Association between nutrition literacy and diet quality among adolescents and young adults in the rural district of Mayuge, Eastern Uganda | BMC Public Health

Study design and setting

This was a population-based cross-sectional study conducted as part of the ARISE-NUTRINT (Africa Research, Implementation Science, and Education – Reducing nutrition-related NCDs in adolescence and youth: Interventions and policies to boost nutrition fluency and diet quality in Africa) initiative. Through population-based surveys, the ARISE-NUTRINT seeks to enhance understanding of dietary and physical activity-related risks for NCDs among AYA in seven SSA countries of Uganda, Tanzania, Burkina Faso, South Africa, Ethiopia, Nigeria, and Ghana, with partners from Europe and North America [23, 24]. However, this paper focuses on data from the Uganda site only, collected within the Iganga-Mayuge Health and Demographic Surveillance Site (HDSS). This HDSS covers three predominantly rural districts of Iganga, Mayuge, and Bugweri in Uganda. Under surveillance since 2005, this HDSS has 18,634 households and 120,000 residents, with AYA comprising 27% of the population [25].

Study population

The study population was AYA aged 10–24 years, sampled from the Iganga-Mayuge HDSS database, with a total of 40,000 AYA residing in Mayuge, Iganga, and Bugweri districts [25]. All AYA within the database and residing in Mayuge district were eligible for inclusion except those with chronic illnesses, which could prevent them from adequate participation, for example, those unable to speak, hear, or cognitively impaired.

Sample size estimation

Anemia was selected as the key outcome for estimating the sample because this condition is among the top five causes of morbidity in both male and female adolescents in SSA, with its prevalence ranging from 20 to 60% [26, 27]. Drawing from previous ARISE Network studies [24, 28, 29], it was assumed that enrolling at least 1,000 participants per country would be sufficient to detect meaningful differences in anemia rates. To account for an expected 20% loss to follow-up [27, 30, 31], the target sample size was increased to 1,200 adolescents. This sample size would ensure 80% statistical power to detect at least a 10% absolute difference in anemia prevalence between any groups compared. The power calculation was done assuming a baseline prevalence of 50%, a conservative approach because it provides the highest variability. Therefore, the design would still have adequate power across different prevalence levels within the 20–60% range.

Sampling procedures

From the Iganga-Mayuge HDSS database, eligible AYA were randomly selected using household-based simple random sampling within each age-sex stratum (10–14, 15–19, and 20–24 years; male and female), applying Probability Proportional to Size allocation to each stratum. In households with more than one eligible AYA, only one was randomly selected to participate using a within-household randomization procedure to minimize intra-household correlation. A list of sampled AYA was generated, including locator information for their households. Those selected were then approached in their households using this locator information, with support from community guides.

Conceptual framework

Analysis for this study was premised on a framework adapted from Nutbeam’s Model of Health Literacy. In this model, nutrition literacy is conceptualized as a sub-component of health literacy, which, in turn, is a component of the broader general literacy [32,33,34]. Nutrition literacy is further categorized into three dimensions: FNL, INL, and CNL [34] and previously in Uganda, this three-level categorization was applied to assess nutrition literacy status of adolescent students in Kampala city [35]. While this previous study used a total of 29 attitude statements [35], the present study refined this approach by adapting 22 statements to measure nutrition literacy among the AYA in rural Mayuge District in Uganda, considering contextual differences, such as lower formal education access, socioeconomic status (SES) as well as reducing time and cost during data collection to meet feasibility requirements. Nutrition literacy was hypothesized as a key factor influencing food choices and eating behaviors [12, 19, 21, 22]. This highlights its role in shaping overall diet quality among AYA, presented in supplementary information: Fig. 1.

Measurements of study variables

The current study utilized a survey questionnaire for the ARISE-NUTRINT initiative. This tool was translated into Lusoga, the common language spoken in Mayuge District, and uploaded onto Open Data Kit for electronic data collection. Trained research assistants conducted one-on-one, face-to-face interviews between April and May 2024. During data collection, the research team conducted daily supportive supervision and spot checks to monitor the performance of research assistants and promptly address any issues. Identified inconsistencies were resolved through immediate feedback and refresher training to reinforce proper procedures and uphold data quality. Additionally, data underwent iterative cross-checks and reviews for completeness, and participants were re-contacted when necessary to fill in missing information.

Diet quality (outcome variable)

In the present study, diet quality was defined as the overall healthfulness of an individual’s diet, categorized into two groups: (i). Healthy category including foods which are diverse in nutrients, healthy, or those whose consumption may reduce the risk of diet-related NCDs. (II). The unhealthy category comprises foods that are less diverse in nutrients and those that are unhealthy or whose consumption may increase the risk of diet-related NCDs.

The Global Diet Quality Score (GDQS) is a food-based index structured as a food-frequency questionnaire, was adapted to measure diet quality [36, 37]. This scale was previously validated in 10 SSA countries (including Uganda), India, China, and the United States of America [37, 38]. The GDQS assess overall diet quality in relation to both nutrient adequacy and risk of NCDs, for instance, the score reflect risks for the triple burden of malnutrition: undernutrition (wasting, underweight and stunting), micronutrient deficiency (anemia, increased infection rates), and overnutrition (overweight, obesity and associated NCDs like CVDs and T2DM) [37, 38]. Previously, higher GDQS scores were reported to be associated with improved nutrient adequacy, better mid-upper arm circumference, and reduced anemia risk among adolescents in SSA, including Uganda [38].

Although the GDQS was validated using food amounts/servings (grams/day) for cross-country and temporal comparisons, the present study adapted this index to measure diet intake using frequency rather than quantity, as it was done in past studies in SSA due to the lack of standardized serving sizes in rural, low-income settings [29, 39]. Internal consistency of the GDQS scale was assessed, and a Cronbach’s alpha coefficient of 0.75 was obtained, indicating acceptable reliability [40].

Based on the GDQS, food consumed was classified into 25 food groups, split into 17 nutrient-rich or healthy, and eight less diverse in nutrients or unhealthy groups. The nutrient-rich and healthy category included dark green vegetables, legumes and nuts, and seeds, fresh fruits, cruciferous vegetables, deep orange vegetables, deep orange tubers, and other vegetables, lean meat, white meat, eggs, low and high-fat milk, whole grains, boiled snacks, matooke, root, and stem tubers). On the other hand, less diverse in nutrients or unhealthy food groups included unprocessed red meat, deep-fried foods, snacks high in salt or oils/fat, refined grains like wheat and white maize flour, processed meat such as sausage, sugar-sweetened beverages and sweets and ice cream, and fruit juice with added sugar. Although white roots and tubers like cassava are high in calories, they are less in micronutrients and contain micronutrient absorption inhibitors [41, 42]. For these reasons, white roots and tubers were included in the unhealthy or less nutrient category.

High-fat milk and dairy products were classified as healthy food groups at all levels of consumption. This was an adaptation of the original GDQS, which categorized high consumption (≥ 3 times per day) of high-fat dairy as unhealthy and scored it negatively [37]. In the context of the rural, low-income setting of the current study, both low- and high-fat dairy were considered healthy, given the important role dairy plays as a nutrient source for adolescents [29].

Participants were asked how often, on average, they had consumed each food group, per week over the past one month and, responses were recorded as 0–1/week, 2–3/week, or 4+/week were used to score diet quality according to food groups consumed more or less frequently. Following the GDQS scoring guidelines, lower points were assigned for less frequent consumption of healthy food groups, while higher points were assigned for less frequent consumption of unhealthy food groups [37]. For example, from the healthy foods; dark green vegetables, legumes and nuts, and seeds, fresh fruits: scored as 0–1 (0 points), 2–3 (2 points), ≥ 4 (4 points); and from the unhealthy groups; deep-fried foods, snacks high in salt or oils/fat, refined grains, processed meat, juice, white roots and tubers. Sugar-sweetened beverages and sweets: 0–1 (2 points), 2–3 (1 point), ≥ 4 (0 points). The detailed scoring plan used is presented as Supplementary Information: Table 1.

For each participant, diet quality scores were summed to obtain an overall GDQS (0 to 49) with higher scores indicating better diet quality. Participants with a GDQS < 15 were classified as having poor-quality diets (associated with increased risk of nutrient inadequacy and higher risk of diet-related NCDs) [37]. Those with scores ≥ 15 were classified as not having poor-quality diets (associated with reduced risk of nutrient inadequacy and lower risk of diet-related NCDs) [37]. This binary categorization of diet quality was considered because it matches with population-level nutrition interventions aimed at improving diets from poor quality to healthy (not poor), and it simplifies interpretation, and facilitates clear messaging to nutrition policymakers and program implementers [43, 44].

Nutrition literacy (independent variable)

In the present study, nutrition literacy (with its three dimensions) was defined as the ability of a participant to access nutrition information and possession of knowledge and a positive attitude to apply the information to make healthy dietary choices. A structured questionnaire, based on the Adolescent Nutrition Literacy Scale (ANLS) [35]. This scale comprised 22 statements measuring FNL, INL, and CNL dimensions across knowledge, attitude, and practices (KAPs) domains. Each item on the ANLS was rated on a five-point Likert scale and divided as follows: FNL: 7 items (score range: 0–7), INL: 6 items (score range: 0–6), and CNL: 9 items (score range: 0–9). The ANLS was developed in Uganda [35], but it has been adapted and applied in Turkey [19, 45,46,47], and North Africa and other Arab countries in Asia [48].

Although the original development study reported a Cronbach’s alpha of 0.54 [35], which falls slightly below the conventional acceptability threshold of 0.60 [40], the adapted items of the ANLS had acceptable reliability (Cronbach’s alpha coefficient of 0.70) [40]. Since Likert scales measure ordinal data with unequal intervals between responses, all items were categorized (scored 0 or 1) to avoid misinterpretation that can arise from treating them as continuous variables [49]. Besides, 12 of the 22 statements on the ANLS are intentionally reversed to improve respondent attention, encourage thoughtful engagement, and reduce acquiescence bias [35], thus enhancing the reliability of scale. Direct statements, for instance, “I am familiar with the concept of a balanced diet,” were scored 0 for strongly disagree, disagree, and neither agree nor disagree, and 1 for agree and strongly agree. Reversed statements for example “I find it difficult to know how I should change my diet when I get dietary advice from the doctor, nurse, or the like” were scored 1 for strongly disagree and disagree, and 0 for neither agree nor disagree, agree, and strongly agree as shown in the supplementary information: Table 2.

Median was used to summarize FNL, INL, and CNL subscales instead of mean, as it respects the ranking of the Likert scores without assuming equal intervals [49]. Overall nutrition literacy was computed by summing subscale scores, with possible totals ranging from 0 to 22 (higher scores indicating higher literacy). Due to limited literature, total scores were categorized into low (≤ 7), moderate (8–14), and high (15–22) based on tertiles, adapting previous studies which used data-driven cut-offs in the absence of validated thresholds [35, 50].

Confounding variables

In addition to nutrition literacy, potential confounders like household socio-economic status (SES), household food deprivation (past 30 days), phone use, and social media use were also recorded during the interviews. Social media use was assessed by asking respondents if they had an account on any of Facebook, WhatsApp, TikTok, or Telegram (yes/no), and those with at least one account on any of these channels were categorized as using social media. Besides, individual ownership of a phone or using a parent’s, neighbor’s, or friend’s phones was reported as phone use. Household food deprivation was assessed using a question, “In the past 30 days, was there ever no food to eat of any kind in your house because of a lack of resources to get food?” with Yes/No as the response options.

The MacArthur Scale of Social Status was adopted to measure the SES of AYA. This scale depicts social status as a 10-rung ladder, where individuals perceive social status within their community and society as a whole [51] (supplementary information: Fig. 2). Participants were asked to state the rung on the ladder on which they felt their family stood at the moment during data collection. On a scale of 0–10, higher scores indicated better SES within their community or society at large (more resources, respect, or influence), where 1–3 was categorized as low, 4–6 as mid, and 7–10 as high SES [51].

Statistical analyses

Data analysis was conducted using STATA version 15.0. To characterize the AYA included in the study, we used frequencies and percentages for ordinal and nominal data and the mean with standard deviation or median plus inter-quartile range (IQR) for the Likert scores based on the normality distribution for the transformed score variables. A chi-square test was run to determine the relationship between the outcome variable and each of the independent variables. Considering diet quality as a binary outcome categorized into poor quality (coded as yes or 1, and not poor coded as no or 0), logistic regression was used to determine the crude associations between diet quality and nutrition literacy. Separate bivariate logistic regression models were run between the binarized outcome and each of the other independent variables, including age, sex, parental education level, phone use, and social media use.

Multicollinearity was assessed using the Pearson correlation coefficient (r), and when r > 0.4, only one of the correlated variables was retained based on its stronger theoretical or empirical justification (for example, known association with diet quality in prior literature) or better model fit. The multivariable logistic regression model was built using a backward stepwise process, where the least significant covariate based on p-value was removed at each step. Statistical significance P < 5% and adjusted odds ratio) were used to determine the magnitude and direction of the association between diet quality and nutrition literacy. Afterwards, non-significant covariates were individually adjusted in the multivariable model to determine their adjusted odds ratios.