Anemia and adverse outcomes in pregnancy: subgroup analysis of the CLIP cluster-randomized trial in India

The CLIP India trial was a prospective, cluster randomized control trial that took place in 12 clusters in Belagavi and Bagalkote districts, rural Karnataka. The units of randomization were primary health care centers and were chosen by the site teams based on a variety of feasibility and logistical considerations. Four clusters (two per arm) were included in a pilot phase from February 1^st, 2014 to October 31^st, 2014. These four clusters were then joined by the additional eight for a two-year definitive phase from November 1^st, 2014 to October 31^st, 2016.

Participants were married pregnant women (ages 15–49) who provided written informed consent for data collection. In both the intervention and control clusters, data was obtained via household and facility-based surveys. These surveys were conducted by community health care workers and research staff to collect socio-demographic characteristics, care seeking behaviours, maternal outcomes and neonatal outcomes. Data was collected at three time points, (i) as soon as possible after enrolment after pregnancy confirmation (focus on obstetric and medical and previous pregnancy history), (ii) as soon as possible after delivery (focus on care seeking, delivery information and pregnancy outcome for current pregnancy), and (iii) within 42 days postpartum (confirmation and update of pregnancy outcome). Data collection was done through the Maternal and Newborn Health (MNH) registry system [21].

Women in the intervention clusters received additional ‘CLIP visits’ during both the antenatal and postnatal periods. These visits were conducted by community health care workers and were guided by the PIERS on the Move (POM) mobile health technology risk stratification tool [22]. Depending on blood pressure measure and other risk factors, POM provided recommendations around need to 1) continue routine care, 2) seek non-emergent care (within 24 h) or 3) seek emergency care (immediately). Further, in cases with high blood pressure (> 160/110) or preeclampsia, oral methyldopa (750 mg) or intramuscular magnesium sulphate (10 g) was administered, respectively. The intervention clusters also had community engagement sessions for women and other community members to promote knowledge about risks and symptoms of pregnancy hypertension.

The inclusion criteria for the analyses were women who had delivered by the end of the trial (October 31^st, 2016) with a hemoglobin measurement at trial enrolment.

Hemoglobin level was collected at trial enrolment, which for the majority of women was during the first trimester. Hemoglobin was assessed using Sahli’s method. Maternal anemia was defined as hemoglobin (Hb) < 11 g/dl and severity was further subdivided based on World Health Organization (WHO) standards into: mild (10.0 g/dL ≤ Hb < 11.0 g/dL), moderate (7.0 g/dL ≤ Hb < 10.0 g/dL) and severe (Hb < 7.0 g/dL).

We analyzed several maternal outcomes. First, we looked at those that have been commonly associated with anemia, including blood transfusion, antepartum hemorrhage, maternal sepsis, and postpartum hemorrhage. Outcomes were self-reported and clinically adjudicated by a group of (non-treating) physicians for accuracy as part of the CLIP trial [20]. The only exception was postpartum hemorrhage, which was collected within the MNH registry, which is based on treating physician diagnosis. Second, we assessed pregnancy hypertension (defined as systolic blood pressure ≥ 140 and/or diastolic blood pressure ≥ 90). Blood pressure data were only available on women in the intervention arm, so analyses of hypertension were restricted to this group. Data were collected during POM visits using standardised methods and a semi-automated, pregnancy-validated digital device (Microlife BP 3AS1-2) [23]. In addition to defining hypertension based strictly from POM measurements, we also used a previously published definition (for this data) to combine information from POM and hypertension reported in trial surveillance data sources to assess an expanded version of hypertension, as well as pre-eclampsia [24]. Pre-eclampsia was defined as gestational hypertension with proteinuria or 1 or more relevant end-organ complications [24].

Secondary outcomes included severe perinatal outcomes including perinatal death, stillbirth, early and late neonatal death, neonatal morbidity, and a composite of these outcomes.

Possible confounders were identified based on expert knowledge and possible relationship between anemia and adverse outcomes [25] (Figure S1). These include trial arm, cluster, maternal age, nulliparity, body-mass-index (measured at enrolment), maternal and husband basic education (as measure of socioeconomic status), gestational age at booking (as a measure of access to care), religion (as a possible proxy for vegetarian diet), and twin pregnancy.

Information on iron supplementation was only collected postpartum, and timing of initiation was not ascertained. We did not adjust for iron supplementation as it occurred after exposure and therefore may be on the causal pathway between exposure and outcome. Furthermore, almost all women (> 99%) who delivered reported taking supplementation, but timing and adherence were not available.

The rates of women with no, mild, moderate, and severe anemia were estimated. Demographic and clinical outcomes were stratified across anemic groups and summarized as counts and percentages for categorical variables and medians and IQR for continuous variables. For each anemic group, iron supplementation rates were compared between women with miscarriages and MTP and those whose pregnancies ended in live or stillbirth. Women with medically terminated pregnancies or miscarriages were excluded from analysis of maternal and perinatal outcomes.

We used Targeted Maximum Likelihood Estimation (TMLE) to estimate risk ratios and risk differences between non-anemic women and women with mild and moderate anemia on our primary outcomes of interest [26]. TMLE has the advantage over traditional regression (e.g., logistic) models in that it combines a propensity score model for the exposure (anemia) and a model for the outcome and if either of these models is correctly specified the estimate of association is unbiased; this is known as a doubly robust estimator [27]. Further, TMLE allows one to use nonparametric algorithms which do not make modeling assumptions (e.g., linearity, no interaction etc.) that are common in standard regression models. In these analyses we used an ensemble (i.e., ‘stacking’) of prediction algorithms. These included standard regression models both with and without interactions, generalized additive models, mean outcome models and tree-based algorithms. These models were used in both the exposure and outcome models. All results are presented with 95% confidence intervals estimated via the delta method and were adjusted for the clustered nature of the trial data via previously published methods [28].

To assess possible dose–response relationship between hemoglobin level and adverse outcomes we included baseline hemoglobin as a continuous variable in mixed effects logistic regression models to estimate adjusted (for all above confounders) dose–response curves. We used restricted cubic splines with 3 degrees of freedom to allow for non-linearity in the hemoglobin/outcome relationship and plotted marginal risk curves for each outcome. Due to software limitations, TMLE is not available for continuous exposures.

We used E-values to assess the strength of an unmeasured confounder (such as dietary or nutritional intake) between baseline hemoglobin and adverse outcomes (on the risk ratio scale) that would be needed to explain away our results [29]. These were applied to all points estimates and lower bounds of the 95% confidence intervals when these values were > 1.0.

All perinatal outcomes were analyzed similarly to the above. All data analysis was conducted using R statistical software version 4.0.3 [30] TMLE was conducted using the drtmle package. Following guidance from the American Statistical Association and current practice in leading epidemiological journals, no null-hypothesis significance tests were performed [31].

Most women in the CLIP India trial had some form of anemia at the time of antenatal care booking, most commonly moderate and rarely, severe. Almost all women received iron supplementation by the time of delivery, regardless of baseline anemia, and in line with Indian guidelines for routine supplementation. In addition, just under 5% received blood transfusion, which was a greater number than those who experienced antepartum or postpartum bleeding, Compared with a normal hemoglobin level, each of mild, moderate, and severe anemia more frequently prompted receipt of maternal blood transfusion. Women with mild-moderate anemia had no excess of adverse pregnancy outcomes, including bleeding and sepsis, although women with severe anemia was associated with an increased risk of PPH. The findings were unlikely to be related to unmeasured confounders, particularly for severe anemia. However, we observed a ‘U-shaped’ relationship between anemia severity and pregnancy hypertension, and pre-eclampsia specifically, with women with mild- or moderate anemia having significantly lower rates.

Given the near 100% reported rate of iron supplementation in delivered women in this cohort, these data represent the estimated effect of early pregnancy anemia on outcomes in a population where public health efforts for supplementation in pregnancy have been successful.

The prevalence of anemia in pregnancy in India is amongst the highest in the world. Estimates have varied from a substantial proportion to most women in pregnancy, as in our study. In South India, Vindhya et al. estimated the prevalence of anemia in pregnant women to be 33.9%, with 48.4% of cases being mild, 49.5% moderate, and 2.1% severe [5]. Other studies have reported even higher rates of maternal anemia in India, up to 88%, similar to our study [2, 5] that demonstrate higher than previously published data from Belgaum and Bagalkote [32].

Many studies in India and elsewhere have examined the impact of maternal anemia on perinatal outcomes, such as stillbirth, low birth weight, small for gestational age, and preterm birth [8, 9, 33‐36]. Far fewer studies have examined the impact of maternal anemia on maternal outcomes, even by systematic review of LMIC data [6, 15]. One recent review found a 190% increase in the odds of blood transfusion in anemic women [37], but they did not stratify by severity of anemia, as in our study.

The relationship between anemia and anemia severity on gestational hypertension and pre-eclampsia has been mixed. A large cohort study from India and Pakistan (110,033 anemic women) found a U-shaped relationship in Indian women (RRs = 1.89 95% CI = 1.12 to 3.18) for severe anemia, but ~ 1.0 for mild/moderate) but not in Pakistani women (RR = 1.18, 95% CI = 0.81 to 1.73 for severe anemia) [15]. Another smaller study case–control from Sudan (606 anemic women) found increases in pre-eclampsia severe anemia (OR = 3.6, 95% CI = 1.4 to 9.1), as well as possible increases with mild or moderate (OR = 1.60, 95% CI = 0.80 to 3.40) [38]. In contrast, Jung et al.’s review of nine studies found little difference in pre-eclampsia among anemic women (OR = 1.15 95% CI = 0.80 to 1.64); however, they did observe a U-shaped dose–response relationship [37] similar to ours and to the Indian cohort referenced above [15]. Mechanisms for the relationship between higher hemoglobin values and pregnancy hypertension may include poor nutrient supply to the placenta due to increased blood thixotropy, and production of reactive oxygen species, together with increased iron [39]. On the other hand, there has been suggestion that increased corticotropin released hormones associated with low hemoglobin may increase maternal and fetal stress, which can cause pregnancy hypertension [37, 40]. That said, published data on biological mechanisms are limited and further studies are needed to better understand this relationship.

It is clear that iron supplementation can increase hemoglobin values in pregnancy, but effectiveness is challenged by poor adherence, continuous access to iron tablets, use of doses lower than necessary, and oral (vs. parenteral) route of administration [41‐43]. However, even when effective, the impact of anemia treatment on adverse maternal and perinatal outcomes remains uncertain. The most recent Cochrane review found that few studies have assessed the impact of iron supplementation on APH, blood transfusion, sepsis, or pre-eclampsia, and most did not focus on anemia in early pregnancy [18]. Our findings suggest that current management strategies for mild and moderate anemia are largely effective in reducing risk to that comparable to women with normal hemoglobin. However, more research is needed to understand the increased risk of hypertensive pregnancy in those with normal hemoglobin, and whether iron supplementation is related.

Our study has several strengths, including our large prospective cohort, standardized collection of blood pressure data using a device validated for use in pregnancy and pre-eclampsia, and detailed documentation of other pregnancy outcomes and possible confounders of the anemia-outcome relationship in a high-quality cluster randomized trial. Also, we report findings in the context of successful public health efforts to provide iron supplementation to all pregnant women in India, where malaria is not endemic.

Our study also has some limitations. First, our sample size of severely anemic women was small, and therefore despite the large effect sizes, caution should be taken in the extrapolation of findings related to this group, as they may be subject to sparse data bias. Conversely, just over 11% of women had normal hemoglobin, and few had a hemoglobin above 15 g/dL to fully evaluate the U-shaped relationship between anemia and pregnancy hypertension. Second, we lacked data on the exact timing of initiation of iron supplementation and associated adherence; stratification by such information may have provided further nuance to our findings. We found low rates of supplementation in miscarried or MTP pregnancies, which occur (by definition) before 20 weeks. Therefore, it is possible that most women began supplementation after this point. Third, we had no information on the reasons for transfusion, but in all groups, the number of women transfused exceeded those with clinical bleeding. Fourth, this is a secondary analysis of data from a clinical trial with different primary objectives, so we did not measure hemoglobin levels throughout pregnancy, or collect information about potential dietary mechanisms. Although we did adjust for vegetarian diet, there are risks of residual confounding due to unmeasured differences (such as diet and nutrition) between anemic and non-anemic women. Our E-values indicate that such variables would need to have strong effect above and beyond our adjustment to explain away the increased risk for severe anemia. For moderate and mild anemia's relationship with hypertension, E-values were more moderate (1.5–2.0), but still unlikely to explain away the U-shape. Finally, as all participants were prescribed iron supplementation, so we cannot comment on the relative benefits and risks of iron supplementation vs. no supplementation.