This study employed a prospective cohort research design, conducted in four different health centres located in Rwanda's Southern Province. Pregnant women in their late second term or beyond, receiving antenatal care services at selected health facilities and voluntarily agreeing to participate in the study, were recruited at baseline from June to July 2019. These individuals were then followed up after giving birth. The study obtained ethical approval from the Institutional Review Board (IRB) at the College of Medicine and Health Sciences (CMHS), University of Rwanda (UR).

The sample size was determined using G*Power software (25). The calculated sample size was 74 participants per group, assuming an effect size of.05, error probability 0.05, a power (1-error probability) of 0.95, and 4 groups. An additional 26 participants were added to each group to account for potential attrition. All participants aged 15 years or older, attending antenatal care services at four randomly selected health centres, and in their sixth month of pregnancy or beyond, were invited to participate in the study with the help of a midwife/nurse in charge of ANC services. Community health workers in charge of maternal and new born care from selected health centres assisted in identifying eligible individuals who did not have scheduled appointments during the data collection period. A total of 396 pregnant women were initially recruited, but due to attrition, the final evaluation included 311 postnatal women (78.54%).

A structured questionnaire was used to gather data on socio-demographic information, risk and protective variables, gestational and obstetric information, and perceived health status. The research instrument included standardized tools to assess postnatal depressive symptoms and maternal social support.

To detect postnatal depressive symptoms, the Edinburgh Postnatal Depression Scale (EPDS) was used. The EPDS has been widely used in LMICs (26). In sub-Saharan Africa, the EPDS has been validated as a reliable tool for assessing both the severity and the likelihood of depression during the perinatal period in various countries and languages (27). An unpublished study of EPDS diagnostic validity with 200 women utilizing postnatal services in Rwanda found the EPDS to have an optimal cut off for depression of 12, and good sensitivity (78.6%) but lower specificity (64.4%), and an overall accuracy of 68.9% when compared to diagnosis by trained clinicians (28). The Cronbach's alpha values were 0.891 in a previous study in Rwanda (24), and 0.894 for the current study.

The Maternity Social Support Scale (MSSS), a modified version of the original scale by Webster and colleagues (29), was used to measure maternal social support. The original MSSS is a 6-item Likert scale that assesses support from spouses, friends, and family. This scale was expanded from 6 to 12 items to account for the cultural context of social support in Rwanda, including support from in-laws, community members, and neighbors. The MSSS has been previously used in research in Rwanda to measure social support in the perinatal period (14, 28). In the current study, a factor analysis identified two subscales: partner support (including three items such as my husband helps me a lot, there is conflict with my husband, I feel loved by my husband); and friend support (including items related to support from friends, neighbours, and the community), which were utilized to measure the level of maternal social support [detailed results reported in (30)]. The Cronbach's Alpha values were 0.775 for the partner support subscale, and 0.870 for the friend support subscale. The study collected socio-demographic characteristics such as age, marital status, education, occupation, and socioeconomic class. In Rwanda, socioeconomic class known as “ubudehe,” is a social stratification program that determines access to social welfare programs and is based on household income (31). The ubudehe system is divided into four groups, but none of the participants in this study fell into the highest category. Therefore, only the lower three categories are mentioned Obstetrics characteristics were also collected, including information on gestational age at birth (preterm or at term); pregnancy intention (whether the pregnancy was wanted), mode of previous deliveries (normal vaginal, vaginal with complications, cesarean section), and parity. Participants were asked to rate their general health on a 5-point scale ranging from very bad to excellent. These health scores were then recoded into a three-item scale indicating poor, neither poor nor good, and good.

Additionally, participants were asked to report any past exposure to negative life events, such asviolence or highly stressful experiences very (e.g., gender-based violence, childhood abuse, stranger violence, loss of any family member, etc.). This variable was recoded as a binary variable, distinguishing between participants who had never encountered any of these events and those who had experienced at least one stressful life event.

The lead investigator provided training research assistants with backgrounds in midwifery and mental health nursing for data collection. These trained research assistants were introduced to midwives/nurses in charge of ANC at research sites. Through community health workers, researchers contacted potential participants and established a data collection schedule. Written informed consent was obtained from all participants, and for participants below 18 years of age, consent was also obtained from their legal guardians or parents (32).

We used SPSS version 25 to analyse the data. The normality of continuous variables was visually assessed. Postnatal depressive symptoms (EPDS ≥12) were the binary primary outcome of interest. Univariate logistic regression analysis was conducted to explore potential correlations with all predictors, obtaining univariate odds ratios (OR) with 95% confidence intervals (CI) and p-values.

Variables with a bivariate relationship significance of p ≤ 0.30 were included in the multivariate logistic regression models using the Enter command. A relatively high significance threshold was set to account for the possibility that weakly correlated variables might have predictive power when analyzed together (33). The final multivariate model (adjusted model) was created using backward-stepwise regression, with a significance level of p < 0.05. The results are presented as adjusted odds ratios (AORs) with 95% CIs.

A total of 396 individuals participated in the study, with 311 (78.54%) completing the follow-up until delivery, while 85 participants (21.46%) could not be reached via phone or through community health workers. The average age of the participants was 30.21 ± 6.61, with nearly half of them (47.90%; n = 149) aged between 24 and 34. The majority of participants (57.90%; n = 180) were legally married. Three-quarters of the participants (75.2%; n = 234) had only a primary education level or lower. Regarding occupation, most participants (80.1%; n = 249) reported engaging in farming or crop cultivation, and 53.1% (n = 164) belonged to the first and second (lowest) socioeconomic classes.

In terms of obstetric characteristics, the majority of participants (93.50%; n = 288) had term births, with 88.1% (n = 274) having given birth to one or three live neonates. Additionally, 75.60% (n = 235) reported that their pregnancy was wanted, while 71.40% (n = 222) had a normal delivery. Participants self-rated their general health, with 45% (n = 140) perceiving it as favorable. Moreover, 69.80% (n = 217) of the participants reported no extreme stressful or negative life events. Regarding maternal social support, 90.90% (n = 239) received good support from their partners, while 69.80% (n = 217) received good support from friends.

EPDS scores ranged from 0 to 30, with a mean of 7.75 ± 6.222. Participants with EPDS scores of 12 or higher were categorized as having postnatal depressive symptoms. For the overall sample, 20.9% (n = 65) of participants were found to have postnatal depressive symptoms.

In bivariate analyses, socioeconomic class was the only demographic variable significantly associated with postnatal depressive symptoms (OR: 0.36, CI: 0.16, 0.84, p = 002. In contrast to previous studies, age, marital status, occupation and education were unrelated to PND symptoms (see Table 1). Surprisingly, having no partner was not related to elevated PND symptoms.

In terms of obstetrics, maternal and health variables, postnatal depressive symptoms were linked to gestational age at birth (preterm vs. term delivery) (OR = 0.35; CI = 0.14; 0.88, p = 0.03) and wanted pregnancy (OR = 2.39, CI = 1.33, 4.29, p = 0.004) but not parity or mode of delivery (see Table 2). Postnatal depressive symptoms were lower in participants with neither poor nor good (OR = 0.28, CI = 0.14; 0.57, p = 0.001) or good (OR = 0.90; CI = 0.40; 0.20, p = 0.001) health status, and higher in those who experienced stressful events (OR = 4.38; CI = 2.47, 7.77; p = 0.001), had poor partner support (OR = 7.41, CI = 3.06; 17.87, p = 0.001), or poor friend support (see Table 2).

Independent predictors of postnatal depressive symptoms that had a bivariate relationship significance of p ≤ 0.30 were included in the multivariate logistic regression. Socioeconomic status, age, and pregnancy intention did not predict depressive symptoms and were not retained in the final model. The final model explained 0.35% of the variance. Postnatal depressive symptoms were significantly predicted by participants' perceived health status, social support, and prior exposure to negative life events (see Table 3). Specifically, participants who reported neither poor nor good health (AOR = 0.28, CI = 0.11–0.72, p = 0.007) or good health (AOR = 0.14, CI = 0.05–0.37, p = 0.001) were less likely to have scores in the depression range. Poor partner support (AOR = 4.22, CI = 1.44–12.34, p = 0.009) posed a high risk, while friend support (AOR = 0.47, CI = 0.23–0.98, p = 0.04) served as a significant protector. Participants who reported prior exposure to any violence or negative life events were two times more likely to develop postnatal depressive symptoms (AOR = 2.94, CI = 1.37–6.29, p = 0.005).

The study has certain limitations. The attenuated variance in some key predictors of PND within the sample may have limited the ability to identify relationships between risk and protective factors and elevated EPDS scores. Self-selection and selective attrition could have resulted in the exclusion of vulnerable or at-risk individuals from the study or the postnatal assessment. Moreover, women whith elevated EPDS scores were referred for mental health services but were still retained in the study. Although only a small number of women accepted treatment, this may have influenced the relationships between risk factors and EPDS scores.