coefficient after energy adjustment suggested the nutrient was significantly linked to total energy intake, whereas a decrease indicated that the initial correlation was likely inflated by energy consumption.

To evaluate ranking agreement, we performed a cross-classification of quartiles for energy and all nutrient intakes. Participants were categorized into quartiles to determine the proportion of individuals classified into the same, adjacent, or opposite (extreme) quartiles by both methods. In accordance with established criteria, a proportion of ≤10% in the opposite quartile was considered to represent good agreement.^{(Reference Lombard, Steyn, Charlton and Senekal24)} Bland–Altman plots were utilized to assess the level of agreement and the presence of systematic bias for energy and macronutrients. These plots displayed the difference between the two methods (FFQ - 3DRR) on the y-axis against their mean ((FFQ + 3DRR)/2) on the x-axis. The mean difference indicated whether the FFQ tended to over-or underestimate intake, while the limits of agreement (LOA) (mean ± 1.96 SD) reflected the degree of precision. A random distribution of data points around the zero-mean difference line indicated an absence of proportional bias, whereas consistent clustering above or below the line suggested that the methods were not interchangeable. All statistical analyses were performed using IBM SPSS Statistics version 23, and all statistical significance was set at p < 0.05.