This study is a secondary analysis of data collected during the course of a cluster randomized, double-masked, placebo-controlled, community trial (JiViTA-1) of the effects of vitamin A or beta-carotene supplementation on all-cause, pregnancy-related mortality [15, 16]. The ClinicalTrials.gov identifier for that previous study was: NCT00198822, registered on 9/12/2005. JiVitA-1 was conducted between August 2001 and October 2007.
Study setting
This trial took place in 19 rural unions of Gaibandha and Rangpur Districts in northwestern Bangladesh, comprising an area of ~ 435 km2. The site was selected to be typical of general living conditions in rural Bangladesh and eastern Gangetic region of South Asia. At the time of the study, about 40% of the married women of reproductive age in the area were literate, with < 5% having completed the 10th grade and < 40% were involved in income-generating work on their own. Husbands were mostly employed in subsistence agriculture, hired themselves out as daily wage laborers or ran a small business. Less than 15% of households had access to electricity, and households were mostly made of earth or tin.
Pregnancy ascertainment
Married women of reproductive age (13–44 years) who were neither sterilized nor menopausal and living with their husbands were recruited and placed under pregnancy surveillance [14]. All women were visited every five weeks to ask about their menstrual history during the past 30 days and offered a urine test if amenstrual to identify incident pregnancies. After obtaining verbal consent (due to illiteracy among many participants), pregnant women were enrolled in the trial, visited for an enrollment interview and began receiving supplements as per their allocation.
Data collection
Participants were interviewed by trained interviewers at inclusion (early pregnancy, median gestation age 9 weeks) to collect information on household socioeconomic factors (education, asset ownership, house construction), maternal nutritional status, diet, morbidity history, previous pregnancy history, wantedness of the incident pregnancy, hygiene, physical work history as well as tobacco and alcohol use [17]. Women were tracked through the course of their pregnancies and all women were interviewed again at 3 and 6 months post-partum and asked about labor and delivery, diet, morbidity history, work history and tobacco and alcohol use. In addition, from February 2004 to July 2017, a module of questions related to depression and suicide was introduced into the study and asked independently at two different times of eligibility, during the latter half of pregnancy (Median = 24.4 wks, Mean = 25.2 wks, SD = 2.1), and at ~ 6 months post-partum(Median 32.1 wks, Mean 32.4 wks, SD = 1.06 wks). All interviews were conducted one-on-one in women’s homes and sufficient community counseling ensured that interviews were private and confidentiality could be maintained. If a woman could not be interviewed alone at a specific time, the interviewer rescheduled the visit until it could be conducted privately.
Data were collected at enrollment about unintended pregnancy from the woman for herself (“Did you want to become pregnant now?”) and for her husband (“Did your husband want you to become pregnant now?”). Options included “wanted now”, “mistimed” or “unwanted”, asked approximately one week after a positive pregnancy test (Median = 1.0 wks; Mean = 11.6 wks, SD = 2.4). Pregnancy-intention discordance was defined from the mother’s report of her intentions and those of her husband. For the purpose of the couple’s discordance analysis, mistimed and unwanted pregnancies were combined and defined as “unwanted.” Using ‘wanted now’ and the collapsed category of “unwanted” pregnancy (for the purpose of the discordance analyses), couples’ discordance was defined as a non-match between the mother and her husband’s intentions (not either both “wanted” or both “unwanted”).
Because there was no validated depression scale in Bangladesh at the time of JiViTA-1, depressive symptom items were adapted from the nine-item Patient Health Questionnaire (PHQ-9) and the Center for Epidemiologic Studies Depression Scale (CES-D) [18, 19]. Selection of items (six prenatal and five postnatal) was based on pretesting in the study area. Questions were translated by a professional translator and an independent back-translation was done between Bangla and English to ensure that the meaning of the questions was retained. Women in the area struggled to grasp the meaning of some questions on the PHQ-9 and the CES-D, and so only questions that were clearly understood in focus group discussions were included. Two standard questions about suicidal ideation and suicide attempts were also included.
Maternal depressive symptoms during the latter half of pregnancy were measured using the following six symptoms reported in the last 30 days (yes/no): feeling sad all the time; becoming more forgetful; crying all the time; having thoughts of hurting yourself; sleeping more than before; and trying to hurt yourself. Postpartum maternal depressive symptoms were measured by asking women the first four items used in the prenatal period (listed above), along with a fifth symptom, not wanting to bathe or eat for several days in the last 60 days.
Socio-demographic and other covariates included: living standards index in quartiles (1st = low, 4th = high), maternal education (none, 1–9, ≥10 years), maternal literacy (yes/no), maternal age at enrollment (≤19, 20–29, ≥30 years), maternal nutritional status during pregnancy (mid-upper arm circumference in the third trimester: < 21.5; ≥21.5 cm), anemia in the first trimester (symptoms of breathlessness at rest resulting in inability to work from the World Health Organization maternal screening questions), and infection in the first trimester (either urinary tract infection or pneumonia or both). [20] The living standards index was created using Principal Components Analysis, combining several household asset variables (toilet facilities, type of walls, kitchen, and roof, the number of clocks, living rooms, closets, beds, radios, irrigation pumps, televisions, rickshaws, and having electricity) [21]. Participants were classified in vitamin A, beta-carotene, and placebo groups representing their micronutrient supplementation assignment in the original trial.
The study protocol was reviewed and approved by the Johns Hopkins Bloomberg School of Public Health Institutional Review Board (IRB, #H.22.01.01.11.A1) and the Bangladesh Medical Research Council (BMRC, BMRC/ERC/1998–2001/2405). The informed consent process included, when so desired by the subjects, husbands, in-laws and sometimes community members. Individual consent was sought from the participating women and documented.
Data analysis
Descriptive demographic and pregnancy intention statistics were summarized for women in the subsample of the trial cohort who received prenatal and postnatal depressive symptoms assessments.
We restricted our analyses for prenatal depressive symptoms to women interviewed after the 23rd week and before the 37th week of gestation, in order to make sure that women were interviewed during a common time window, after a spontaneous loss (miscarriage) and before the very end of the pregnancy when hormonal changes related to pending delivery might affect their responses. Analyses were further restricted to women with singleton live births. The Cronbach alpha for prenatal symptoms with all items was 0.57 (unexplained variance = 0.67), and 0.63 (unexplained variance = 0.61) after removal of the item ‘sleeping more than before.’ The remaining five items were summed to achieve a total score of 0–5, with a score ≥ 3 considered our higher level of depressive symptoms. Only women with data on all five items were included in the analysis. Cronbach’s alpha was 0.71 (unexplained variance = 0.49) for the five postpartum depressive symptoms. Items were summed and a cutoff of ≥3 was our basis for dichotomizing depressive symptoms based on the fact that this was the cutoff that conservatively optimized the number of women with symptoms (8% prenatally, 13% postnatally) that would be expected in rural Bangladesh.
Bivariate risk ratios (RRs) and 95% confidence intervals (CI) were calculated using a generalized linear model with binomial error structure and logarithmic link function. Unintended pregnancy included mothers’ and their husbands’ intentions separately (wanted and mistimed versus unwanted) and discordance between intentions (comparing unwanted by the mother/perceived as unwanted by the husband; unwanted by the mother/perceived as wanted by the husband; wanted by the mother/perceived as unwanted by the husband; wanted by the mother/perceived as wanted by the husband). Adjusted RRs were calculated for intentions and depressive symptoms at both time points including potential confounders and covariates of theoretical significance. We did not adjust for prenatal depressive symptoms in models for postpartum depressive symptoms due to the fact that we consider prenatal depressive symptoms potentially on the causal pathway, e.g. the negative effects of unintended pregnancy might first cause depressive symptoms during pregnancy which are sustained after the birth.
All models adjusted for geographic clustering sampled in the original study design and mothers’ assigned vitamin supplementation group in the trial. All adjusted analyses at six months postpartum controlled for child sex. We were unable to include arm circumference, anemia and infection variables in all models simultaneously because of small numbers in some categories. Quasi-likelihood under the Independence Model Criterion Goodness of Fit tests indicated that arm circumference and anemia yielded the best fit; these two variables were included in our multivariable models.