Menu
Menu
Menu
Menu
Investing in adolescent girls’ nutrition is vital for health and for breaking the intergenerational cycle of malnutrition and deprivation, but limited knowledge on the type, timing, and efficacy of interventions delays progress. We describe the design of a 26-wk randomized placebo-controlled trial with multiple-micronutrient-fortified biscuits (MMBs) among adolescent girls in northeastern Ghana. Apparently healthy, premenarche (n = 312) and postmenarche (n = 309) girls (10-17 y) were randomly assigned to receive the following for 5 d/wk: 1) MMBs (fortified with 11 vitamins and 7 minerals) or 2) unfortified biscuits. Data included plasma micronutrient status, anthropometry, body composition, cognitive function, psychosocial health, fertility, dietary intake, and sociodemographic and socioeconomic covariates, complemented with in-depth interviews (n = 30) and 4 focus group discussions. We hypothesized an increase in plasma ferritin and retinol-binding protein with a resultant increase in hemoglobin, cognition, vertical height, and psychosocial health. Our study seeks to investigate the efficacy and optimal timing of a multiple-micronutrient food intervention program for adolescent girls. The RCT was registered prospectively with the Netherlands Clinical Trials Register (NL7487).
We conducted the study in the Mion District, in Northeastern Ghana. The district is located in the eastern corridor of the Northern Region of Ghana between latitude 90–35° north, 00–30° west, and 00–15° east. The district shares boundaries with the Tamale Metropolis, Savelugu Municipal, and Nanton District to the west; Yendi Municipal to the east; Nanumba North and East Gonja districts to the south; and Gushegu and Karaga districts to the north. The district capital is Sang, the largest community in the district. The district covers a surface area of 2714.1 square km and has a population density of 30.1 persons per square kilometer (72). The area has a typical tropical climate with 2 main seasons: a dry season (November–March), characterized by high temperatures, and a single rainy season (April–October). According to the 2010 Ghana Population and Housing Census (72), Mion District had a population of 81,812, with the majority (91.1%) living in rural locations. In 2010, the average household size in the district was 9.3 persons per household. According to the GSS, ∼20% of the district’s female population was aged 10–19 y (72). The district’s main ethnic groups are Dogombas and Konkombas, and ∼61.8% of the district population professes Islam. Over 90% of the people depend on agriculture for their livelihood. In 2010, the district’s literacy rate was 28.7% for both sexes, implying a very high illiteracy rate in the district (72). The study protocol was approved by the Navrongo Health Research Centre Institutional Review Board (NHRCIRB323). The RCT was also registered prospectively with the Netherlands Clinical Trials Register (NL7487). We purposively selected Mion District as it is relatively new, carved out of Yendi Municipal Assembly in 2012. Hence, data on nutrition and health in the district are scanty. Moreover, the district is mainly rural (∼91%), and our secondary analysis (30) suggests a very high prevalence (64.6%) of anemia among adolescent girls in the rural northern savannah agro-ecological zone. Last, the district capital is only ∼1 h drive from Tamale, the regional capital and location of the University for Development Studies (UDS), which coordinated the fieldwork. Figure 2 is a map of the district with locations of the selected communities where the study was conducted. Map of Mion District, Ghana, with the communities included in the Ten2Twenty-Ghana study. MMDAs, Metropolitan and Municipal Assemblies. The study started with an extensive cross-sectional survey (n = 1057), 2 mo before a follow-up double-blind, placebo RCT. Herein, we refer to the cross-sectional survey as the “survey,” and we describe the methods for the survey and RCT in this article. For ethical reasons, a nontargeted approach that did not distinguish anemic and nonanemic girls was used to include a random subsample (n = 620) of girls from the RCT survey. The nontargeted approach was also justified by the high prevalence of anemia (64.6%) among female adolescents in the rural northern savannah agro-ecological zone of Ghana (30). The nontargeted approach was previously used in similar efficacy trials and proved to be effective (43, 73, 74). The girls were randomly assigned to 2 parallel treatment arms receiving nutrition/health education (5 different occasions) with a 5-d weekly MMBs or UBs for 26 wk. Similar studies (40–45) reported significant effects of MMFs on children’s and adolescents’ micronutrient status when consumed between 5 and 7 d for 3–12 mo with an average duration of 6 mo; this informed our decision to administer the treatment 5 d/wk for 6 mo. Figure 3 shows a schematic overview of the RCT. To estimate mean nutrient intake and the proportion of the population at risk of nutrient inadequacy, a random subsample (n = 310) of subjects from the RCT (including both pre-and postmenarche girls) was selected for a quantitative 24-h dietary recall (24hR). Out of the first 24hRs, a random sample of 100 girls was selected for a second 24hR, allowing us to adjust for the random day-to-day variation in dietary intake. For triangulation purposes, we also conducted 1 focus group discussion in each of the RCT arms at the endline. Likewise, in the extensive cross-sectional survey, we conducted qualitative in-depth interviews (n = 30) and 2 focus group discussions. Supplemental material indicates the SPIRIT checklist for the study. Design of a 26-wk, double-blind, randomized placebo-controlled trial among adolescent girls in Ghana. DDS, Dietary Diversity Score; MMB, multiple-micronutrient fortified biscuits; UB, unfortified biscuits; 24hR, quantitative 24-h dietary recall. The target population for the study included premenarche and postmenarche adolescent girls. The girls were seemingly healthy, nonpregnant, and nonlactating adolescents aged 10–17 y, residing in the Mion District in the Northern Region of Ghana. To be enrolled into the survey, girls had to meet all of the inclusion criteria, and those who did not meet the exclusion criteria of the RCT were eligible to participate in the RCT (Table 1). Inclusion and exclusion criteria for Ten2Twenty-Ghana study1 The survey sample (n = 1040) was estimated to detect a minimum mean difference of 0.30 in math and verbal skills z scores between stunted and nonstunted adolescent girls (60) using the RMASS program (http://www.rmass.org/) (75). The RCT sample size calculation was based on 80% power, a 1-sided hypothesis at 5% significance level for 3 variables: hemoglobin (Hb), serum ferritin (SF), and serum retinol. The SD for Hb in this population was 12.9 g/L (for both anemic and nonanemic girls) and 8.4 g/L (for only anemic girls) (30). Therefore, to detect a difference in mean Hb of 3.83 g/L between the MMB and UB groups required 141 girls per group for a nontargeted approach and 122 girls per group for only anemic girls. Based on an SD of 20.1 µg/L for SF from a previous study (76), 57 girls per group were required to detect a mean difference of 9.5 µg/L for SF between MMB and UB groups. Last, the SD of serum retinol from a previous study was 0.29 µmol/L; hence, 23 girls per group were required to detect a mean difference of 0.22 µg/L between MMB and UB groups. Expected mean differences for Hb (3.83 g/L), SF (9.4 µg/L), and serum retinol (0.11 µmol/L) are biologically plausible (43), which are within the range of our estimates. We considered the larger estimate (n = 141) of the 3 variables (Hb, SF, and retinol), and considering a maximum attrition rate of 10% during follow-up a minimum sample of 155 girls/group was considered for the RCT. With premenarche and postmenarche girls randomly assigned to the parallel arms of the RCT, the study had, in total, 4 groups, implying a total of 620 adolescent girls were required for the RCT (310 premenarche and 310 postmenarche). Last, we estimated the sample size for the 24hR with the 1 random sample formula considering a 95% CI, an estimated width of 10.13 mg, and an SD of 28.9 mg for iron intake, as well as an estimated width of 50.5 µg retinol equivalents (RE) and an SD of 113.2 µg RE for vitamin A intake (77). For both iron and vitamin A intake, the estimated sample was 130 girls and, considering an attrition rate of 20%, this was rounded up to 150 girls for each menarche/age cohort. Using the rule of thumb recommended by Rothman (78), a subsample of 50 per cohort from the first recalls was included in a repeated recall to allow for adjustment for random day-to-day variation in intake at the population level. Focus groups typically have 6–12 members plus a moderator (80), but Wyatt et al. (81) indicate that focus groups with 4–6 children are most effective in yielding valuable information because duplicate responses are less common and smaller groups are easier to control. According to Wyatt et al., children may be reluctant to talk in larger groups. Hence, we sampled 5 girls for the composition of each focus group in the survey and the RCT endline. Finally, Rothman’s (78) rule of thumb was used to decide on 50 RCT-nonenrolled girls as a control group for the RCT-enrolled girls for the body-composition analysis. In Mion District, where the study was carried out, there are 70 primary and 11 junior high schools. The latter were excluded due to an already ongoing iron and folate supplementation project called GIFTS (Girls’ Iron Folate Tablet Supplementation) among adolescent girls in junior high schools (82). While the Ghana Education Service (GES) has zoned Mion District into 6 clusters, for the Ten2Twenty-Ghana project 2 clusters were excluded based on their inaccessibility (remoteness). The remaining 4 clusters had 41 primary schools, and we ranked these schools in descending order based on the size of their female child enrollment using secondary data on enrollment obtained a priori from the GES in Mion. We purposively selected all the urban primary schools (n = 4) and the larger rural primary schools (n = 15) for screening until the minimum sample required for the survey (n = 1040) was met. In each school, we screened all the girls using a 16-item screening questionnaire including personal and household identification and demographics, menarche status, pregnancy and lactation status, health condition, use of any medical drug, iron supplements, and participation in a study with drugs, supplements, or food. Subsequently, girls who met the survey’s inclusion criteria were invited to participate after obtaining their assent and their parents’/guardians’ informed consent. During the enrollment of subjects into the RCT, we added participants who were not postmenarche at in the survey using age >13 y, ensuring that we had enough sample size for randomization into the postmenarche group of the RCT. The cutoff age (>13 y) was chosen in conformity with the average age at menarche in Ghana from the literature (83–85). Thus, the postmenarche group in the present study includes postmenarche girls at screening or who were expected to become postmenarche during the RCT. Probability proportion to size was used to select a random subsample of girls from the survey who did not meet the exclusion criteria of the RCT in 2 stages. First, we generated random numbers (between 0–1) by school and menarche group in Microsoft Excel (MS Excel, version 2016; Microsoft Corporation). The random numbers were sorted in ascending order (lowest to highest); the first set of participants from the menarche group of each school was enrolled until the sample size required for the school’s menarche group was met. Any girl who dropped out during the enrollment was replaced with the next girl in the list from the same school and menarche group in the ascending order until the sample requirement was met. Subsequently, a second set of random (between 0–1) numbers was generated for the girls enrolled in the RCT in MS Excel. All enrolled subjects with random numbers <0.5 were assigned to a yellow color code, while subjects with random numbers ≥0.5 were assigned to a red color code. The first step of the probability proportional to size approach described for the RCT was again used to randomly select 155 girls from each of the RCT menarche groups for the first 24hR. For the repeated 24hR, another random selection process like the preceding was used to select 100 girls from the sample for the first 24hR. At the RCT endline, the probability proportion to size approach was once more used to select 50 RCT-nonenrolled girls for body-composition assessment. In-depth interviews are known to be labor intensive, and most studies utilizing in-depth interviews are based on 10%, measurements were repeated and obvious outliers removed. The CVs for the various indicators were as follows: SF, 2.3%; TfR, 3.6%; RBP, 3.6%; CRP, 5.8%; and AGP, 8.1%. Certified quality-control samples from the CDC/Atlanta and Bio-Rad Liquicheck controls (Bio-Rad) were used to calibrate the results. Plasma samples from the 2 time points (RCT baseline and endline) were analyzed at the same time. Plasma zinc was analyzed with atomic absorption spectrometer, while folate and vitamin B-12 were analyzed with HPLC later. We assessed the current intake of a subsample of the girls enrolled in the RCT with a quantitative 24hR using the USDA standard multiple-pass procedure (89). To enable adjustment for random day-to-day variation in dietary intake, we repeated the quantitative 24hR in a subsample (n = 100) of the girls with a first quantitative 24hR on nonconsecutive days to avoid dependency of intake. Trained interviewers conducted the dietary interviews at home, 1 mo preceding the RCT. We randomly assigned subjects to all days of the week and interviewers to account for differences in intake between days and interviewers. No interviewer could interview the same subject twice. In the standard multiple-pass procedure, the girl was first asked to mention all foods and drinks, including snacks that she consumed in and outside the home (including school) the previous day. She was then asked to describe the ingredients and cooking methods of any mixed dishes. The primary caregiver and/or the person who prepared home meals the preceding day was asked to help the girl list and estimate ingredients for mixed dishes prepared at home. We recorded the actual weight of a duplicate portion of each food, beverage, and ingredients of mixed dishes using a digital kitchen scale (Soehnle Plateau, model 65086) precisely to 2 g with a maximum capacity of 10 kg. In the absence of duplicate portions in the household, amounts were estimated as their monetary value equivalents, weight-to-weight with other foods (e.g., amount of sugar estimated with refined corn flour), in volumes, food models (small, medium, or large), or as household units in priority order. The research team agreed a priori on models for food such as onion, tomatoes, and garden eggs, which were carried alongside. We estimated the total volume of each mixed dish cooked at the respondent’s household and the volume of this dish consumed explicitly by the girl to determine the proportion of the dish she consumed. The amount of ingredients consumed from mixed dishes by the girl was estimated by multiplying the proportion consumed with the total amount of ingredients used to prepare the dish. We also recorded each food ingredient’s frequency of intake (for mixed dishes) or food item for the last 7 d preceding the interview day. For shared-bowl eating, the girl’s usual intake for such dishes and the number of persons who ate from the shared bowl were recorded in the logbook. The 24hR ended by probing the girl for likely forgotten foods—notably, fruits, sweets, and snacks consumed on the recall day. Standard recipes and school feeding recipes were generated to estimate grams of ingredients consumed from mixed dishes eaten outside the home or from the school feeding program. Estimates of these recipes were obtained by averaging 3 recipes of different vendors and different school feeding matrons/cooks. The vendors and school feeding matrons/cooks were each selected from different localities and schools. Moreover, we developed conversion factors to convert monetary values, weight–weight measures, volumes, food models, and household units to their weight (grams) equivalents following Gibson and Ferguson (89). Last, we conducted a market survey in 4 different markets in each of the study clusters. We determined the mean price per 100 g of edible food for each listed food in the 24hRs using the average price and weight of foods obtained from each of the surveyed markets. Height and weight were measured in duplicates to the nearest 0.1 decimal with the Seca stadiometer and digital weighing scale, respectively, in the survey and at the RCT endline. Standard anthropometry guidelines were followed (90) in the assessment. Height and weight were transformed into height increment, attained height, z scores (height-for-age, BMI-for-age), and BMI. The z scores will be computed with the WHO AnthroPlus software with the WHO growth reference for adolescent girls aged 10–19 y. In the survey and at RCT endline, we also assessed body composition with bioelectric impedance using the Bodygram Plus Analyser (Akern, Germany) (91). In the body-composition assessment, subjects laid in a backward position with their arms by their side on a field camp bed for 3–5 min to ensure uniform distribution of body fluids before the assessment. The girls’ feet and wrist were wiped with nonalcoholic wipes before placement of the bioelectric electrodes for the appraisal. The electrical resistance (Rz) of the tissues and capacitive resistance of the cell membranes (XC) in whole numbers were recorded on a form and later input into Bodygram Plus Analyser’s software(Akern, Germany) for the computation of body composition. Body-composition estimates were to the nearest 0.1 decimal. They included fat mass (kilograms), fat-free mass (kilograms), muscle mass (kilograms), skeletal muscle mass (kilograms), body cell mass (kilograms), total body water (liters), extracellular water (liters), and intracellular water (liters). The program also computes indices and percentage to the total body weight for these estimates. In the survey and at RCT endline, we assessed the girls’ dietary patterns with a qualitative 1-mo food-frequency questionnaire (FFQ). A 10-food-group indicator (92) was adopted for the FFQ. Likewise, we assessed the frequency of the consumption of energy drinks using a list of energy drinks a priori collected through a market survey (Abdul-Razak Abizari, University for Development Studies; unpublished data 2018) at the RCT endline. A single qualitative 24hR was also used to assess the Dietary Diversity Score (DDS) of the girls using the 10-food-group indicator (92) in the survey and at RCT endline. Furthermore, the girls’ households’ food-security status was assessed with the Food Insecurity Experience Scale (93). In the survey and at RCT endline, age at menarche was assessed by recall and pubertal development stage by a 5-item Pubertal Development Scale questionnaire (94, 95). A semi-structured questionnaire assessed relationships (sexual) of the girls. Psychosocial health outcomes were assessed in the survey and at RCT endline with validated scales including self-reported health-related quality of life (HRQoL) using KIDSCREEN-27 (96, 97), subjective health complaints (98), self-esteem (99), self-efficacy (100), and life satisfaction (98, 101). Furthermore, the assessment included body image of subjects using the Stunkard figure rating scale (102) in the survey. Finally, we included and assessed depression using the Children’s Depression Inventory (103) at the RCT endline. The data included secondary data collected from the schools on the school attendance of the girls and of their grades in English Language, Mathematics, and General Sciences in the academic year prior to the study (September 2017 to July 2018) and at the end of the RCT (September 2018 to July 2019). The academic data were collected for the overall sample (n = 1057) from the survey at both time points. At both time points (survey and RCT endline), we assessed the cognitive function of the girls with the NIH toolbox cognition battery (NIH-TCB) (104, 105). The NIH-TCB is a recognized and standardized test tool for measuring cognitive function. The test is computerized on iPads (Apple), and the scores are automated at the end of each test. The tests appeared as games the girls had to play, but since our subjects were generally from a rural setting, we recognized that they might be limited in playing computer games. Hence, they could point to the right answer on the screen, with the interviewer clicking for them instead. We assessed 5 domains of cognitive function, which were found based on the literature to be relevant to adolescents’ neurological development (106, 107). The 5 domains included episodic memory with the Picture Sequence Memory Test, working memory with the List Sorting Working Memory Test, attention with the Flanker Inhibitory Control Attention Test, processing speed with Pattern Comparison Processing Speed Test, and executive/shifting function with the dimensional Change Card Sort Test. A set of unscored trial tests preceded the actual tests; the unscored test allowed the girls to practice before the actual test. In the survey, we adopted the “Young Lives” questionnaire (68, 108) to assess the time use, labor participation, and earnings of the girls. A life history calendar (109) mapped the labor participation of the girls. Last, the questionnaire included the girls’ expectations and aspirations for marriage, family formation, education, and work. In the survey, the anthropometric assessment included the height and weight (nearest 0.1 decimal) of the girls’ mothers for whom BMI and maternal height would be used. The data also included the mothers’ participation in household decision making using the Demographic and Health Survey 8-item final decision-making index (110). Life history calendars (109) also captured data on the mothers’ fertility and labor participation. Moreover, we enumerated household members with a household rooster including their sex, relationship to the index girl, age group, education, occupation, and literacy, ensuring that we can compute various household-related indices. Finally, the International Wealth Index (111) captured data on the household’s socioeconomic status. In the survey and at RCT endline, focus group discussions were conducted by 2 trained research assistants who had previous experience with focus groups. They were trained to probe, listen, and record in writing as well as using a digital recorder the expressions of the girls. One of them moderated the discussions while the other recorded the discussions both digitally and in a notebook. Topics for discussion included the knowledge, attitudes, and practices (KAP) of the girls regarding relationships, reproductive health, risk behaviors, and dietary habits. The discussions also delved into the aspirations, expectations, and life satisfaction of the girls. In the survey, the focus groups also explored the KAP of the girls with regard to their body image. A visual storytelling technique was incorporated into the focus group discussions. The girls’ data generated in the focus group discussions included digital records, notes, and worksheets used for sketches. According to Mack et al. (112), in-depth interviews are optimal for collecting data on individuals’ personal histories, perspectives, and experiences, mainly when sensitive topics are being explored. Boyce and Neale (113) also posited that the approach provides detailed information about a person’s thoughts and behaviors and in exploring new issues in-depth. We used in-depth interviews to explore rich insights into the girls’ lives and understand their motivations, expectations, aspirations for the future, their life satisfaction, relationships, risk behaviors, and the challenges confronting them in their everyday lives. We also solicited information on their usual dietary patterns and the reasons for adherence to these dietary patterns. Several measures were taken to ensure the internal validity of the data. We recruited and trained field research assistants as well as supervisors with relevant field experience who could speak at least 1 of the key local dialects (Dagbani or Likpakpa) fluently. The training included 5 d for the survey, 3 d for the 24hR, 1 d for the focus group discussion, and a 3-d refresher for the endline. Due to the sensitive nature of questions regarding menarche, relationships, and sexuality, all interviewers administering the one-to-one questionnaire were women recruited from the UDS. In the field, supervisors checked and validated all questionnaires for consistency and completeness. A Microsoft Access template was designed and used for the data entry. The data template was coded numerically, such that implausible values in coded categorical data were impossible. All data entry clerks received a 5-d training by an Information, Communication Technology expert who oversaw the data entry. The entries were merged into a single MS-Access file and the data exported into different SPSS templates based on data themes. Data cleaning was performed in the SPSS templates in the field. The entries of 449 out of 1057 (42.5%) and 202 out of 589 (34.3%) questionnaires were verified entirely in all of the data files in the survey and at RCT endline respectively. Data analysis will be conducted with SAS 9.4 (SAS Institute, Inc.) and IBM SPSS (version 25), where necessary. Frequencies and percentages are used to describe baseline summary statistics for categorical data while means ± SDs will be used for continuous, normally distributed data. Skewed continuous data will be presented as medians and IQRs. Data normality will be visually explored with histograms with normality curves, boxplots, and Q-Q plots. Baseline differences in proportions between biscuit groups will be determined with chi-square or Fisher’s exact test, as appropriate. One-factor ANOVA or its nonparametric version (Mann-Whitney U test) will be used to determine differences in means between biscuit groups for descriptive population statistics. Summary statistics will be presented for sociodemographic, anthropometric, and micronutrient indicators at baseline by biscuit group in the RCT to describe the study population. The RCT data analysis follows the intention-to-treat approach with a sensitivity analysis following per protocol (compliance ≥80%). Compliance is defined by the amount (grams) of biscuits consumed expressed as a percentage of the expected total amount that should have been consumed for the entire RCT. The effects of the fortified biscuits on micronutrient status will be analyzed using linear mixed models (LMMs) with maximum-likelihood estimations. LMMs are more robust in handling unbalanced and missing data; the models are also better able to handle the assumption of independence and homogeneity of slopes in the data (114). As our study population was selected from 4 clusters, 19 different schools, and different classes, LMM analysis is preferred over ANCOVA to adjust random hierarchical variables related to the cluster, school, and class of the girls. Similarly, we will use LMM analysis to examine the intervention’s effect on cognition, body composition, and the psychosocial health outcomes (HRQoL, self-efficacy, self-esteem, and life satisfaction) of the subjects. A “Step-up strategy” (115) will be used in building the LMMs. The analysis of body composition includes the effect of the fortified compared with the unfortified biscuits, as well as the effect of being enrolled and not enrolled in the RCT. However, for dichotomous/categorical outcome variables, Cox proportional hazard models will be used to examine the incidence rate and prevalence risk ratio. Cox and Poisson models with robust variance are reportedly better alternatives than logistic regression (116, 117). We hypothesize that the fortified biscuits would significantly affect micronutrient status and the secondary outcomes; hence, a 1-sided hypothesis at 5% significance and 95% CI will be used in the analysis. We will adjust for a set of identified sociodemographic and socioeconomic confounding variables in all associations in the statistical analysis. A confounder will be defined as any variable that differs significantly between the biscuit groups at baseline or any variable contributing at least a 10% change in the crude effect estimates after adjustment (78, 118). All missing data will be imputed if >5% of data are missing using multiple imputation methods in SAS, assuming that the data are missing at random (119). Although no interim analysis was planned, the decision to conduct interim analysis was dependent on reports from the field on AEs and SAEs. The data safety monitoring committee had access to reports of the AEs and SAEs and could request for an interim report. Compl-eat software (www.compleat.nl) of WUR will be used to estimate individual nutrient intake. Nutrient intake will be adjusted for random day-to-day variation in intake using the Statistical Program to Assess Dietary Exposure (SPADE) (120). To determine the population at risk of nutrient inadequacy, we will use the harmonized average requirements proposed by Allen et al. (121). Optifood linear programming (122, 123) will be used to develop and evaluate affordable alternative FBDGs that can fulfill or best meet adolescent girls’ nutrient requirements. Pubertal timing may influence dietary habits/patterns of adolescent girls. For instance, mid-adolescent, compared with early adolescent, girls consumed fewer protein- and vitamin-rich foods in India (124). In addition, the nutrient requirements, notably iron for postmenarche girls, are higher than in premenarche girls. Hence, in the formulation of the FBDGs, stratified analysis by menarche status will be conducted. Last, principal components analysis will be used to identify dietary patterns of the girls. We will use the inductive thematic analysis approach (125) in analyzing all qualitative data from the in-depth interviews and focus group discussions. The analyses will focus on the similarities and differences in the themes within transcripts. This method provides a rich and detailed account of data and the themes emerging from the data (126). Analyses include transcriptions of digitally recorded discussions, field notes, and worksheets from the girls in the focus groups. We will conduct open-ended coding on each text unit (e.g., sentence or paragraphs) and coding the “raw” participant data, such as quotes. The different categories will be sorted into potential themes and all of the relevant coded data extracts will be collected within the identified themes. Coding and categorizing will be done using ATLAS ti (version 8.0; Scientific Software Development) data-management software, which will facilitate the retrieval of coded chunks of transcripts. The protocol was approved in January 2019 by the Navrongo Health Research Centre Institutional Review Board (NHRCIRB323). The RCT was prospectively registered with the Netherlands Trials Register (https://www.trialregister.nl/trial/7487) with registration number NL7487 in February 2019. A data safety monitoring committee comprised of 3 independent persons with relevant experience in nutrition trials reviewed the trial’s safety monthly during implementation. Before the study, a stakeholder meeting was held with the Mion District Assembly, the GES and GHS, and all heads of the selected schools in the district capital Sang. Written permission was next obtained from the GES in the district. We also undertook a community entry sensitization with all of the opinion leaders, the School Management Committee, the Parent-Teacher Association, and teachers of all the selected schools. Last, in the survey and RCT, we invited parents of the eligible girls for sensitization and education about the study at the school; their signed/thumb-printed informed consent for their female child’s participation was then sought. Data collected remain confidential, and study results will be reported in aggregated form so that participants remain anonymous. Only members of the RCT team had access to participants’ records. RCT assistants also signed a written statement to maintain the confidentiality of any personal information from trial participants with whom they may become acquainted.
