Study design and setting
A community-based prospective cohort study design was carried out among pregnant women in five urban settings (Hossana, Shone, Gimbichu, Jajura and Homecho) in the Hadiya zone, South Ethiopia. In this study, a total of 18 kebeles (lowest administrative unit in Ethiopia) were included.
For this study, cohorts of pregnant women were enrolled at the end of 1st trimester of confirmed pregnancy (after 12 weeks of gestation) via house-to-house identification and registration every three months, for a total of nine months. An enrolment was done from July 08, 2019 to March 30, 2020 by trained midwives. During the recruitment, study participants were included in the study based on the eligibility criteria for the exposure variable (IPI). The inclusion criteria were women who: were pregnant at the time of recruitment, had a live birth during the most recent childbirth, and were able to recall the date of last childbirth. Women who had a prior live birth ≥ 60 months earlier, a recent stillbirth, a recent abortion, and those who did not show a willingness to be followed were excluded.
A sample size of 2424 (exposed = 1212 and non-exposed = 1212) was calculated in Epi Info StatCalc version 188.8.131.52 software using the formula for Cohort study design, assuming % of unexposed (24–59 months interval) with outcome (low birth weight) = 8.52%, % of exposed (< 24 months interval) with outcome (low birth weight) = 11.97%, RR of 1.41, ratio of unexposed to exposed 1:1, 1-alpha of 1.96 (two sided) and 1-beta of 0.842 from previous study in Tanzania . However, from July 08, 2019 to March 30, 2020, a total of 2578 pregnant women were enrolled through house-to-house identification, and the enrolled pregnant women were followed until September 30, 2020. Of 2578, 1273 were exposed groups; 769 had IPI < 18 months and 504 had IPI 18–23 months. The remained 1305 were unexposed group (IPI 24–60 months). This categorization was based on World Health Organization recommendation for pregnancy spacing .
Variables and definitions
The dependent/outcome variables were preterm birth, term low birth weight and perinatal deaths.
Preterm birth is defined as a baby born alive before 37 completed weeks (28–36 weeks) of gestation.
Term low birth weight is a live born baby’s weight at term birth (≥ 37 weeks of gestational age) less than 2500 gms.
Perinatal deaths are fetal death after 28 weeks of gestation up to 7 days, which include both stillbirth and early neonatal death.
The exposure variable was inter-pregnancy interval (a time elapsed from live birth to subsequent conception or woman’s last menstrual period) .
Confounding variables were socio-demographic, and economic and reproductive variables such as maternal age, maternal education, husband education, maternal occupation, wealth status, mode of previous delivery, age at first childbirth, parity and pregnancy intention.
Before baseline data collection, the questionnaire was prepared from existing related literature (published articles and Ethiopia Demographic and Health Surveys) based on the study objectives [2, 3, 24]. English version was translated to Amharic version by two native speakers of Amharic language (one was public health and the other was English language and literature in professions). Then back translation to English was done by another two individuals who could speak English (again one was from public health and the other from English language and literature). The questionnaire was pre-tested on 50 pregnant women in Durame town where the actual study population is culturally related. The investigators have amended the pre-test for unclear terms and order of questions. Baseline data about sociodemographic, economic and reproductive variables including the main exposure variable (IPI) were collected at the household level during enrolment via face-to-face interviews. Ten trained midwives collected data and five public health professionals made supervisions. The data collectors at each health facility were assigned and the list of participants was given for each of them. Outcomes (preterm birth, term low birth weight and perinatal deaths) were collected during labor, and delivery and from clients’ charts before discharge was made.
Preterm birth was ascertained as a baby born alive before 37 completed weeks (36 weeks plus 6 days) but after 28 weeks of gestation. Gestational age was computed by subtracting the date of delivery from the date of childbirth and expressed in weeks. Then, gestational age was categorized as < 37 weeks (1 = preterm birth), and as ≥ 37 weeks (0 = otherwise).
Term low birth weight was ascertained as a weight of a baby who born alive at term (≥ 37 weeks of pregnancy) with a birth weight of < 2500 g. The weight categorized as < 2500 g (1 = low birth weight) and as ≥ 2500 g (0 = otherwise).
Perinatal deaths were ascertained as the death of the baby after 28 weeks of gestation and within 7 days postpartum (28 weeks up to 7 days). It includes both stillbirth (fetal death from 28 weeks of gestation up to the time of delivery, with no signs of life at birth such as fetal heartbeat, breathing, and movement) and early neonatal death (death of a live birth baby within 7 days after delivery). If at least one of them, either stillbirth or early neonatal death, happens then it was categorized as perinatal deaths (1 = yes, 0 = no).
The exposure variable (IPI) was ascertained by asking women about the date of most recent childbirth and the last menstrual period. IPI was computed by subtracting the date of recent childbirth from the date of the last menstrual period (LMP). For women who had difficulty in recalling the date of LMP, Ultrasound was used to estimate gestational age. LMP was computed by subtracting the duration of gestation, and then the value of IPI was calculated . To be in line with the World Health Organization recommendation, women with IPI < 24 months were categorized as an exposed group and IPI 24–60 months as an unexposed group. During the analysis, we further categorized IPI < 24 months (the exposed group) into IPI < 18 months and 18–23 months, and then compared with IPI 24–60 months (the unexposed group).
Potential confounding variables are those variables that have an association with an exposure (IPI) variable and the outcomes (preterm birth, low birth weight and perinatal deaths). These confounders were identified by prior theoretical knowledge and literature [9, 14, 16,17,18, 25]. An additional figure file shows this in more detail (see Additional Fig. 1). The potential confounders were ascertained as follow: reported age at interview was measured in completed years and categorized into the 5-year interval. Educational status was measured as no formal schooling, primary education (1st – 8th grade), secondary education (9th -12th grade) and higher education (> 12th grade or certificate, diploma and above). The occupation was measured by asking the women the main occupation that they routinely do. Parity was measured as the number of times a woman gives birth, irrespective of the outcomes of birth (live birth or stillbirth). Age at first childbirth was measured as reported age at the time that a woman had her first child. Mode of previous delivery refers to whether a woman has given the most recent childbirth spontaneous vaginal delivery or by other routes (caesarean section and instrumental) of delivery. Pregnancy intention was measured as whether a woman has the intention to be pregnant or not at the time of conception. Wealth index was measured using household assets for urban residence, which consists of the following items: owner of the house, number of rooms, the material of the roof, material of the floor, material of the exterior wall, source of drinking water, type of latrine, type of cooking materials (1 = electricity, 0 = wood/charcoal/biogas/natural gas, etc.), source of income, and presence or absence of; cell phone, refrigerator, radio, television, stove, chair, table, watch, modern bed, bicycle, bajaj (three wheel vehicle), motorcycle, car, donkey/horse cart, and bank account. Each item was categorized into two (1 = yes and 0 = no). Latrine and water sources were categorized as an improved and unimproved facility based on the world food program and World Health Organization recommendations. Principal component analysis was done to generate the components. Finally, the ranking was done into three quintiles (low, middle and high).
Data were entered in Epi-data version 3.1 software and exported to R version 4.0.5 software for the analysis. Before the analysis data cleaning and recoding were done for all variables. Categorization and recoding for continuous variables were done using information from related literature. Frequencies and percentages, using cross-tabulation, were calculated for categorical variables and discreet continuous variables. For missing data, a complete case analysis approach was applied.
Bivariable and multivariable regression analysis
A generalized linear model for binary outcomes was used to assess the association of IPI with perinatal deaths. In the multivariable model, IPI was adjusted for all possible confounding variables, and a significant association with perinatal deaths was declared using a 95% confidence level and p < 0.05. For the other perinatal outcomes (preterm birth and term low birth weight), a multilevel generalized linear model for binary outcomes was considered due to the presence of the clustering effect, as described underneath in the model specification section. Multivariable multilevel model adjustment was done for both individual and community-level confounding variables.
Model specification (Multilevel generalized linear model)
This study applied a multilevel analysis technique to account for the hierarchical/clustering nature of data. The clustering variable was kebele with a cluster size of 18. In this analysis, binary response variables (preterm birth and term low birth weight) were considered for multilevel modeling, independently. A two-level multilevel generalized linear model for binary outcomes was applied in which individuals (level 1) were nested within communities (level 2). The level 1 model represents the association of individual-level factors, including IPI, with the outcome variables. The level 2 model represents the influence of community-level confounding factors on the outcomes. Four models were fitted as follows:
Model-I: It is an intercept-only model or model with no covariate inserted. It is used to check the variability among the communities (cluster-to-cluster variation) and used to give information to whether there is justifiable evidence to consider a random-effect model. Model-II: It is a multivariable model adjustment, containing only individual-level factors, including IPI. Model-III: It is a multivariable model adjustment, containing only community-level factors. Model-IV: It is a final adjustment model, containing both individual and community-level potential confounding factors.
The effect of IPI (fixed-effect) on preterm birth and term low birth weight was expressed by using adjusted relative risk (ARR) with 95% confidence intervals and its public health impact was interpreted using attributable fraction (AF) and population attributable fraction (PAF). AF and PAF are calculated from the adjusted RR (Appendix).
The measures of variation (random-effects) were reported by using Intra-class (community) Correlation Coefficient (ICC) which explains the percentage of variation by community-level factors (level 2). Proportional Change in community Variance (PCV) was also used to express the percentage of changes in the community-level variance between the null model (model-I) and the successive models.