Maternal Factors Promoting Normal Linear Growth of Children from Impoverished Rwandan Households

Study setting, design, and population

This study was carried out in the districts of Kayonza, Nyaruguru, Burera, Rutsiro, and Gasabo from the provinces of eastern, southern, northern, western, and the City of Kigali, respectively. These districts were purposively selected due to their elevated prevalence of stunting.

This study used a cross-sectional design and a quantitative approach. We included children aged six to twenty-three months and their mothers who lived in the study districts. Inclusion criteria considered the child coming from a low-income family (category 1&2 of Ubudehe) [15], being a singleton, and being born full-term. Eligible participants who appeared to be too sick to participate were not considered to take part in the study.

The sample size of 807 participants was calculated using Fischer formula [16], considering a design effect of 2, a 20% loss to follow-up, and a precision level of 5%. In this formula, n indicates the sample size, and Z represents the abscissa of the normal distribution at a confidence level of 95% (Z=1.96), with the prevalence p of (33.1%), the complement of p, q = (1-p), and the desired accuracy of 5%.

We used a three-stage cluster sampling technique, where administrative villages served as the primary sampling unit, and households served as the unit of analysis. Extensive lists of all eligible candidates (mothers and children ages 6-23) who met the inclusion criteria were created from each chosen village in partnership with community health workers. After lists became available, systematic random sampling was used to select 807 mother-child dyads.

Qualified and experienced enumerators were recruited and trained for three days: one day for theory, one day for practice, comprising the pretest, and one day reserved for the adjustment of the data collection tool. Community health workers from the visited villages helped us to identify eligible households. Twenty homes in one sector of the Gasabo district participated in a pretest; the results were used to update and correct the data collection instruments. We chose one sector within the Gasabo district due to the same attributes as the research area. However, the study did not cover this sector.

Data was collected from 30^th August to 23^rd September 2021 by twenty-five enumerators grouped into five teams. Each team was composed of six enumerators and one supervisor. Enumerators worked in pairs to help each other during the data collection. One team composed of three pairs was assigned to collect data in one District and coordinated with one supervisor. Data collection was conducted in the participants’ households. In order to avoid data entry errors and save time for data entry, quantitative data were gathered using a data collection tool installed in tablets with an electronic data management system called Kobo Collect.

A portable stadiometer was used to measure the child’s length. The remaining data were gathered through a structured questionnaire. All day-to-day collected data were sent to the central system, where quality checks were done daily to ensure completeness of data. The primary investigator and the supervisors oversaw every aspect of the data-gathering process.

Study variables

The main outcome variable of this study was linear growth, and it was presented in the form of height-for-age Z-scores as a continuous variable. The main predictors of this study were the mother having income-generating activity, maternal education, maternal depression, household decision-making (HHDM), number of ANC visits, use of family planning method, family planning types, and mode of delivery. The potential confounding variables were child age, engagement in works for both parents, good handwashing practice, lack of vegetable gardens, and the total number of livestock (inclusive of chicken, goats, and cattle).

To assess maternal depression, we used the EPDS, a ten-item tool with a 0-3 Likert scale and a maximum score of 30 [17]. When it comes to interpretation, the higher the value, the worse the depressive symptoms. The EPDS tool has been widely used throughout sub-Saharan Africa [18]. In addition, it has been validated for examining the extent and potential risks of depression during the perinatal period in numerous countries and languages [18]. It has also been verified as an antenatal depression screening tool in a previous Rwandan study [19]. In the present study, the EPDS was shown to be reliable, with Cronbach’s alpha value was 0.840.

For household decision-making, we used a three-item scale to measure the women’s household decision-making, with the total score ranging between zero and nine [20]. For interpretation, a higher score reflects more maternal involvement in decision-making. For this study, the tool was deemed reliable, attaining a Cronbach’s Alpha value of 0.895.

Data analysis

Data were analyzed using STATA 15 software. In order to export the data into stata for final cleaning and analysis, a data dictionary was developed. After data cleaning, the Shapiro-Wilk Test was used to check for the normality of the dependent variable. Initially, descriptive analysis helped to summarize study variables using frequencies and percentages or means and standard deviations.

The next level involved specifying significant variables considering the main outcome variable. Given that the outcome variable was not normally distributed, we opted to use the Kruskal-Wallis, Mann-Whitney U, and Spearman Rank Correlation tests for bivariate analysis and Huber’s approach for robust linear regression for multivariable analysis. This approach allows for vigorous estimation of the model coefficients, accommodating potential outliers and providing reliable findings in the presence of diverse variable types [21]. Only significant variables from bivariate analysis (p-value ≤ 0.10) were included in the robust regression analysis.

The robust regression analysis models were fitted in three steps. In the first step (Model I), significant variables with p-value ≤0.10 from the bivariate analysis were entered one by one. In the second step (Model II), the significant variables (p-value ≤ 0.05) related to maternal factors were entered one by one into the Model after controlling for other significant factors from Model I, including child age, good and washing practice, and the number of livestock. In the third stage (Model III), we entered all significant variables from Model II with p≤0.05 after controlling the same variables as in Model II. Only significant variables were considered independent factors of normal linear growth. The results were reported as a beta coefficient (ꞵ) with a 95% confidence interval (CI) and p-value ≤ 0.05. In addition, after finding independently significant variables, we thoroughly analyzed their combined effects. To be more precise, we created two separate models to look into the relationship between a combination of household decision-making and ANC visit numbers, a combination of household decision-making and the mother having income-generating activity, a combination of the numbers of ANC visits and the mother having income-generating activity, as well as the interaction of all three variables together. The process used to construct these interaction models followed the same exacting guidelines as our earlier analysis.

Ethical considerations

The Ethical approval was issued by the Institutional Review Board of the College of Medicine and Health Sciences, University of Rwanda (N°. 335/CMHS IRB 2021, which amended N°. 178/CMHS IRB 2022). Additionally, a research permit was obtained from the Rwanda Ministry of Local Governance and the National Institute of Statistics of Rwanda. In addition, the mothers signed their written consent to participate voluntarily.

Linear growth among children

Height-for-age Z-scores, reflecting linear growth for the sample, exhibit a mean of -1.38 ±1.78. As shown in Figure 1, curve A represents a normal distribution, while curve B reflects the study sample’s growth curve. Besides, curve B demonstrates a left-skewed tendency with a concentration of children aged 6-23 months from impoverished Rwandan households reflecting relatively lower height-for-age Z-scores. This negative value suggests a tendency towards linear growth faltering. However, even though the present mean displays a negative skew, a large number of children still fell within the normal range (above minus two standard deviations), indicating that a significant number of children had a likelihood of improving their height-for-age Z-scores despite the unhurried improvement.

• Download figure
• Open in new tab

Figure 1. Height-for-age Z-scores (linear growth)

Sociodemographic characteristics of study participants

As detailed in Table 1, of the mothers who completed the survey, the mean EPDS score was 10.84 ±5.91 out of a maximum of 30. This mean demonstrates a moderate level of depressive syndromes within the study population. Furthermore, the mean HHDM score was 7.44 ±2.09 out of a maximum of 10. The mean here shows a relatively high level of involvement of the mothers in household decision-making among the study population.

In relation to socioeconomic status, the mean livestock owned by households was 1.48 domestic animals. Only 7.7% of mothers said they had income-generating activities, with even fewer 3.7%) claiming to be only the ones who work in the household, and 11.03% did not receive formal education.

In terms of reproductive health, around half of all women (49.7%) did not use contraceptive methods before getting pregnant. The mean number of ANC visits was 3.50 ±0.863, while Rwanda recommends at least 4 ANC visits. Further, most mothers gave birth in a health facility, with only 1.9% reporting giving birth outside a health facility. Post-delivery, 13.8% of mothers claimed to experience breast discomfort during lactation. In addition, about half of all surveyed households (51.1%) had a vegetable garden, while 74.1% demonstrated good handwashing practices, and 86.3% reported having both parents working.

Using the same table, we examined the relationship between the sociodemographic characteristics of mothers and the normal linear growth of children. The following variables were significant (p-value < 0.05): child age (p-value<0.001), maternal education (p-value= 0.005), EPDS (p-value=0.004) and HHDM (p-value<0.001), ANC visits numbers (p-value=0.001), the mother has income-generating activities (p-value<0.001), the mother works alone (p-value= 0.019), experience breast discomfort during lactation (p-value= 0.037), mode of delivery (p-value= 0.004), the total number of livestock (p-value=0.024), and good handwashing practice(p-value<0.001).

Multivariable analysis: maternal factors promoting a normal linear growth of children

Table 2 shows the results of multivariable analysis into three models, with the final model results presented here. The results suggest that maternal engagement in income generation, higher scores of maternal involvement in household decision-making, and more ANC visits are all positively associated with higher height-for-age Z-scores. To be more specific, the mother’s being engaged in income-generating activities increased by 0.640 to the Z-scores of height-for-age of the child (ꞵ=0.640 [0.0269−1.253], p-value= 0.041). In addition, with an increase in one unit of household decision-making, there is an increase of 0.147 in Z-scores of height-for-age of the child (ꞵ=0.147 [0.080− 0.214], p-value< 0.001). Further, in line with the increase in one antenatal care visit for the woman, there is an additional 0.189 units to the Z-score of height-for-age (ꞵ=0.189 [0.025− 0.354], p-value= 0.024).

Table 3 presents the combined effects of all significant variables. The findings showed that a combination of HHDM and ANC visits was significant (ꞵ=0.032 [0.019− 0.045], p-value<0.001). This coefficient means that for each unit increase in the combination of household decision-making scores and number of ANC visits, a corresponding increase of 0.032 units in linear growth was anticipated.