Pre-conception and prenatal factors influencing gestational weight gain: a prospective study in Tigray region, northern Ethiopia

The analyses for this study were performed on data from the KIlte-Awlaelo Tigray Ethiopia (KITE) cohort, a prospective cohort study in Kilite-Awlaelo Health and Demographic Surveillance Site (KA-HDSS) conducted between February 2018 and January 2019 [23]. KA-HDSS is located in the eastern zone of the Tigray region, northern Ethiopia. The site has 113,760 residents in ten rural and three urban kebeles (the smallest administrative units). Women of reproductive age account for 24% of the population. Within the surveillance site, about 4550 pregnancies are expected per year. Most of the population live in rural settings, and agriculture is the primary source of income.

Ethiopia has a three-tier health care system with health posts at the forefront of primary care. There is one health post in each kebele staffed by two to three Health Extension Workers (HEWs). Health posts provide promotional and preventive services under the umbrella of the ‘health extension package.’ The primary delivery modality of the package is through home-to-home visits. The health extension package consists of 16 components concerning family health services, disease prevention and control, and hygiene and environmental sanitation. The components include maternal and child health, family planning, nutrition, proper and safe waste disposal, and food hygiene and safety measures. Health Extension Workers train the households in their catchment area on health extension package and follow the progress of implementing the package after the training [24].

The sample size was calculated to achieve the objectives of the prospective study with respect to pregnancy outcomes in relation to nutritional status. The primary outcome was low birth weight, and the target was to be able to discriminate an estimated proportion of 24.6% low birth weight among women with Mid-Upper Arm Circumference (MUAC) ≥ 23.0 cm and a proportion of 32.6% among women with MUAC < 23.0 cm, as a cut-off to define undernutrition [25], with an alpha of 5% (2-sided), 80% power, and a 10% drop-out rate. The total sample size was calculated at 1100. With this sample size, differences in continuous outcomes > 0.2 standard deviations (SD) could also be detected.

Identifying unmarried pregnant women is complicated by lack of registration that is also required to facilitate follow-up. In the study area, identification of women is based on identifying their respective households, which requires knowing the name of the household head (the husband). Therefore, only married women were included in the study, which was also an opportunity to collect husband information. As height continues to increase during the adolescence period, women aged < 18 were excluded to minimize bias in nutritional data.

An interviewer-administered questionnaire and anthropometric measurements were used to collect the data. Additionally, secondary data were extracted from the KA-HDSS database and antenatal records. The questionnaire was adapted from the literature [18, 25‐29] and was pretested on a convenience sample of 55 pregnant women in Tahtay-Maichew, the central zone of the Tigray region. Qualified Health Extension Workers collected the data. The detailed descriptions of the measurements at different time points are provided below.

Age, residence (urban or rural), religion (Orthodox, Catholic, Muslim or others), educational status (no formal education, primary education, or secondary and above education), occupation (farmer, housewife or others), household size, and economic status were extracted from the KA-DHSS database. The database is updated every six months except for the economic status. For socioeconomic proxy indicator variables, adjustments were made at inclusion when there was a change since the last update. Then, economic status was determined by generating wealth index quintiles designating the lowest to the highest economic status from the proxy indicators. The socioeconomic proxy indicator variables included housing characteristics, access to improved water and sanitation facilities, and ownership of household assets, land, and livestock [30].

Access to improved drinking water sources refers to access to piped water on-premises, public taps or standpipes, tube wells or boreholes, protected dug wells, protected springs, and/or rainwater collection. Similarly, access to an improved sanitation facility was defined as access to an unshared toilet facility, pit latrine with a slab, ventilated improved pit latrine, or flush toilet [31]. Besides access, the time needed to fetch improved drinking water from the nearest source was collected at inclusion. Then, the self-reported time was dichotomized at a cut-off point of 30 min, with the time needed not exceeding 30 min indicating better service [31].

Moreover, access to the health facility was measured at inclusion by asking ‘How long does it take to go to the nearby health facility and back home?’. To assess the implementation of the health extension package, data were collected at inclusion by checking if the women’s respective households were certified as model households or not. Model household is a proxy for implementing the health extension package described above. For a household to be certified as a model, the requirement is receiving short-term training on the health extension package and implementing the package after the training [24]. Additionally, a short form of the International Physical Activity Questionnaire (IPAQ) was used to measure physical activity at inclusion, and the data were summarized as per the scoring protocol [29, 32]. Self-reported history of pre-pregnancy illnesses and perceived work burden were also collected at inclusion. Work burden was rated as easy, moderate, or difficult.

As part of the assessment of reproductive characteristics, the number of previous pregnancies, parity, history of abortion, and history of stillbirth, were extracted from the KA-DHSS database. Other reproductive factors like age at first marriage, age at first birth, a previous inter-birth interval in months, and history of preterm birth, delivery by C-section, and severe perinatal hemorrhage were collected by interview at inclusion. Based on this information, a history of adverse pregnancy outcomes was defined as having experienced one or more of the following: abortion, stillbirth, preterm birth, severe perinatal hemorrhage, or delivery by C-section.

Women were also asked to report on the four-item HITS (Hurt, Insult, Threaten, and Scream) scale at inclusion to assess intimate partner violence. Each question was scored from 1 to 5, and a total score > 10 was used as suggestive of violence [33]. Similarly, women’s empowerment was assessed by asking nine questions addressing three dimensions of empowerment: economic, socio-familial, and legal. The economic empowerment asked the relative income to husband, control over men’s income, control over women’s income, and decision-making on large household purchases. Likewise, the socio-familial empowerment was scored based on decision-making on family visits, decision-making on women’s health and attitude towards domestic violence. In contrast, the legal empowerment assessed women’s legal entitlements over land and house [34‐36]. Then, coding each as 1 or 0, the scores were totaled as women empowerment scores (0 to 9). Furthermore, the question “At the time you became pregnant, did you wanted to get pregnant then, wanted later, or did not want at all?” was asked at inclusion to assess the index pregnancy plan. Accordingly, if the intention was not to get pregnant then, the pregnancy was considered unplanned.

Agrobiodiversity was captured by querying women about a list of food crops and livestock products their households produced in the past year. Then, by counting the number of product groups consisting cereals, roots, and tubers; legumes and nuts; oilseeds; fruits; vegetables; dairy; egg; and meat and poultry, a sum score of agrobiodiversity ranging from 0 to 8 was obtained [37]. Additionally, the amount of produce of each crop, expressed in quintiles was asked, and total harvest volume was calculated by adding the amounts reported for all crops.

A 24-h dietary diversity score was obtained by asking women if they consumed a list of food groups with ‘yes’ or ‘no’ response options. The sum yielded a dietary diversity score ranging from 0 to 10, with scores ≥5 indicating adequate diet diversity [27]. Moreover, to measure household food insecurity, women were asked how often nine specific food insecurity associated conditions, if any, happened in the previous month (0) not at all, 1) rarely, 2) sometimes, or 3) often) [28]. The answers were aggregated to a food insecurity score between 0 and 27. If the responses to all occurrence questions were no or if the affirmative response was only to “did you worry that your household would not have enough food” in a rare frequency of occurrence, households were classified as food secure [28].

Partner support was rated by the five-item Turner Support Scale with each item scored from 0 to 3 [38], and a sum score < 10 was defined as low. Support from significant other social sources was also rated using Oslo-3 Social Support Scale, and scoring ≤8 was considered low [39]. Both were summed up as a total support score. Distress was obtained using the ten-item Edinburgh Postnatal Depression Scale [40], the seven-item anxiety subscale of the Hospital Anxiety and Depression Scale [41], and the four-item Perceived Stress Scale [42]. The depression and anxiety scales were rated from 0 to 3, while the stress scale was scored from 0 to 4. Summing the standardized depression, anxiety, and stress scores, a total distress score was obtained. Additionally, a cut-off point ≥13 as suggestive of high depressive symptoms [40] and ≥ 8 for high symptoms of anxiety and stress were used. To indicate the level of distress, the presence of high symptoms in one, two, or all of the three domains, i.e., anxiety, depression, or stress, were considered.

Anthropometric measures; weight to the nearest 100 g as measured before pregnancy, and height to the nearest 0.1 cm were collected at inclusion using a Seca scale and height-measuring board. Likewise, MUAC to the nearest 0.1 cm was measured using MUAC-measuring tape. All measurements were taken twice and averaged. Pre-pregnancy BMI (pre-pregnancy weight (kg)/[height (m)]²) was categorized as underweight (BMI < 18.5), normal weight (BMI = 18.5 to 24.9), or overweight (BMI ≥ 25.0), and so was BMI calculated from weight and height measured at inclusion. MUAC below 21.0 cm was defined as undernutrition [43].

At this stage, data on self-reported stressful life events that occurred in the past year, attendance of prenatal care, illness during pregnancy, and pregnancy complications were obtained [44]. Prenatal care was defined as adequate plus (five or more visits), adequate (four visits), or intermediate (two or three visits) for women who began prenatal care at ≤16 weeks of gestation. Prenatal care was defined as inadequate if started at >16 weeks or was received only once [45].

Moreover, weight and MUAC were measured as they were measured at inclusion. Also, data on human immunodeficiency virus (HIV) infection, urine analysis, rhesus factor, stool examination, venereal diseases, hepatitis B infection, haemoglobin, and other illnesses were extracted from antenatal records when available. Based on the measures at prenatal care booking, haemoglobin < 11 g/dL was defined as anaemia.

Data were entered into Epi-Data 3.1 and analyzed with STATA (Version 11, Stata Corporation, College Station, Texas, USA). Proportions and means (SD) or medians (interquartile range [IQRs]) were used to describe the characteristics of the study population. Gestational weight gain, obtained by subtracting pre-pregnancy weight from weight measured between 32 to 36 weeks of gestation, was classified based on the 2009 IOM guideline. As excessive weight gain was rare, weight gain was re-categorized into inadequate or adequate. Student’s t-test or Mann Whitney U-test, as appropriate, was used to comparing the mean of continuous variables between women with adequate and inadequate weight gain. To compare the distribution of categorical variables by adequacy of weight gain, a Chi-squared test was used.

The assumption of linear association between gestational weight gain and the independent variables was preliminarily tested with ANOVA comparing mean weight gain by categories of each independent variable. If this test suggested non-linearity, spline regression was applied (Stata adjust-rcspline package), and each independent variable was partitioned into two continuous variables, below and above the knot value (K), using the mkspline command [46]. The coefficient B₂ for the second spline variable represents the change in the effect of the variable above K as compared to below K. The effect of a unit increase in the value of the variable above K can be obtained as B₂ − B₁, where B₁ is the coefficient of the lower partition. The knot value resulting in the best-fitted linear spline regression model, as apparent by the lowest root mean square of errors, was determined by checking the different values of the respective independent variable around a knot value estimated by viewing the linear spline regression curves. Finally, the two variables with their intercepts were regressed against gestational weight gain. If the coefficient for the second spline variable was statistically significant, this was considered to indicate that the effect of values above the knot was significantly different from below the knot, and we concluded that the association was non-linear. When the linear spline regression fitted better than quadratic and cubic models, the two variables and the second intercept were included in the analysis.

Finally, the probability of achieving adequate gestational weight gain by optimizing the foremost factors was estimated based on a multivariable logistic regression model using marginal standardization [47]. All independent variables that were significantly associated with adequate gestational weight gain (P < 0.05, tested two-sided) in the univariable analyses were included in the subsequent multivariable logistic regression model. Residence and MUAC were correlated with other variables and were not included.