A cross-sectional survey was conducted among adolescents who resided in six randomly selected areas of Dembecha Woreda, Northwest Ethiopia. Data were collected from 10 December 2021 to 20 January 2021. The study area consists of three agroecology zones (low-, mid-, and high-lands), which are conducive to diversified agricultural production.

Adolescents aged 10–19 years were the target population for the current study. These populations are overlooked for nutritional intervention in developing countries, including Ethiopia, as well as in the current setting. Adolescents with spinal curvature, who could not stand properly or walked with a wheelchair, did not participate in the study.

A single population proportion formula was used to determine the total number of adolescents in this study, based on the following assumptions: a 5% error margin, a 95% confidence interval, and a proportion of 44.6% for the consumption habit (P) of dark green leafy vegetables in southwest Ethiopia (25) with a design effect of 1.5. After adding a 10% non-response rate, the final sample size was 627.

The study area was purposefully selected due to its unique three agroecology zones (low-, mid-, and high-lands), which represent the northwest region of the country. Then, from each agroecology zone, two research settings, or Kebeles (the lowest administrative unit in the case of Ethiopia), were randomly selected as study areas for data collection.

Subsequently, from a randomly selected study setting, eligible households with at least one adolescent were registered, and a sampling frame was prepared. The number of adolescents assigned is proportional to the size (population) of a particular site. Finally, 357 adolescents from high-land, 77 from mid-land, and 188 from low-land agroecology zones were interviewed.

A pretested, structured interviewer-administered questionnaire was used for data collection. The tool was adapted from peer-reviewed articles and customized to the local context. Data on the dietary patterns of adolescents were collected using a validated food frequency questionnaire (FFQ) and contextualized to the setting over 1 week. To develop the FFQ assessment tool, a list of local food items that were consumed at different times during meals and snacks was collected by conducting a pilot study on 30 adolescents for 2 non-consecutive weekdays and 1 weekend day (27). The FFQ was organized in a semi-quantitative manner and administered using 24-h dietary recall methods.

Anthropometric data on adolescents’ weight and height were measured using a standard weight and height scale with precisions of ±0.1 kg and ± 0.1 cm, respectively, based on WHO protocol (28). During measurement, adolescents wore lightweight clothing without shoes to minimize error. Trained human nutrition graduates collected data through interviews and anthropometric measurements.

Data quality was ensured in all phases of the research activities. The tool was translated from English into an Amharic version (the local language). Training was provided to data collectors on the sampling procedures and interview techniques. The questionnaire was pretested using 5% of the total sample size at one site, which was not included in the actual study. In addition, before conducting actual dietary data collection from each sample, the FFQ was tested for internal reliability in 30 adolescents. The test results showed Cronbach’s alpha coefficient (α) of 0.79.

This result indicates that the FFQ is reliable for measuring the dietary patterns of adolescents. During the measurement of anthropometric weight and height data, calibration of the instrument was performed after each measurement. In addition, local food-colored pictures were used to minimize recall bias. The research lead investigator closely followed the day-to-day data collection process to ensure the completeness and consistency of the administered questionnaire.

Statistical analysis was carried out using SPSS version 25 and WHO Anthro Plus software. A descriptive analysis was conducted to characterize the data using frequencies, percentages, means, and standard deviations. Before conducting further analysis, extreme values and the normality distribution of continuous variables were checked.

Dietary patterns were determined using exploratory factor analysis (EFA). Kaiser–Meyer–Olkin (KMO; value of p > 0.05) and Bartlett’s test of sphericity (value of p<0.05) were used to check the adequacy of the sample-to-factors ratio. The existence of a correlation between food items was determined for each step of factor analysis (29). Then, EFA was run under orthogonal rotation with the varimax method to select interpretable and independent dietary patterns. Factors with a communality above 0.5 and an eigenvalue above 1 explained by principal components were used to decide the final dietary patterns (30).

Nutritional outcomes of adolescents were determined based on height for age (HA), and body mass index (BMI) for age z-scores was computed using WHO AnthroPlus software (31, 32). A chi-square test was conducted to evaluate the effect of the identified dietary patterns on the burden of stunting and thinness of adolescents. Significant differences were determined with a value of p of < 0.05.

To identify factors for the identified dietary patterns, the first binary logistic regression analysis was carried out to measure the existence of an association between dietary patterns and factors. Then, factors that showed an association between binary logistic regression analysis and those with a p-value of <0.3 were fitted to the multivariable logistic regression model to identify significant factors. Significant factors were reported with a p-value of less than 0.05.

Based on factor loadings, dietary patterns were determined using EFA. Based on the score, possible different dietary patterns were developed (33, 34). Then, each dietary pattern was classified into quartiles (Q) based on their contribution to each pattern, assuming an increasing order from Q1 to Q4 (35). Finally, Q1 and Q2 were combined to represent low consumption habit (0), while Q3 and Q4 represent high consumption of dietary habit (1) (35).

Adolescents’ nutritional status is determined based on height for age and body mass index for age using WHO AnthroPlus software (32). Then, HA z-scores < −2 SD were categorized as “stunted” and < −3 SD were categorized as “severely stunted,” and BMI for age z-scores < −2 SD was categorized as “thin” and > +1 SD was categorized as “overweight/obese” using the WHO cutoff points (31).

Ethical clearance was secured from the Institutional Review Board (IRB) of Addis Ababa University from the College of Natural and Computational Sciences. The study aims, possible risks, benefits, and privacy were described verbally to the study participants. Participation in the study was voluntary, and we gave them full autonomy to participate. To maintain confidentiality, the names and other identifiers of the study participants were not recorded in the data collection tool. Data on the anonymity and confidentiality of the study participants were assured. Informed verbal consent was obtained from parents of adolescents aged <18 years. Direct verbal informed consent was obtained from adolescents aged >18 years.

From the calculated sample size of 627 participants, 622 were involved in the study, with a response rate of 99.20%. The mean age of adolescents was 15.2 ± 2.02 years. More than half (60.10%) of the participants were girls. In total, 82.30% of adolescents’ mothers had no formal education, 11.90% attended elementary education, and the remaining 5.80% attended secondary education and above. Regarding the wealth index, 40.50% had lower thirds, 34.40% had medium thirds, and the remaining 25.10% had higher thirds (Table 1).

The food lists were grouped into common food groups based on their similarity in nutrient contributions. Eight food groups were identified from the collected dietary consumption data. These food groups consist of traditional, legumes, tubers, whole grains, foods with high carbohydrate content, fruits and vegetables, animal sources, and traditional alcoholic beverages. Based on the EFA factor loading analysis, in the first component, traditional and legume diets had higher scores of 0.962 and 0.964, respectively. In the second component, fruit and vegetable diets, high carbohydrate (tea and white bread), and whole grain diets were identified with scores of 0.710, 0.653, and 0.519, respectively. In the third component, animal-source foods (eggs, meat, milk, and milk products) and traditional alcoholic beverages were identified with factor loadings of 0.746 and 0.625, respectively (Table 2).

Based on EFA analysis, three major types of dietary patterns were identified, which explained 58.64% of the variance in adolescent diets. The label assigned to each component was based on the items with high factor-loading scores and the interpretability of the factors. These include (1) the traditional type, explained by 26.71%; (2) mixed type, explained by 16.63%; and (3) animal-source foods with traditional alcoholic beverages, explained by 15.30% (Figure 1). From the identified dietary patterns, traditional, mixed, and animal-source dietary patterns were identified as healthy types; high-carbohydrate and traditional alcoholic beverage drinking patterns were identified as unhealthy types.

The prevalence of stunting was 26.30% (95% CL: 22.70, 30.20%), thinness was 9.60% (95% CL: 7.30, 12.40%), and overweight was 1.80%. Of the adolescents, 22.80% were moderately stunted, and the remaining 3.60% were severely stunted. This community-level undernutrition indicated that one in four and one in 10 adolescents were stunted and thin in their nutritional status, which became significant public health problems.

The burden of stunting was 15.12% versus 11.21 and 19.40% versus 6.94% for the lower and higher traditional and mixed dietary patterns, respectively. In contrast, 7.30% versus 19.00% of stunting was observed in higher and lower animal sources with alcoholic dietary patterns and consuming habits of adolescents, respectively (p < 0.05). However, there were no significant differences in thinness between the lower and higher consumption habits of the identified dietary patterns (p-value >0.05; Table 3).

Adolescents who resided in low-land agroecology (AOR = 2.4; 95% CL: 1.24, 4.81) and mid-land agroecology (AOR = 0.48; 95%: 0.31, 0.74), the early age range of 10–13 years (AOR = 0.6; 95% CL: 0.46, 0.99), and had access to animal sources (AOR = 1.64; 95% CL: 1.04, 2.60) were significantly associated with a lower consumption habit of traditional dietary pattern (Table 4).

A mixed type of dietary pattern was the second component of a dietary pattern, which was explained by 16.63% of all dietary patterns. Among sociodemographic factors, adolescents who were male (AOR =2.28; 95% CL: 1.56, 3.34), resided in mid-land agroecology (AOR = 1.80; 95% CL: 1.18, 2.74), high-land agroecology (AOR = 2.14; 95% CL: 1.19, 3.87), whose mothers had no formal education (AOR = 2.38; 95% CL: 1.35, 4.19), lived with large family size (AOR = 1.69; 95% CL: 1.02, 2.80), and had poor nutritional knowledge (AOR = 2.83; 95% CL: 1.12, 4.05) had a significant association with lower consumption of a mixed type of dietary pattern (Table 4).

A third type of dietary habit was explained by a 15.30% variance. This dietary pattern was characterized by the consumption habits of animal-source foods (milk, eggs, and meat) with traditional alcoholic drinking. Traditional alcohol consumption is an unhealthy dietary pattern. Overall, this type of dietary pattern lacks vitamins and minerals that are found in vegetables and fruits. Being female adolescents (AOR = 1.87; 95% CL: 1.27, 2.74), those who were residing in high-land agroecology (AOR = 2.50; 95% CL: 1.37, 4.56) and had access to irrigation land (AOR = 0.68; 95% CL: 0.47, 0.99) had a significant association with lower consumption of animal sources with traditional alcoholic beverage dietary patterns (Table 4).