Study design
The study was an extension of a population-based birth cohort, the child outcomes in relation to maternal mental illness in Ethiopia (C-MaMiE) study [13]. For the current analysis, three contiguous birth cohorts were included: the original C-MaMiE cohort, cohort A (born in the preceding 12 months) and cohort B (born in the following 12 months). At assessment time-point 0 (T0), children were all aged 6.5 years (SD 0.04) and then followed up at two further assessment time-points (T1 and T2) for an average total period of 2.2 to 3.5 years, depending upon the birth cohort. The prospective relationship between the primary exposure and outcomes was assessed separately for two waves of the cohort: from T0 to T1 (maternal report of drop-out) and from T1 to T2 (school records of drop-out, absenteeism and achievement).
Study setting
The C-MaMiE study is located within the Health and Demographic Surveillance Site (HDSS) of the Butajira Rural Health programme, established in 1986 [14]. The Butajira HDSS includes nine rural administrative sub-districts of different ecological zones and one urban sub-district in Butajira Town. Butajira is a predominantly rural area found in the Gurage Zone of the Southern Nations, Nationalities and Peoples’ Region (SNNPR) of Ethiopia, 135 km away from the capital city, Addis Ababa. The Zone is characterised by high population density with substantial ethnic and linguistic diversity. The local economy is based on mixed farming of cash crops (khat and chilli peppers) and staples (maize and “false banana” or Ensete ventricosun).
Educational context
In Ethiopia there is no school fee for governmental schools; however, parents are expected to cover costs for school uniform, exercise books and food [15]. In the study area, less than 0.3% of the study children attend private schools. The country has no standardized measure of school readiness but depends on self-reported age of 7 years. Primary education has two cycles; basic (grades 1–4) and general (grades 5–8) between the ages 7 and 14 years [15]. Coverage of primary school was estimated to be 85.5%, with 10.1% drop-out overall, but higher drop-out in grade one (16.8%) and 6.7% grade repetition in 2015/2016 [16]. Except for two national examinations at the completion of grades 10 and 12, and one regional examination at completion of grade 8, the academic performance of students is assessed by the class teacher using non-standardized tests.
Study participants
The original C-MaMiE cohort was established between July 2005 and February 2006 in the Butajira HDSS [13]. At recruitment, inclusion criteria for the women were: age between 15 and 49 years, ability to speak Amharic, resident of the HDSS and in the third trimester of pregnancy. Fewer than 3% of eligible women were excluded at recruitment because of lack of fluency in Amharic language. A total of 1065 out of 1234 eligible women (86.3%) were recruited and followed to date. When the C-MaMiE children were aged 6.5 years, the cohort size was augmented using the HDSS to identify cohort A (n = 543; 94.9% of eligible) and cohort B (n = 717; 92.8% of eligible) with application of identical inclusion criteria.
Assessment time-points
Exposures assessed at T0 (2012/2013 academic year) and T1 (2013/2014 academic year).
Outcomes assessed at T1 and T2 (2014/2015 academic year) at a mean age of 9.3 (SD 0.5) years. See Additional file 1 for a graphical depiction of the cohort waves and mean age for each cohort.
Measures
Outcomes
Absenteeism The total number of days of absence was obtained from the daily school attendance record at T2.
School drop-out This was operationalised as the proportion of students who had enrolled at the beginning of the academic year (September) but who had dropped out of school before the end of the academic year (June) and obtained from maternal report at T1 and from school records at T2. Presumed reasons for drop-out at T1 were also obtained from mothers. Children who drop out of school at T1 can be re-enroled and, therefore, drop-out again at T2.
Academic achievement The averaged grade point over two semesters of the Ethiopian school year was measured at T2. The class teacher grading of academic achievement is non-standardised and incorporates mastery of content, class participation and interaction, conduct, homework, progress over time and school attendance.
Primary exposure
Child EBD was measured at T0 and T1 using the brief screening, parent-report version of the Strengths and Difficulties Questionnaire (SDQ) for emotional and behavioural difficulties in children and adolescents aged 4 to 16 years [17]. An approved Amharic version of the SDQ is available [18]. Within the C-MaMiE cohort, the SDQ has been found to have construct and convergent validity when used as a continuous scale [19]. In keeping with the recommendation of the developer, we applied a score of 14 and above to indicate high emotional and behavioural symptoms, but we were not able to relate this to mental health problems as the criterion validity of the SDQ against clinical diagnoses has not been established in Ethiopia.
Potential confounding factors
All potential confounders were assessed at the time-point preceding the outcome measure (T0 for T1 and T1 for T2).
Maternal common mental disorder was measured using the World Health Organization (WHO) 20-item version of the Self-Reporting Questionnaire (SRQ-20) [20] which has been validated in this setting [13].
Stressful life events over the preceding 6 months were measured using an adapted version of the List of Threatening Experiences [21].
Socio-economic status (SES) was measured using self-report of house roof composition (corrugated iron vs. thatched), the experience of hunger in the preceding month, and the existence of emergency resources for times of crisis.
Paternal substance use report of the frequency of alcohol or khat use by the father was obtained from maternal self-report.
Child nutritional status height was assessed by project data collectors using a portable stadiometer with a movable head piece for height. Using the World Health Organization (WHO) reference population [22], height-for-age z scores were calculated with the WHO Anthro software [23]. Although weight was also assessed, height-for-age and weight-for-age were collinear. As height-for-age is a cumulative indicator of nutritional status [24], the height-for-age z scores were preferred for the analysis.
Demographic characteristics age of the mother, marital status, literacy level, birth order and sex of the child were obtained from self-report of the mother.
Data management
Data collection procedures interviews with the women and anthropometric measures of the child were carried out in the woman’s home, or in the surrounding area, according to their preference. Child anthropometric measures were also conducted in school when convenient. The project data collectors, who had all completed secondary school education and above, were given 3 days of intensive refresher training on the administration of instruments. The questionnaires were piloted before commencing data collection and discrepancies in ratings were discussed to ensure that the data collectors had a common understanding.
Maintaining data quality supervisors monitored data quality and identified missing data in the field. Random quality checks were performed on a sample of assessments. Data were double entered with EpiData [25] by project data clerks on the day of data collection, where possible.
Statistical analyses
Stata software version 12 [26] was used for the analysis. The analysis was hypothesis-driven with potential confounders specified a priori. SDQ total score, SDQ high (≥ 14) score and total scores of SDQ sub-scales were the primary exposures. Logistic regression was used for school drop-out (binary outcomes), zero-inflated poisson regression was used to model absenteeism (count variable, with excess zeroes), and linear regression was used to model academic achievement (continuous outcome, normally distributed). Estimates of association were presented with their respective 95% confidence intervals. An exploratory analysis was carried out to examine whether there was effect modification by cohort between the primary exposure and outcomes.