Study design, setting, and participants
A community-based, two-arm, parallel cluster randomized controlled trial was conducted among pregnant women receiving prenatal care at health facilities in Robe and Goba towns, Bale Zone, Southeast Ethiopia, from February to December 2021. Details of this study have been published25. In brief, cluster randomization was used over individual-level randomization to decrease information contamination and for pragmatic reasons, as urban health extension workers (UHEWs) operate in clusters26. Robe and Goba towns, located 430 and 444 km from Addis Ababa city, respectively, were the chosen sites. In the municipalities of Goba and Robe, there were 1832 and 2048 pregnant women, respectively. The source population was all pregnant women attending antenatal care (ANC) in the Robe and Goba towns. The study population included all first- and early-second-trimester (the time between 12 and 16 weeks of gestation) pregnant women attending ANC in the Robe and Goba Towns. First- and early-second-trimester (the time between 12 and 16 weeks of gestation) pregnant women who were permanently residents of the study area were included in the study. Pregnant women with gestational diabetes mellitus or pregnancy –induced hypertension were not included in the study.
Sample size estimation and techniques
Using G-Power software version 3.1, the sample size was calculated by making the following assumptions: an effect size of 0.25, a 95% confidence level (CI), a precision of 0.05, and a power (1 − β) of 80%27. The calculated sample size was 120. The ultimate sample size was 264 after taking the largest sample size into account, along with a design effect of 2 and a 10% non-response rate. Nonetheless, 454 were drawn (intervention group = 227, control group = 227) since the computed sample size for one of the broader study’s other objectives was higher25. Data on births compiled by UHEWs were used to estimate the number of pregnant women in each cluster. Robe and Goba towns have 36 and 24 clusters, respectively. Using a probability proportional to size allocation, the sample size was assigned to each cluster. The systematic sampling technique was used to select pregnant women. In the event that a woman missed her interview due to being out of home, the next eligible pregnant woman in the serial number was contacted. The pregnant woman who had been absent from the interview was contacted the next day (Supplementary Fig. 1).
Randomization, intervention allocation, and blinding
The gestational age was calculated by asking about the beginning day of the last menstrual period, and the pregnancy was confirmed using a urine human chorionic gonadotropin test. Robe and Goba towns were chosen at random. Clusters were randomly allocated to the intervention and control groups. Pregnant women residing in Robe Town received the intervention, whereas those residing in Goba Town did not receive the nutrition education interventions. After pregnant women were evaluated for eligibility, the primary author randomly assigned clusters to the intervention and control groups in a 1:1 ratio to make a balance of clusters. The allocation sequence was produced using simple randomization techniques, including coin tossing.
Nutrition education interventions
Nutrition education was delivered in Afan Oromo and Amharic. An organized work schedule, counselling cards, and nutrition education were provided to the intervention group. The core messages for the lessons were generated utilizing the health belief model (HBM) and theory of planned behavior (TPB) theoretical principles16,28. These messages were taken from those recommended by the Ministry of Health (MOH), Ethiopia29.
Following the gathering of baseline data, pregnant participants in the intervention group received nutrition education for six sessions. Not all women received all 6 sessions; however, nearly 96 out of 100 received 5 of 6. Recruitment was done during the period when animal-source foods were allowed (i.e., during the non-fasting period). Following recruitment at their homes in each cluster, respondents received nutrition education for 30–45 min per session. Six nurses with Bachelor of Science (BSc) degrees delivered nutrition education, while two Master of Public Health (MPH) specialists supervised the nutrition education sessions. The core contents of the session were: increasing knowledge about iron-rich food sources, IFA, iodized salt, meal frequency, and portion size with increasing gestational age; food groups; taking day rest; reducing heavy workloads; enhancers and inhibitors of iron absorption; increasing utilization of health services; and interrupting the intergenerational life cycle of malnutrition; increasing pregnant women’s perceptions of under nutrition and factors leading to it; poor eating practices causing inadequate dietary intake and disease; a diet adjustment; a food-based strategy; diversifying, enriching, and standardizing knowledge regarding FV intake; identifying obstacles and finding solutions to them. By engaging pregnant women in the assessment and analysis of their own FV difficulties using participatory approaches, learning by doing encourages pregnant women to devise their own solutions. Customize the strategy to address barriers such as cost, accessibility, preparation, time, and taste preferences. For example, consider inexpensive FV choices; lowering the perceived obstacles to creating an FV; motivating participants to find solutions to the obstacles; specific food taboos (meat and eggs); enhancing participants’ perceptions of control and intention; enhancing participants’ hand washing proficiency; and enhancing participants’ knowledge and attitudes on the capacity of pregnant women to adjust feeding patterns (Supplementary Table 1).
Nutrition education sessions included presentations, discussions, demonstrations, and picture-based exercises. Key messages, realistic activities, and the GALIDRAA (greet, ask, listen, identify, discuss, recommend, agree, and make follow-up appointments) processes were all identified by the trainers as crucial counselling abilities. However, no concealment was adopted in the trial due to the distinctive features of the cluster RCT and the nature of the intervention being studied. Because the two towns were so far apart, the study was not blinded. Pregnant women were made aware of the intervention yet were blinded to the research hypothesis. After the pregnant women were enrolled, reasonable attempts were made to encourage their retention and full follow-up for the duration of the trial by providing them with incentives to reduce missing data. Periodic conversations about compliance with the intervention during routine meetings and home visits by trainers served to retain interest in the study. After two weeks of nutrition education sessions, post-intervention measurements were assessed at 36–38 weeks. Moreover, home visits were planned to lessen the strain of follow-up visits among pregnant women.
No set schedule was given to the control groups. They did, however, receive standard health care. At the end of the trial, a brief intervention was given to the control group to ensure fairness and achieve a high level of postrecruitment satisfaction. Family health (family planning, nutrition, and vaccination services), disease prevention and control (human immunodeficiency virus/acquired immune deficiency syndrome, sexually transmitted infections, tuberculosis, malaria, and first aid care), hygiene and sanitation, waste disposal management, water supply, food hygiene and safety, control of insects and rodents, personal hygiene, and health education are among the 16 components of Ethiopia’s routine health extension programme packages30.
Data collection
An interviewer-administered, structured questionnaire was used to collect data. The data collection was paper-based. The data collection instruments were modified from the Ethiopian Demographic and Health Survey (EDHS) and previous studies5,31,32,33. For the two groups, baseline and final assessments were performed. Prior to the intervention, information on sociodemographic, economic, substance abuse (alcohol, smoking, tea, or coffee), and reproductive history was gathered. Before and after the intervention, data on nutritional issues, intimate partner violence, physical exercise, healthcare delivery systems, knowledge, practice, HBM, and TPB tools were gathered.
The dietary diversity score (DDS) was computed using a qualitative 24-h dietary recall, as previously described25. The DDS score was determined using nine food categories to reflect the sufficiency of the diet’s micronutrients. All food and beverages consumed the previous day, both inside and outside of participants’ houses, were asked to be recalled. Food groups that were consumed during the reference period were given a score of “1”, and those that were not consumed were given a score of “0” for the nine groups: (1) starchy foods; (2) dark green leafy vegetables; (3) vitamin-A-rich fruits and vegetables; (4) other fruits and vegetables; (5) beans, nuts, and seeds; (6) meat and fish; (7) fats and oils; (8) milk and milk products; and (9) eggs. The food groups ingested during the reference period were added together and ranked into tertiles, with the highest tertile denoting a high DDS and the two lower tertiles denoting a low DDS34.
Principal component analysis (PCA) was used to generate a wealth index. Twenty-one variables entered into PCA included the availability of a water source, a latrine, a bank account, different types of living houses, livestock, agricultural ownership, and items of household property5,35. Details are published elsewhere25.
Twenty-seven previously approved questions were used to assess the state of food security. Families with fewer than the first two, two to ten, eleven to seventeen, and more than seventeen food insecurity indicators, respectively, were classified as food secure, mildly, moderately, and severely food insecure, respectively36,37.
Perceived susceptibility (3 questions), perceived severity and perceived benefits (4 items each), perceived barriers (5 items), cues to action and self-efficacy (4 items each) were individually evaluated using the sums of a 5-point Likert scale (1 = strongly disagree to 5 = strongly agree)28 and TPB constructs: attitude and subjective norms (3 items each), perceived behavioral control (2 items), and behavioral intention (7 items)18. The factor scores were summed and divided into tertiles. Perceived susceptibility, severity, benefit, barriers, cues to action, self-efficacy, positive attitude, subjective norm, perceived behavioral control, and behavioral intention were all labelled “yes” in the highest tertile but “no” in the two lower tertiles.
The importance of fruits and vegetables was also assessed using a ten-item knowledge test38. A respondent received a 1 if they responded correctly; otherwise, they received a 0. The scores were then calculated and ordered in tertile order. Last, a high degree of nutrition knowledge was assigned to the top tertile, while a low level of nutrition knowledge was assigned to the two lower tertiles. After the data collectors were trained, they measured the MUAC of pregnant women. After 36 weeks and up to the time of birth, end-line data were obtained.
Outcome assessment
Mid-upper arm circumference (MUAC) was measured in this study to estimate the nutritional status of the women39,40. Because MUAC changes minimally during pregnancy, it is considered a better indicator of pregnant women’s nutritional status than body mass index (BMI), because pregnancy-related weight gain affects the reliability of using BMI to assess pregnant women’s nutritional status. MUAC measurements were taken on the left arm of subjects to the nearest 0.1 cm using flexible and nonstretchable measuring tape, using standard procedures40. Pregnant women with MUAC ≥ 23 cm were considered well-nourished, while those with MUAC < 23 cm were classified as undernourished40,41. Details have been described (Supplementary Table 2).
Data quality control
The questionnaire was initially created in English, translated into the local languages, “Afan Oromo” and “Amharic,” and then back-translated into English by language specialists to guarantee the consistency of the results. The questionnaire was pretested on 5% of the total sample size of study participants, and the questionnaire’s face and content validity were examined by an epidemiologist and a biostatistician25. Eight BSc data collectors and two MPH professionals each received training on the study’s goals, data collection tools, and ethical considerations to minimize interviewer bias. Supervisors rigorously monitored the data collectors every day to ensure that the questionnaire was successfully completed, and they promptly intervened if it was not. To increase the response rate, the study participants were questioned at their residences.
Data processing and analysis
The data were checked for completeness, consistency, and accuracy and entered into, cleaned, and analyzed using SPSS for Windows version 20 and STATA version 14. Descriptive statistics, including frequencies, percentages, means, standard deviations, and standard errors, were generated for the selected predictors and covariates. Details of model assumptions have been described (Supplementary Table 3). The baseline characteristics of the intervention and control groups were assessed using the chi-square test. The independent sample t test and paired t test were used to compare MUAC between and within the intervention and control groups, respectively. The difference in difference (DID) estimated the difference in the change in the mean value of the end-line and baseline of MUAC42.
We employed a linear mixed effect model (LMM) to evaluate the intervention effect on MUAC, accounting for the clustering effect. The identification of clusters and respondents was analyzed as a random effect in the analytic model. The intervention’s effectiveness was evaluated using time and intervention interaction.
Four models were fitted. The null model (model without predictors), model I (MUAC + group), model II (MUAC + group, time, group × time), and model III (MUAC + groups + predictors and covariates) were all fitted. The intraclass correlation coefficient (ICC) for MUAC in the null model was 0.795, indicating the variability of the conditions attributed to the clustering effect. The Deviance (− 2 LL), Akaike’s information criterion (AIC), and Bayesian information criterion (BIC) values were used for model comparison. The deviance value for Model III was the lowest, indicating that the full model for MUAC was a best-fit model. As a result, Model III was used to make interpretations. The effect size was expressed as an estimate (β), along with the SE and 95% CI. Sensitivity analysis using per protocol analysis was conducted. However, there was no difference in the effect size. Initially, randomly assigned pregnant women were examined in the groups to which they were assigned (intention-to-treat analysis principle). Pregnant women who discontinued due to adherence failure or relocation were included in the intention to treat analysis. The statistical significance of the association was declared at a p value of less than 0.05, and tests were two-sided.
Ethical approval
The current study was ethically approved by Jimma University’s Institutional Review Board before it began (Protocol #: IRB000296/2012). The health offices provided an authorization letter. All methods were carried out in according with the relevant tenets of Helsinki Declaration and good clinical practice43. Each respondent provided informed written consent. The respondents’ privacy and confidentiality were ensured throughout the data collection and administration procedures. The trial for the study was retrospectively registered on Pan African Clinical Trials.gov with a registration number of PACTR202201731802989 on 24/01/2022. The study was reported following the Consolidated Standards of Reporting Trials (CONSORT) 2010 statement44 (Related manuscript Table 1).
link