Effect of postpartum anaemia on maternal health-related quality of life: a systematic review and meta-analysis

This review’s protocol was registered online with PROSPERO (CRD42020206618) following the Preferred Reporting Items for Systematic Reviews and Meta-analysis [13].

We searched PubMed, Embase, Cochrane and Scopus through the HINARI website. For PubMed, we used English MeSH keywords: “anaemia”, “iron deficiency”, “health-related quality of life”, “depression”, “bonding” and “postpartum women” with attention to possible synonyms, spelling variants, and correct use of truncation and Boolean operators (Additional file 1). Using search terms for PubMed, we developed a search strategy for Embase by entering one term at a time, and a correct term was selected on Emtree (Additional file 1). We restricted our search to studies of human females published in English from databases inception until August 2020. After that, the results were directly exported into EndNote reference management software (Endnote 2017), and all duplicates were removed. We also manually searched references of included articles for additional relevant studies.

Studies were included according to PICO: 1) Population: epidemiological studies (randomised and non-randomised-controlled trials, cohort, case–control and longitudinal cross-sectional studies) that reported either the effect or association between anaemia or iron deficiency during the postpartum period [14] and measure of HRQoL either fatigue or depression or mother–child interaction; 2) Intervention/Exposure: any form of postpartum iron supplementation or haematological test confirming anaemia/iron deficiency and questionnaire confirming PPD or fatigue or mother–child interaction; 3) Comparison: placebo or standard treatment or those that showed HRQoL indicators in anaemic and non-anaemic women and 4) Outcome; effect of (experimental studies) or association (observational studies) of anaemia or iron deficiency on fatigue, depression and mother–child interaction. Other study designs such as cross-sectional, case series, narrative reviews and commentaries were excluded. Additionally, qualitative study designs were excluded from this review.

A standardised, piloted data extraction form was used to extract data from included studies. Two independent reviewers (EM and NP), working in parallel, screened titles and abstracts of the identified articles. After that, full articles were retrieved for further evaluation. Discrepancies between the two reviewers were resolved through discussion. Disagreements between the two were resolved through discussion with MNM and KP. For studies with dichotomous outcomes, we extracted the number of events and participants in each group. We extracted the effect measure, which included both crude and adjusted ratios with their respective 95% confidence intervals and p-values. We extracted means and standard deviation for continuous outcome with normally distributed data while medians, range and p-value of the non-parametric test were extracted for continuous skewed data. We also extracted correlation coefficients for correlation studies and median or mean change from baseline for longitudinal studies.

Two reviewers (EM and NP) independently assessed the risk of bias in the included studies, and discussions resolved disagreements. For randomised trials, we used a revised Cochrane risk-of-bias tool for randomised trials (RoB-2). Thereafter the risk of bias in the individual study was judged as either “low risk” or “moderate risk” or “high-risk bias” [14]. The Newcastle–Ottawa Scale was used to assess the risk of bias for cohort studies and case–control studies [15]. We considered studies rated with ≥ 7 stars as good (moderate risk) [16]. We adopted and modified a tool for evaluating the risk of bias in non-randomised studies of interventions (ROBINS-I) for longitudinal observational studies [17]. We assessed bias due to confounding, selection of participants, missing data, measurement of outcome, and selecting the reported outcomes. We dropped two domains that assess bias due to intervention classification and deviations from the intended interventions as these were deemed not applicable [17].

We graded the quality of evidence using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) [18]. The assessment criteria included; risk of bias, indirectness, imprecision, inconsistency and publication bias. Thereafter, the strength of evidence was grouped as “high”, “moderate”, “low”, or “very low”.

Data analysis was performed using Stata SE for Windows V.14.1 (StataCorp, College Station, Texas, USA). Data has been grouped and analysed separately depending on study design (observation and experimental) and whether the study has reported the effect of maternal anaemia on either fatigue, depression or mother–child interaction. We presented risk ratios (RR) for study level comparison of binary outcomes and mean difference (MD) for the pooled continuous data. We used Mantel–Haenszel random-effects model to calculate pooled RR and 95% confidence interval (CI). A random-effects model was used to calculate pooled MDs. We used recommended formulas to calculate the estimated sample mean and SD from the sample size, median, range and/or interquartile ranges [19]. Wherever fatigue was reported with multiple scales, such as on the multi-dimension fatigue inventory (MFI) scale, the overall “general fatigue” was selected. The I² statistic was used to quantify statistical heterogeneity. We used Cochrane recommendation to interpret I² statistics as “might not be important” (0—40%), “moderate” (30 – 60%), “substantial” (50—90%) or “considerable” (75%—100%). We planned to assess publication bias using funnel plots, but this was not possible because, in each meta-analysis, there were less than ten studies. Thus, we used visual inspection of the confidence intervals. The findings were considered to be statistically significant if the reported P-value was < 0.05.

The searches in Embase, PubMed, Cochrane Central Trial, and Scopus databases through the Hinari website identified 7,547 citations, of which 82 (1.1%) full articles were extracted and assessed for their eligibility. Of the 82 articles, 27 (32.9%) met the eligibility criteria. We further included one article identified by searching the references of the included articles (Fig. 1). Tables 1, 2 and 3 summarise the characteristics and methodological quality of the eligible studies.

Out of 18 studies that reported the effect or association between postpartum anaemia and depression, 15 studies were planned to be included in a meta-analysis. Meyer et al. (1995) [43] was not included as the study reported symptoms of postpartum blues and not depression. Holm et al. (2019) [24] and Güven et al. (2020) [44] were excluded due to a lack of sufficient information. Authors for both studies were contacted but did not provide the missing information by the time submission was made. However, all the three excluded studies reported a significant decrease in depression scores with a corresponding increase in haematological parameters (Table 1).

Ten of the remaining 15 studies were observational studies [9, 21, 25‐32] and 5 were RCTs. [20, 22, 23, 33, 34] The comparison of depression scores in eight observational studies was dichotomous. One study [32] was not included in a meta-analysis as it was not consistent with zero events in the non-exposed group. The pooled results of the remained seven studies [21, 25‐32] showed that iron deficiency or anaemic women were 1.66 times more likely to experience symptoms of depression than the non-anaemic or iron-replete women and the findings were statistically significant [(heterogeneity estimate: I² = 67.0%, p < 0.01), and RR = 1.66 (95% CI: 1.28; 2.15)] (Fig. 2). Two observational studies [29, 31] measured depression as a continuous outcome, and their findings were not pooled together but reported separately (Table 1).

Similar findings were observed in four of five RCTs [20, 22, 23, 25] that measured depression as a continuous outcome. The pooled results in these four studies [n = 739 (treatment group = 372 and control group = 367)] showed a significant reduction in depression scores for women who were given intravenous iron than oral iron or placebo group based on random mixed model (MD = -1.48, 95% CI: -2.53; -0.42). However, heterogeneity was considerable high (I² = 61.0%, p = 0.05) (Fig. 3). Perello et al. (2014) [21] was not included in the meta-analysis as the outcome measure depression was reported differently with others ( categorical outcome) and findings in this study are present in Table 1.

Eleven studies reported either the effect or association between postpartum anaemia and fatigue (Table 2). Four studies [27, 32, 38, 39] were excluded from a meta-analysis due to insufficient information. Westad et al. (2008) [36] reported marked improvement in physical fatigue scores corresponding with rapid iron stores replenishment in women in the intervention group than the control group at 4, 8 and 12 weeks postpartum (p = 0.02, p = 0.02 and p = 0.03 respectively). However, it was excluded from analysis as the author did not provide missing information until submission.

Pooled continuous data from three RCTs [20, 22, 24] (Fig. 4) showed statistically significant reduction of fatigue scores in women who received iron supplementation or vitamins that led to haematological increase than the control (MD: -1.85, 95% CI: -3.04; -0.67) but with considerable heterogeneity (I² = 65.0%, p < 0.06). Fatigue scores were reported as a categorical outcome in two studies. Hatzis et al. (2003), [37] in their matched intervention study in Greece, reported that women who received recombinant human erythropoietin reported fewer clinical symptoms of fatigue (p = 0.0012) compared to oral iron. Conversely, Chandrasekaran et al. (2018) [9] reported no significant association between anaemia and maternal functional status (OR: 1.03, 95% CI: 0.34, 2.94) (Table 2).

Four studies reported the effect of postpartum anaemia on mother–child interaction (Table 3). Murray-Kolb et al. (2009) [40] reported that at nine months’ post-intervention, women who were not anaemic and the IDA-ferrous groups significantly improved scored significantly better on maternal sensitivity, non-hostility, and structuring scales and child responsiveness scale than did the IDA-placebo group (p-value = 0.007), whose iron stores remained low. Similarly, Perez et al. (2005) [41] reported that anaemic mothers in the IDA-placebo group had negative statements towards their infants, less goal setting and responsiveness than mothers in non-anaemic and IDA-ferrous groups (p < 0.05) at nine months postpartum.

Unlike the above findings, Hamm et al. (2020) [23] reported that significant improvement in Hb levels in women who received multiple units of RBCs (8.7 g/dl versus 7.8 g/dl) did not significantly improve maternal attachment scores. Dearman et al. (2012) [42], in their pilot case–control study enrolled 115 women (Hb < 10.5 g/dl = 57 and non-anaemic = 58) and reported no statistical difference in maternal perception of mother-infant bonding between the anaemic and non-anaemic group.

We used the GRADE approach to judge the strength of the evidence. Table 4 has provided the GRADE summaries with an overall quality evaluated as “moderate”. The risk of bias was judged to be moderate, and inconsistency was assessed using the heterogeneity statistics.

Our study has two major limitations. Firstly, we limited our search to human studies published in English language only because none of the authors is conversant with other languages. Secondly, we did not produce funnel plots to assess publication bias in the included studies, as less than ten studies were included in each meta-analysis.