Context of parent trial

Conducted in northwest rural Bangladesh between 2007 and 2011, the JiVitA-3 community-randomized controlled trial enrolled 44,567 pregnant women to assess the effect of daily antenatal supplementation with multiple micronutrients, compared to iron-folic acid, on six-month infant mortality (Clintrials.gov #NCT00860470) [46]. The JiVitA-3 study area comprised 435 square kilometers of rural northwest Bangladesh, including 19 unions of Gaibandha and Rangpur Districts in Rangpur Division [47]. The study area was selected as representative of typical rural populations in Bangladesh based on population density (~1000 people per square kilometer), rural, agrarian characteristics (villages surrounded by rice fields, linked by unpaved roads), and economic and public health indicators [47]. As of 2010, Gaibandha District had a population of 2.3 million people, 51% of which was female and 49% male [48]. Literacy among females was reported at 63% [48].

Women were eligible for the parent trial if they could become pregnant (were of reproductive age 13–45 years, married and living with their husbands, not sterilized or menopausal, and whose husbands were not sterilized) [47]. These criteria have been used in trials in this same study area to define women at risk of becoming pregnant [49, 50]. At the outset, a census of all households within the study area was used to generate a comprehensive list of non-pregnant married women meeting above criteria. Eligible women were visited by field workers every five weeks to administer pregnancy tests for missed menses. Women who were identified as pregnant based on urine tests were asked for consent to enroll in the parent trial.

Enrolled women were visited weekly by trained female workers to ascertain pregnancy outcomes (live births, stillbirths, induced abortions, or miscarriages). Those who had a live birth or a stillbirth were administered a birth assessment form two weeks after delivery (standard deviation: 6 days) and asked whether they experienced a series of complications during labor and delivery. All women with a known pregnancy outcome were interviewed one month following the end of their pregnancy and asked whether they had experienced a series of symptoms in the four weeks before, during, or after the pregnancy outcome. Women were administered structured modules using local terms, pretested for comprehension in this setting, to assess symptoms. All women with pregnancy outcomes were asked: “At any time during pregnancy, delivery, or in the first four weeks after delivery, were you so sick that you believe you nearly died?” Data from these interviews was used to define the morbidity categories.

Definitions of morbidity categories

The eclampsia category consisted of report of convulsions (not related to epilepsy) in the absence of high fever, while sepsis was defined as high fever along with foul-smelling vaginal discharge or lower abdominal pain. Report of the baby having been stuck or labor pains greater than 24 hours, along with the use of saline with injection or injection only, was used to define obstructed labor. Women who reported profuse bleeding at birth or retained placenta were categorized as having had postpartum hemorrhage. For women with induced abortions or miscarriages, those who reported vaginal bleeding (other than blood spotting) heavier than a normal period in the three days after a pregnancy outcome were classified with hemorrhage (Table 1). Although the validity of definitions for abortive outcomes was not available from previous studies, we included IMPAC criteria (e.g. bleeding heavier than a normal period) to help distinguish abnormal bleeding that may result from induced abortion [51].

Women who reported that felt that they nearly died anytime during pregnancy, delivery, or 30 days after their pregnancy outcome as well as symptoms of eclampsia, hemorrhage, sepsis, or obstructed labor were considered “near misses.” Sensitivity and specificity values for the near miss category were not available from prior validation studies. Since IMPAC guidelines for obstetric complications exclude injuries [51], women who reported physical injuries, including those sustained from accidents, during pregnancy, delivery, or in the four weeks after their pregnancy outcome were excluded.

Women not meeting the criteria for obstetric complications or near misses as shown in Table 1 were categorized in the non-complicated pregnancy category. If women reported symptoms consistent with more than one morbidity, they were included in multiple groups. While the terms “obstetric complication” and “near miss” are defined by the criteria described above and shown in Table 1, the use of the term “maternal morbidity” in this analysis refers broadly to any illness symptom reported during pregnancy, delivery, or the 30 days following a pregnancy outcome.

Definitions of independent variables

We focused on biological, socioeconomic, and psychosocial factors associated with obstetric complications and/or near misses in the literature. Upon enrollment in the parent trial, women’s socioeconomic characteristics were assessed, including ownership of household assets such as the number of bicycles and construction of homes. To measure socioeconomic status, we used an asset-based wealth index constructed using principal component analysis of household assets and construction materials, as described elsewhere [58]. Enrolled women were also interviewed about their literacy, defined as the ability to read or write a letter in Bengali (the local language).

At enrollment, demographic data such as number of previous pregnancies and the outcome of these pregnancies was also collected. This data was used to define adverse obstetric history as report of an adverse outcome (stillbirth or miscarriage) during the previous pregnancy. Women’s age was self-reported at enrollment. Parity was defined as the number of births a woman reported. Interviewers trained in obtaining anthropometric measurements measured the mid-upper arm circumference (MUAC) of women thrice, with the median value used for analysis. We included standard cutoffs used in the literature to distinguish women at higher thresholds of risk for complications due to age (<18 years and >35 years), parity (nulliparous), and wasting malnutrition in pregnant women (MUAC less than 21.5 cm) [25, 32, 44–50, 59, 60].

Upon enrollment in the parent trial, women were also asked whether they wanted the index pregnancy and whether their husbands had wanted the pregnancy. A ranking of couples’ pregnancy wantedness was created, ranging from both partners wanting the pregnancy, only one partner wanting the pregnancy, and neither partner wanting the pregnancy.

Data analysis

To explore the data, we tabulated frequency distributions for categorical variables and used Chi-squared tests to assess differences between the multiple morbidity categories. Multinomial logistic regression was used to obtain relative risk ratios (RRR) and 95% confidence intervals (CI) to estimate the relationships of independent variables with three morbidity categories defined as those with non-complicated pregnancies (referent category) and those who had obstetric complications and near misses, respectively. Robust cluster estimates were generated to account for cluster-randomization in the parent trial using the Huber-White sandwich estimator [61]. We retained variables strongly associated with development of obstetric complications in the literature (women’s age, obstetric history) as well as potential confounders (year and season of pregnancy outcome) in the base model. Selection procedures based on Akaike Information Criteria values were used to identify optimal regression models. Model fit was assessed by comparing the log likelihoods of the fitted model to the intercept-only model.

Sub-analyses by the type of complication reported (hemorrhage, eclampsia, sepsis, or obstructed labor), was done using logistic regression adjusted for year and season of pregnancy outcome, as recorded in the parent trial. All data analysis was performed in Stata 11 [62].

The JiVitA-3 trial and all interim follow-up studies received ethical approval from the Bangladesh Medical Research Council (BMRC Reference Number 458) and the Johns Hopkins School of Public Health Institutional Review Board (IRB 00000570). The JiVitA-3 trial is registered under ClinicalTrials.gov NCT00860470.

Characteristics of study sample

As a subset of the pregnant women enrolled in the JiVitA-3 parent trial, this analysis drew from the 42,796 women who reported pregnancy outcomes between December 2007 and June 2011 (Figure 1). Approximately 61% of women had live births (n = 26,320) and 3% had stillbirths (n = 1,216), while 25% (n = 10,406) of women had induced abortions and 11% (n = 4,891) had miscarriages. The 42,214 women who completed maternal morbidity interviews (99% of the women with pregnancy outcomes) were considered the analytic cohort for this paper.

Using the definitions in Table 1, 25% (n = 10,380) of women were classified as having obstetric complications, 2% (n = 1,004) with near misses, and 73% (n = 30,830) with non-complicated pregnancies. Hemorrhage was identified as the leading complication, reported by 12% (n = 5,232) of all pregnant women, followed by sepsis (8%, n = 3,259). Of the 27,241 women with live births or stillbirths, 11% (n = 2,950) reported obstructed labor and 1% (n = 328) reported eclampsia.

Bivariate and multivariable analyses

Sub-analyses

Main findings

Using population-based data on symptoms assessed with minimal recall bias from a rural setting of Bangladesh, our analysis shows that a quarter of the women reported obstetric complications, with two percent reporting near misses. Biological risk factors and lack of pregnancy wantedness were associated with increased risk of obstetric complications, while parity was the only factor that was significantly associated with reported near misses. MUAC less than 21.5 cm was a risk factor for reported hemorrhage or sepsis.

Interpretation

The estimates in our study are consistent with the proportions of self-reported complications found in other studies from South Asia. Population-based studies published since 2005 have indicated that between 15% and 38% of women experienced an obstetric complication during their most recent pregnancy [13, 14, 21, 52, 53]. The prevalence in our study (25% of women of reproductive age with recent pregnancies reporting obstetric complications) falls within the middle range of these estimates. Our numbers are closest to the proportions estimated in West Bengal and in Uttar Pradesh, India where 23% of women with pregnancies in the four years prior to survey (n = 3,111) and 21% of women with pregnancies in the year prior to survey (n = 410), respectively, were considered to have intrapartum obstetric complications [11, 12]. This similarity may arise from the common, though not identical, definitions used for obstetric complications.

For individual complications, the proportion of women reporting hemorrhage in our study (12%) falls within the range of the 5% to 24% of surveyed women of reproductive age who reported “excessive bleeding” in studies conducted in South Asia [7, 10–13, 16–18, 20, 53]. Our estimates of sepsis and eclampsia (8% and 1% of women, respectively) lie in the lower range of the proportions reported in previous studies [11, 12, 15, 17, 24, 54]. The lower proportions in our study could arise from our use of more exclusive definitions as well as shorter recall periods.

The prevalence of self-reported near misses recorded in this study (2%) is similar to the 2% of women admitted to intensive care units or delivery wards considered to have near misses in hospital-based studies in Nepal [63], Morocco [64], and Brazil [65, 66]. The prevalence of obstetric complications in our study (25%) is similar to 23% of women considered to have obstetric complications in a hospital-based study in Matlab, Bangladesh, a demographic surveillance site where rates of institutional deliver are higher (50%) than the national average (29%) for institutional delivery [67]. Although our definitions were not based on clinical criteria, we may have captured severe morbidity since we restricted definitions to structured symptoms shown to have high validity in previous studies [55–57].

In our analysis, biological factors appeared to increase the risk of all reported obstetric complications. Research from the JiVitA study area has suggested that adolescents who become pregnant may be unable to meet the increased nutritional demands of pregnancy and lactation [68], which could increase risk for obstetric complications. In this setting, child marriage is strongly tied with early childbearing, with girls who marry early also likely to drop out of school and have low decision-making power in their families [44, 69]. Biological factors such as adverse obstetric history and nulliparity, which have been associated with risk of complications in previous studies, were all significant risk factors in this analysis [11, 24–26]. Women’s age above 35 years has also been associated with elevated risk for pregnancy-related mortality, as shown in a prospective community-based study on 25,580 pregnancies in rural Nepal conducted by Christian et al. [32]. The consistency of biological risk factors in this analysis with previous studies may suggest validity of the assessed morbidities to detect obstetric complications.

Maternal mid-upper arm circumference less than 21.5 cm, a cutoff that may be indicative of wasting malnutrition in pregnant women [32], appeared to increase the risk of reported hemorrhage or sepsis. Christian et al. found that rural Nepalese women with MUAC less than 21.5 cm had a 1.5 increased risk of pregnancy-related mortality [32]. Studies have shown associations between wasting malnutrition and weakened immune system response to infection [68, 70], as well as associations between acute malnutrition and anemia in pregnancy [71]. Data from rural northwest Bangladesh suggests a high prevalence of anemia in early pregnancy, with 26% of 42,896 pregnant women between 2001 and 2007 estimated to be anemic [60]. In this analysis, a high proportion of pregnant women (33%) were estimated to have MUAC below 21.5 cm.

Women’s report of neither partner wanting the index pregnancy was associated with increased risk of reported obstetric complications. Studies have suggested that pregnancy wantedness may indicate the importance placed on the pregnancy and the care and attention given to women throughout pregnancy [22, 23]. Lack of pregnancy wantedness has been associated with decreased attendance at antenatal visits [22, 23]. Only 39% of women in this study indicated that both partners wanted the pregnancy, suggesting substantial unmet need for contraception in this population.

In this analysis, a quarter of the women reported having an induced abortion. While abortions performed under safe conditions incur low risk of complications, unsafe abortion can lead to complications such as hemorrhage, sepsis and chronic morbidity [72]. In Bangladesh, menstrual regulation, the vacuum extraction of intrauterine content to stimulate menstruation and terminate pregnancy [73], is a legal procedure. While menstrual regulation has been credited for a marked decline in abortion-related deaths over the last few decades, abortions induced by untrained TBAs remain a major cause of morbidity in developing countries [72]. A previous study in rural Bangladesh showed that women reported seeking crude methods of pregnancy termination when unable to pay for safer medical abortions [74]. Safe abortion procedures coupled with post-abortion care and follow-up referral, as indicated, need to be advocated and made more accessible to women wishing to terminate their pregnancies in rural areas. Future studies may compare maternal outcomes for women with safe versus unsafe methods of induced abortion.

Although antenatal care attendance is widely recommended for identification of potential risk factors, encouragement of health behaviors, and planning for complications that may arise [29], less than one-third of women (27%) who had live births in our study attended any antenatal visits. Increasing attendance at ANC visits could increase awareness of potential complications among pregnant women as well as encourage emergency preparedness and planning.

At the near-miss level of morbidity, socioeconomic and psychosocial factors were non-significant compared to the biological risk factor of parity. Using data from hospital-based studies, authors have suggested that socioeconomic factors may cease to be important for near misses due to their severity [75, 76]. For women who died from pregnancy-related causes, research from Nepal also indicates that socioeconomic factors did not significantly impact the risk of death, possibly due to the severity of these conditions and the relatively uniform distribution of socioeconomic status in rural areas [32].

Strengths and limitations

This analysis used prospective data on structured symptoms comprising obstetric complications and near misses, assessed within two to eight weeks to minimize recall bias. Detailed morbidity modules facilitated the exploration of results by complication (hemorrhage, obstructed labor, sepsis, and eclampsia) and by type of pregnancy outcome. Data on socioeconomic, biological, and psychosocial factors of interest were collected and available for analysis. The large sample size provided sufficient power for main analyses and sub-analysis.

Since the data in the parent trial were not exclusively collected to determine risk factors for reported complications, we were unable to make direct causal inferences. However, the presented associations exhibited plausible relationships and coherence with previous studies. Since our analyses were based on women’s report of complications rather than clinically confirmed complications, these definitions may have low validity for estimates of prevalence of clinically relevant conditions. However, we applied definitions of obstetric complications recommended by WHO IMPAC guidelines for community settings in order to reduce potential misclassification. Although we were unable to validate the specific definitions used in this study for this population, we have provided sensitivity and specificity values, when available, for similar definitions used in resource-poor settings.