The present analysis was carried out in the context of a parent study titled “Evaluation of dietary practices and assessment of nutritional status and associated risk factors for metabolic syndrome in the young adult population of Rwanda”.

Study setting

Rwanda is a country of 10,169 square miles (26, 338 km2), with a 13,246,394 total population by end-August 2022 [19, 20]. Rwanda is divided into four provinces (Eastern, Northern, Southern, and Western), and Kigali City. Only 28% of Rwandans live in urban areas, and the young population (below 30 years) represents 65.3% [21]. The country is landlocked, with a landscape made of small hills, most topped by a village and enjoys two rainy seasons which makes it a mild climate in general. The light rainy season starts in September and ends in December, while the-heavy one starts in February and ends in April. In the fiscal year 2021–2022, the national per capita Gross Domestic Product (GDP) was 907.0 USD [20]. About 69% of Rwandan households are engaged in agricultural activities in terms of crop production and/or animal husbandry. Despite the large economic growth of the country, food systems in Rwanda remain rural and traditional [1]. And most of the households are food insecure, a result of the high population density, poverty, and the dependence of poor rural households (83% of a total number of households) on subsistence farming on small and low productive lands [13, 22, 23].

Sampling strategy

The main sampling frame for this survey was based on data and cartographic materials from the Rwanda 2012 Census [24]. The primary sampling units were the census enumeration areas (EAs), which are small operational areas, with the boundaries and landmarks clearly identified. Rwanda is divided geographically into 4 Provinces and Kigali City, 30 districts, 416 sectors, 2,148 cellules and 14,837 villages. A village may have one or more EAs. In the case of this study, a sampling frame of villages was used, and the primary sampling units were the individual villages.

Study design and period

The cross-sectional survey was conducted from end January 2023 to early April 2023 in rural, semi-urban, and urban communities of Rwanda.

Study participants and inclusion and exclusion criteria

Adults, reportedly aged from 18 to 35 years, from Kigali (the capital) and the four regions of Rwanda participated in the survey. Participants were from both sexes and were included in the study after providing an informed consent. They were eligible if they were permanent resident in the selected study areas for at least three years, registered in the village office, accepted home visits, and willing to visit the health center for blood sample collection. Participants who suffered from any chronic disease, had physical or mental disability, or were pregnant or lactating (for women) were not included in the study.

Sample size estimation

The estimated minimum sample size for household survey was 225 per region, following the formula of the General Household Survey [25, 26].

$$n = \right)} \mathord \right)} }}} \right.\kern-\nulldelimiterspace} }}$$

Where:

? = sample size.

? = 0.50 (recommended when prevalence is unknown), baseline level of the selected indicator.

? = 1.96 (at 95% Confidence Interval).

? = 0.0653 (adjusted margin of error to achieve 230 households per province).

The estimated required sample size was therefore adjusted to the five regions: N = 225*5 = 1125. Assuming a non-response rate of 8%, the final sample size was adjusted upward to 1223 households. We aimed to include 15 EAs per region. Overall, 82 EAs, including 1,218 households, were selected.

Socioeconomic and dietary data collection

Data related to participant’ socio-demographics (age, sex, marital status, literacy, education, religion) and economics (household ownership and assets, income, and socioeconomic category) were collected after eligibility criteria were met and the participant consent was obtained. The socioeconomic (Ubudehe) program classifies households into four numeric categories (I to IV), ranging from the poorest to the wealthiest [27]. Dietary intake data were collected using a semi-quantitative FFQ which was developed and validated for the purpose of the present study.

The FFQ included 113 food items that are commonly consumed in Rwanda. These foods were listed in the order of their categories that included (1) cereals and cereal products; (2) starchy roots and tubers; (3) sugar and sweets including soft drinks; (4) milk and dairy products; (5) eggs and their products; 7) fish and seafood; 7) meat and their products; 8) legumes, nuts, and seeds; 9) green leafy vegetables; 10) other vegetables and their products; 11) fruits; 12) oils and fats; 13) deep-oil fried foods, 14) salt and other salted products; 15) spices and condiments; and 16) alcoholic drinks. The FFQ considered the recall time frame of the previous year assuming that the consumption patterns of participants over one year could be considered as their usual food intake. The participant was asked whether and how frequently they consumed the food as the following: (1) never, (2) once to twice per year, (3) once to twice on the 6 months, (4) one to twice on the 3 months, (5) once monthly, (6) 2–3 times per month, (7) once per week, (8) two to three times per week, (9) four to six times per week, (10) once per day, and 11) more than one time per day [28]. If the respondent indicates that they consumed the food item, they were then asked to report the average serving size in comparison to what they perceive as the average intake within the community, categorized as small, medium, or large.

All data were collected digitally and managed using REDCap electronic data capture tools hosted at the Ghent University [29, 30], offline using the REDCap Mobile App [31].

Pilot study – food frequency questionnaire development and food list validation

Sample size and participants. A pilot study was carried out before the present study to validate the list of food items used in the FFQ. One hundred participants, from both sexes were selected in urban Kigali and the same number of participants from rural communities from the four regions of Rwanda. Participants in this pilot study were acquainted with cooking practices to be able to list all the ingredients that were used in the preparation of the dishes/meals (e.g. type of oil used, and other ingredients). Additionally, participants were 18–55 years old, and not pregnant or lactating for women because of the different nutritional requirements and dietary intakes. Pilot study participants were randomly selected based on their respective household’s registration at village level.

Pilot study procedure. The pilot study validated a semi-quantitative FFQ that assessed the dietary intake of adults in Rwanda, during one year, using the multiple-pass 24-hour recall method [32]. Briefly, two FFQs were administered to all the participants, one on the enrollment day and the second, after one year. In four consecutive rounds, every three months, two multiple-pass 24-hour recall data were collected on weekdays and on weekends. The FFQ administered contained a list of 167 food items and was developed in consultation with local researchers who are familiar to the foods and meals in the different regions of Rwanda.

The multiple-pass 24-hour recall data were collected on the same participants following the standard method described by Gibson and Ferguson [33]. The participants were asked in the first pass to list all solid and liquid foods, beverages, and mixed dishes they had consumed during previous 24 h, starting from previous morning to the morning of interview. In the second pass, the participant was asked to provide details on the processing and the cooking method of the foods, beverages, and mixed dishes. In the third pass, the quantities of the consumed foods, beverages, and mixed dishes were estimated by different methods including direct weight, standard unit (such as bottle of 1 L or a pack of 500 g), and proxies including rice and water. Detailed information on prepared mixed dishes that were not already collected as a standard recipe, were collected from the participant during the non-standard recipe pass. Finally, in the fourth pass, the list of reported foods, beverages and mixed dishes was reviewed with the participant to make sure that no food was mistakenly added or no food was consumed but forgotten by the participant.

FFQ food list validation. A list of foods consumed by the study participants in the study area was constructed based on the foods reported in the multiple pass 24-hour recall data. Mixed dishes were further mapped to their ingredients and the list was added to the food list. This resulted in a dataset containing lists of foods consumed by all the participants, along with their frequency of consumption recorded over the four rounds of recall interviews.

The list thus developed included 127 food items with their frequency of consumption. The list was compared with the food lists of the two FFQs in order to have an exhaustive list of commonly consumed foods. The retention in, the addition to, or the exclusion from the final FFQ food list was based on the following criteria: (1) retaining food items that are consumed by a large proportion of people; (2) adding raw food items that have been listed in the multiple pass 24-h recall, by a large proportion of participants; and (3) dropping foods that do not provide any nutrient/energy (water).

Data processing and statistical analysis

Data were exported from REDCap to Stata for analysis (STATA/IC, Version 15 [34]). Asset index was constructed based on the information on baseline ownership of the house and of a set of assets (radio, television, telephone, computer, refrigerator, bicycle, light source, cooking material, or water source) and house characteristics (cooking place, exhaustion fan, toilet, number of rooms) using multiple correspondence analysis [35]. Descriptive statistics were presented as means and standard deviations (mean ± SD) or proportions for continuous and categorical variables, respectively.

Further, food intakes as collected by the FFQ were summarized using two indicators: the GDQS and dietary patterns.

Development of food consumption score that combines frequency and serving size. In a first step, the consumption frequency of each food was considered. The consumption frequencies were numbered, and these numbers were considered as consumption scores, as follows: Never = 1, 1–2 times per year = 2, 1–2 times per six months = 3, 1–2 times per three months = 4, once per month = 5, 2–3 times per week = 6, once a week = 7, 2–3 times per week = 8, 4–6 per week = 9, once a day = 10, and more than once a day = 11. In order to have fewer categories, we recoded the frequency of food intake as follows: respondents who reported never consuming a food item were assigned a score of 0. Those who reported consuming it 1–2 times per year, 1–2 times per six months, or 1–2 times per three months were assigned a score of 1. Those who consumed a food item once per month, 2–3 times per month, or once a week were assigned a score of 2, while those who had that food 2–3 times per week, 4–6 times per week, or at least once daily received a score of 3. Moreover, we assigned scores of 1, 2, or 3 based on the reported portion sizes consumed, representing small, medium, or large portions, respectively. When the frequency score is multiplied by the portion size score, each food group can be assigned a score ranging from a minimum of zero to a maximum of nine. The portion sizes in the FFQ were not adjusted based on the portion sizes collected from the repeated multiple-pass 24-hour recalls.

Global Diet Quality Score. We used the recently developed food-based indicator, GDQS, that captures dietary risk of non-communicable diseases [36]. The GDQS is a validated novel food-based metric of diet quality based on the Prime Diet Quality Score (PDQS) [37]. The GDQS includes 25 food groups, 16 of which are classified as healthy, 2 as unhealthy in higher amounts, and 7 as unhealthy [36, 37]. Healthy food groups are assigned more points for higher consumption (0 points for food consumption scores ranging between 0 and 2, 1 point for food consumption scores 3–5, and 2 points for food consumption scores 6 and above). Scoring is reversed for unhealthy groups (more points are given for lower consumption). Healthy food groups included citrus fruits, deep orange fruits, other fruits, dark green leafy vegetables, cruciferous vegetables, deep orange vegetables, other vegetables, legumes, deep orange tubers, nuts and seeds, whole grains, liquid oils, fish and shellfish, poultry and game meats, low-fat dairy, and eggs. Unhealthy foods in excessive amounts included high-fat dairies and red meat. Unhealthy foods included processed meat, refined grains and baked goods, sweets and ice-cream, sugar-sweetened beverages, juices, white roots and tubers, and deep-fried foods. Our questionnaire did not capture deep-orange tubers and low-fat dairy products for their consumption was not common as indicated in the developed FFQ. We considered deep-fried foods as an unhealthy food without distinction of whether they were bought or cooked at home. Solid-fat and spreads, alcohols, and tea and coffee were not scored following the validated method of Bromage et al. [37].

In the second step, the GDQS was calculated by summing the scores of 23 food groups consumed by the participant [37]. Consequently, GDQS ranged between 0 and 45. Further, the classification of GDQS was based on the classification suggested by Bromage et al. as follow: lowest GDQS (scores lower than 15), medium GDQS (GDQS ranging between 15 and 23), and the score of 23 or more (≥ 23 GDQS) was considered as high GDQS [37].

Dietary patterns. Scores of frequency of consumption of the food items were assigned to each food item as described above. From initial analysis of the pilot study, food groups included cereals, legumes, dairy, roots and tubers, fruit, vegetables, fat and oil, deep-fried foods, meat, fish, eggs, alcoholic drinks, soft drinks, fruit juice, tea and coffee, sweet foods, and herbs and spices [38]. In this analysis, few foods with large consumption or eaten as they are remained as such (such as pizza and pasta). Added salt and sugar were also considered separately. This resulted in a total of 26 food groups and items that included cereals, rice, bread, pasta, pizza, chapatti, roots and tubers, fruit juice, energy and sweet drinks, jam, cakes and cookies, added sugar, coffee and tea, dairy, eggs, fish, meat and organs, plant-based proteins including legumes and tofu, seeds and nuts, green leafy vegetables, fruits, oils and fats, deep-fried foods, salt and bouillon cube, spices, and alcoholic drinks. Within each food group, individual consumption scores for foods were determined based on their proportionate contribution to the total score of the group. For instance, if the total score for grains in the respondent intake data is Y, and the cumulative score for sorghum is X, the sorghum score would be calculated as the initial score multiplied by the ratio X/Y. Scores for food groups or individual items remained within the range of 0 to 9.

Exploratory factor analysis (EFA) with the orthogonal varimax rotation was employed to derive dietary patterns [39]. EFA helps to combine food groups based on the degree to which they are correlated with one another, with the aim of identifying food groups that account for the largest amount of variation in diet between individuals, and hence reduce the complexity of the data. The number of factors extracted was 12 and the number of parameters was 246.

To calculate factor scores, the sum of the products of factor loading coefficients for each food group and item and the retained factors was computed. Promax rotation was employed to optimize the factor structure and enhance interpretability. The Kaiser-Meyer-Olkin (KMO) test was used to measure the sample adequacy, with a KMO value above 0.50 considered satisfactory. Additionally, the Bartlett test of sphericity was used to assess correlations among food groups, following a Chi-Square distribution and significance at a p-value less than 0.05 [40]. For all participants, a total of 12 factors were retained. Factors contributing to the pattern were selected based on criteria, requiring eigenvalues greater than 1 and rotated factor loadings of at least 0.3. Consequently, three factors were chosen for all participants. The total variance explained was calculated.

Statistical analyses. The number of total subjects that consumed each of the food groups was calculated. The proportion of urban, semi-urban and rural, female and male, high and low socio-economic status, Kigali and the other regions, social categories, educational level, and age groups in the three GDQS categories, high, medium and low, were also calculated through cross-tabulation. Statistical differences between groups were tested using logistic regression. Similarly dietary patterns were assessed in the whole population and by group. Differences were tested using analysis of variance, ANOVA. Post hoc analysis was conducted using Bonferroni test.

Finally, we conducted multiple linear regression to assess the associations between the three dietary patterns and participant’s residency (rural, urban or semi-urban), age and sex, province, socio-economic status, social category, and education level. A p-value of less than 0.05 was considered statistically significant.