Common perinatal mental disorders and post-infancy child development in rural Ethiopia: A population-based cohort study

Objective: To investigate whether maternal common mental disorders (CMD) in the postnatal period are prospectively associated with child development at 2.5 and 3.5 years in a rural low-income African setting. Methods: This study was nested within the C-MaMiE (Child outcomes in relation to Maternal Mental health in Ethiopia) population-based cohort in Butajira, Ethiopia, and conducted from 2005 to 2006. The sample comprised of 496 women who had recently given birth to living, singleton babies with recorded birth weight measurements, who were 15 to 44 years of age, and residing in six rural sub-districts. Postnatal CMD measurements were ascertained 2 months after delivery. Language, cognitive, and motor development were obtained from the child 2.5 and 3.5 years after birth using a locally adapted version of the Bayley Scales of Infant Development (3rd Ed). Maternal CMD symptoms were measured using a locally validated WHO Self-Reporting Questionnaire. A linear mixed-effects regression model was used to analyze the relationship between postnatal CMD and child development. Results: After adjusting for confounders, there was no evidence for an association between postnatal CMD and overall child development or the cognitive sub-domain in the preschool period. There was no evidence of effect modification by levels of social support, socioeconomic status, stunting, or sex of the child. Conclusions: Previous studies from predominantly urban and peri-urban settings in middle-income countries have established a relationship between maternal CMD and child development, which contrasts with the findings from this study. The risk and protective factors for child development may differ in areas characterized by high social adversity and food insecurity. More studies are needed to investigate maternal CMD’s impact on child development in low-resource and rural areas. The C‐MaMiE study (Child outcomes in relation to Maternal Mental health in Ethiopia) is a population‐based cohort study [30]. Participants were recruited and assessed in pregnancy and underwent repeated assessments with the index child. Measurements in this analysis were taken at birth and 2 months, 2.5 and 3.5 years after delivery. The C‐MaMiE study was conducted in the Health and Demographic Surveillance Site (HDSS) in the Butajira area, Ethiopia. The Butajira HDSS is 130 kilometres south of the capital, Addis Ababa, and was established in 1986 as part of the Butajira Rural Health Programme [31]. At the time of the study, the HDSS population was 49,943, with 13,268 women of reproductive age [30]. One general hospital exists in Butajira town, and a second hospital is located 8 km outside of town. In addition, four primary health centres and seven health posts serve the HDSS population. Rural residents rely on a livelihood based on mixed farming of cash crops, mainly khat and chilli peppers, maize as the subsistence grain, and false bananas. Parts of the HDSS are food insecure because of a combination of overpopulation and intermittent drought [32]. Health‐centre‐based nurses and experienced project data collectors were trained for 10 days by the project co‐investigator (G.M.) to administer the Bayley III. The co‐investigator has a Master’s degree in applied statistics and experience working with the Bayley Scale in Butajira and was supported by an Ethiopian consultant paediatrician (B.W.) and an Ethiopian psychiatrist (A.A.). The paediatrician took a prominent role in observing the administration of the complete Bayley Scales by trainees, giving feedback, and discussing the findings in detail with the trainees. The data collectors and local female high‐school graduates surveyed using the HOME scale and structured demographic questionnaires. Both nurses and C‐MaMiE data collectors administered the Bayley III with comparably high reliability (Chronbach’s α > 0.7) in a previous validation study [33]. For the C‐MaMiE study, a sample of 1065 women was recruited from 1234 eligible women (86.3%) in the Butajira HDSS between July 2005 and February 2006 [30]. HDSS enumerators identified participants during their routine quarterly surveillance interviews. After giving verbal or written informed consent, the participants were interviewed by data collectors in their own homes. Women aged 15 to 49 years, able to speak Amharic, residing in the HDSS, and in their third trimester of pregnancy were eligible. Women with a known severe mental disorder, such as psychotic or bipolar disorder, or an emergency health condition during enrolment were excluded. The cohort was restricted to women with singleton, living births for the analytic sample, with birth weight measured within 48 hours of delivery, from rural sub‐districts (kebeles), and who had maternal CMD symptoms assessed 2 months postnatally. At the time of recruitment into the study, around 90% of deliveries took place at home [34]. In six rural sub‐districts, a community worker was trained to measure birth weight within 48 h of birth in the woman’s home [24]. Child development was measured with a composite of three sub‐scales (cognitive, motor, and language development) on the Bayley Scales of Infant Development, third edition (Bayley III). The Bayley III has been translated into Amharic and validated in Butajira with this cohort [33]. Items lacking cultural validity were adapted (e.g., pictures adapted for contextual relevance) or dropped (e.g., involving scissors or stairs). No time limit was imposed for the completion of items. Mokken scaling, a method based on non‐parametric item response theory, was used to create a hierarchical scale for the raw scores at both time points [35, 36]. Postnatal CMD was measured 2 months after birth, using the WHO 20‐item version of the Self‐Reporting Questionnaire (SRQ‐20) [37]. The SRQ‐20 functions as a screening tool that assesses the presence or absence of depressive, anxiety, and somatic symptoms in the previous month. The measure has been used in other Ethiopian studies [38, 39] and was validated with this cohort as a continuous measure [40]. A conceptual framework was developed for this analysis based on previous theoretical models and literature on the risk factors for maternal CMD and child development (Figure ​(Figure1)1) [21, 24, 41]. The selection of confounders, mediators, and confounders was theory‐driven. Factors were considered confounders if they were hypothesized to have a relationship with postnatal CMD and child development, affecting their relationship. In contrast, mediators were selected if they potentially explained the relationship between postnatal CMD and child development. Finally, effect modifiers were chosen if there was a hypothesis that the effect of postnatal CMD on child development varied across the levels of another variable. The measures were assessed at pregnancy, birth, and 2.5 and 3.5‐year timepoints (Figure S1). Conceptual framework for the association between postnatal maternal common mental disorders and child development During pregnancy, the following confounders were obtained through self‐report and included maternal age, parental education level, and parity [4]. Assets comprised ownership of 11 resources (e.g., land, house, crops). Socioeconomic status (SES) [4] was confirmed with Mokken scaling [38, 39] and included self‐report of hunger in the last month, indebtedness, lack of access to emergency resources, and perceived lower relative wealth. Marital discord [5] was summarised using Mokken scaling [24] and included self‐report of inadequate help from husband, relationship quality, frequency of quarrels, and perception of problematic alcohol consumption by the husband. Exposure to violence [4] assessed women’s experience of physical violence since birth. Social support [11] comprised women’s perception of the support received with housework and children. The sex of the child [43] was obtained at birth. Obstetric complications [30] summed the responses to instrumental or operative delivery, duration of labour greater than 24 h, and bleeding or fever after delivery. Birth weight [44] was measured within 48 h of delivery using SECA 725 scales to an accuracy of 10 g [35]. Home environment, child growth or stunting, and child illness were measured 2.5 and 3.5 years after birth. Home environment [45] was measured using the original Home Observation for Measurement of Environment (HOME) scale [46]. The HOME measure of environmental stimulation was not formally validated for the setting. Because of difficulties with the contextual adaptation of the HOME, we relied on the sub‐scales based on observation of mother‐child interactions. The other sub‐scales were more challenging as they assessed aspects of a stimulating environment (i.e., number of books and time spent watching television) that were difficult to adapt for this low‐resource, rural setting. The instrument measures the amount and quality of stimulation and support provided to a child. Sub‐scales include a responsivity and an acceptance scale focused on the parent’s attentiveness to the child and negative interactions. Height‐for‐Age Z scores [47] were calculated using WHO standard growth curves to define children as stunted at two Z‐scores below the median. Lower scores are indicative of higher levels of stunting in a child. Height‐for‐age has been argued to function as a better measure of cumulative undernutrition and more predictive of impaired child development [48]. A standard piece of medical equipment for height measurements, a stadiometer, was used to measure height with an adjustable headpiece. Child illness [11] was assessed through maternal recall for the presence of diarrhoea, fever, and severe illness episodes in the past 6 months. SES [11], social support [11], stunting, and sex of the child [17], were conceptualised as potential effect modifiers, and stunting [9] and home environment [12] as potential mediators. The analysis was conducted using Stata Version 16. Participants’ characteristics with missing data on the primary outcome were compared with those remaining in the cohort, using Pearson chi‐squared tests, t‐tests, and Wilcoxon rank‐sum tests. The multivariable analysis of the association between postnatal maternal CMD symptoms and total and cognitive development outcomes was hypothesis‐driven. A mixed‐effects linear regression model with a random intercept was fitted. Model fit was tested using likelihood ratio tests after adding random slopes. An interaction with time was included to estimate the association between postnatal CMD and the change in child development between 2.5‐ and 3.5‐year time points. All conceptualised confounders were added into the multivariable model sequentially, clustered by socio‐demographic, maternal and child, and environmental characteristics. The final model included all a priori confounders (Figure ​(Figure1).1). Effect modification was investigated by including interaction terms for SES, sex of the child, stunting, and social support. Home environment and stunting were individually added to the final model to assess for exploratory evidence of mediation. The National Ethical Review Committee for Ethiopia and the Research Ethics Committee of King’s College London in the U.K. approved the C‐MaMiE study. All participants gave informed consent. Literate women provided written consent, and non‐literate women indicated their consent with a thumbprint. Women received reimbursement for healthcare costs, and participants suffering from severe mental disorders, including psychotic or bipolar disorder, were referred to the local psychiatric unit and covered transportation costs. At baseline, women experiencing violence were directed to a local, community‐based non‐governmental organization for services.