Study settings

The study was conducted in all zones of the Tigray region except the Western zone due to insecurity (Fig. 1). According to the projection of the 2007 population census, the population of Tigray is approximately 7 million, and approximately 80% of the population comprises rural dwellers. Tigray had a well-established healthcare system before the start of the war in November 2020, with 1,011 public health institutions. Public healthcare services in Tigray are provided through two referral hospitals, 14 general hospitals, 24 primary hospitals, 231 health centers, and 743 health posts. However, more than 80% of health facilities were completely or partially damaged or looted during the war [20, 23]. This study was conducted from January 16–February 14, 2024.

Fig. 1

Illustrative map of the study area (Tigray), 2024

Study design

A household-based cross-sectional study design was used.

Population

All lactating mothers in the six zones of the Tigray region composed the source population for this study. The selected lactating mothers who were lactating for less than six months in the selected districts of the six zones of the Tigray region composed the study population. Households were the study units where a lactating mother was interviewed.

Sample size calculation and sampling procedures

The sample size was determined using a double proportion formula using Epi-Info 7.2.5 software, with different assumptions to achieve sufficient power to identify the differences that may exist and address non-response rates. A significance level of 95%, 90% power, 37% proportion of undernutrition on exposure, 1.49 odds ratio between exposed and non-exposed individuals, and a non-response rate of 10% were used [18]. Consequently, a final sample size of 1245 was calculated and employed.

A multistage stratified sampling procedure was employed. First, the 78 accessible districts in the six zones of Tigray were stratified by rural-urban residence into 51 rural and 27 urban districts. Then, 24 (16 rural districts and 8 urban districts) districts were randomly and proportionally selected. From each selected district, kebelles (the smallest administrative units) were selected proportionally. The proportional inclusion aimed for a 30% inclusion rate of the kebelles within each district. The number of kebelles selected from each district was proportional to the district’s population size. Hence, 78 total kebelles were randomly chosen from each district, and random sampling was also performed for each kebelle. In this study, fifteen to twenty households were selected randomly from each of the kebelles using a list of the eligible households in the kebelles as a sampling frame. The sampling frame consisted of lactating mothers who had under six months old children.

Data collection tools and procedures

We collected quantitative data using a standard interviewer-administered questionnaire, which included questions related to the sociodemographic characteristics of the household/mother, obstetric-related characteristics, nutritional characteristics of the mother, and water, hygiene, and sanitation (WASH) status. The study employed 8 experienced master’s degree holder supervisors and 24 BSc holder data collectors with extensive experience in data collection.

Nutritional assessments

Mid-upper Arm Circumference measurements (MUAC) were used for quick screening of the nutritional status of lactating mothers among mothers who have children aged less than 6 months from the selected study communities. The MUAC measurement is recommended for pregnant and lactating mothers with a cut-off point from 21 to 23 cm, especially in emergency or humanitarian contexts when measuring weight and height may be difficult [24,25,26]. This study used a cut-off point of 23 cm. The measurement was taken by placing a non-stretchable MUAC tape on the non-dominant arm, usually the left arm, without clothing. The tape was placed at the mid-point between the tips of the shoulder and elbow. The measurement was taken three times for each mother to ensure accuracy, and the average values were taken. The MUAC measurements were recorded to the nearest 0.1 cm.

A questionnaire designed to measure food security status was adopted from the Food and Nutrition Technical Assistance Household Food Insecurity Access Scale guidelines [27]. The questionnaire consists of 27 questions in total. The first nine questions are answered with a yes or no response after the respondent is asked to recall whether the condition in each occurrence question occurred at any point in the past four weeks. If the respondent answered “yes” to a question about the occurrence of a certain condition, they were then asked how often it occurred in the past four weeks. The frequency options were “rarely” (one or two times), “sometimes” (three to ten times), and “often” (more than ten times). Based on these scores, the respondent’s level of food security was classified into one of four categories: food secured, mildly food insecure, moderately food insecure, and severely food insecure, which were ultimately dichotomized into food-secured and food insecure households. This tool is widely validated and has good reliability internationally including in Ethiopia.

Finally, the data were collected using the Kobo toolbox, which is designed to collect real-time data.

Variables of the studyDependent variable

Undernutrition (Yes if the MUAC < 23 = Yes, and if the MUAC ≥ 23 = No).

Independent variables Sociodemographic

Maternal age, educational status, occupational status, marital status, and family size.

Obstetrics related

Parity, gravidity, antenatal care visit, current use of family planning, ever use of family planning, and place of delivery,

Nutrition-related

Nutritional counseling, Iron folic acid (IFA) intake, household salt, number of meals, and household food security (HHFS).

Water, sanitation, and hygiene-related

Basic sanitation, water source, handwashing, and water treatment practices.

Data quality assurance

A standard questionnaire was prepared in the English language, and translation to the local language (Tigrigna) and back-translation to English was done. Extensive training was provided to the supervisors and data collectors on the data collection procedures, sampling procedures, appropriate MUAC measurements, and the ODK/KoboToolBox applications. We pre-tested the questionnaire using Kobo collect in the field, and necessary modifications were made accordingly.

The Kobo toolbox helped us collect real-time data to monitor the progress and any errors on time on the central server. In addition, the automated check-in features such as skip logic, constraints, and other data validation rules enabled us to collect accurate data. Moreover, there was strict field supervision to check the completeness and consistency of the completed questionnaires. After data collection, rigorous data cleaning is conducted to increase the quality of the data.

Data management and analysis

We exported the data collected through Kobo to Stata version 17 software for data cleaning, management, and analysis. Descriptive statistical analyses such as frequency, proportion, mean, and standard deviation, were initially performed. Cross-tabulation of the dependent variable with each independent variable was performed.

Robust (modified) Poisson regression was used to identify the factors associated with maternal nutritional status. We preferred the robust Poisson regression model over the convenient logistic regression model because it directly provides the prevalence rate as a measure of effect. This measure of effect is more straightforward to interpret and communicate in survey (prevalence) studies [28, 29].

Binary robust(modified) Poisson regression was fitted for each independent variable to select candidate variables for the final model. A P-value < 25% was used as a cut-off point to select candidate variables for the final regression model. Finally, a multivariable robust(modified) Poisson regression model was used to identify the independent factors associated with maternal nutritional status using an adjusted Prevalence rate (aPR) and 95% Confidence Interval (CI). The model fitness was checked using different techniques, which revealed that there was no overdispersion suggesting the model fit the data well (deviance goodness-of-fit = 739.5, P-value = 1; Pearson goodness-of-fit = 536.5, p-value = 1).

We checked for multicollinearity among the independent variables using variance inflation factor (VIF) statistics, and there was no problem (1.03–6.57).