Vitamin K supplementation during pregnancy for improving outcomes: a systematic review and meta-analysis

To study supplementation effect of vitamin K (VK) alone or combined with other nutrients administered to pregnant women, we searched Cochrane Pregnancy and Childbirth Group’s Trials Register (till 22 January 2016, updated on 28 February 2018) including other resources. Two review authors independently assessed randomised or quasi-randomised controlled trials for inclusion, data extraction, accuracy, and risk of bias. We included older trials from high-income countries (six; 21,493 women-newborns), judged mostly as high or unclear bias risk. We could not assess high-risk e.g. epileptic women, but healthy women (different gestational ages) received varying VK dosages and duration. We meta-analysed neonatal bleeding (RR 1.16, 95% CI 0.59 to 2.29; P = 0.67) and maternal plasma VK1 (MD 2.46, 95% CI 0.98 to 3.93; P = 0.001). We found many outcomes were un-assessed e.g. perinatal death, maternal bleeding, healthcare utilization. Mostly newborns were included where VK found significantly effective for e.g. serum VK (mother-newborn), maternal breast milk VK. Few trials reported neonatal adverse side effects. The GRADE evidence quality was very low i.e. neonatal bleeding, neonatal jaundice, maternal plasma VK1. The intervention was favourable for maternal sera VK1 but remained uncertain for neonatal bleeding and other outcomes. The existing literature gaps warrant future investigations on un-assessed or inadequately reported outcomes.


Materials and Methods
We developed the present systematic review and meta-analysis following the Cochrane methodology 31 and PRISMA guidelines 32 . Search strategies. The main searches were performed on the Cochrane Pregnancy and Childbirth Group's (PCG) Trials Register from the inception until 2016 January 22 by a PCG search co-ordinator. The search was populated on major electronic databases: CENTRAL, MEDLINE, Embase, CINAHL, including hand-searches of 30 journals and major conference proceedings, recent awareness alert for 44 additional journals and BioMed Central email alert (http://pregnancy.cochrane.org/pregnancy-and-childbirth-groups-trials-register). Two PCG staff screened the initial results and reviewed full texts of the relevant reports. Before adding to the Register, the intervention detail was used for each report to be assigned with a number related to particular review topics. Using the topic number (not keywords), the Registry search was performed for each review. A precise search set was built using this method and was fully incorporated into this review. We performed an update search (28 February 2018; see Supplementary Information for full search strategies). We also searched bibliographic references of the identified trials, including reports and relevant reviews. We applied neither language nor date restrictions. We contacted researchers in this field for additional information, if needed.
Study selection. Inclusion of the eligible trials was guided by the independent assessment of two of the review authors (SS, KS or NH). In case of disagreement, the review authors either discussed among the team or consulted an arbitrator (EO or RM). We considered quasi-randomised trials (Cochrane Handbook; Section: 13.2.1.1) 31 to avoid finding none or a limited number of randomised trials and due to the likely limitations inherited to randomised trial design such as assessment of adverse events that are rare or need long-term observation. We included studies based on the following inclusion criteria: (1) randomised or quasi-randomised controlled trials assessing VK supplementation during pregnancy; (2) trials presented only as abstracts (3) trials that studied pregnant women as a subpopulation and where applicable data could be extracted; (4) regardless SCIENTIFIC RepORtS | (2018) 8:11459 | DOI: 10.1038/s41598-018-29616-y of pregnancy stage, all pregnant women and their babies were included; (5) oral, intramuscular, or intravenous VK were administered; (6) outcomes e.g. neonatal bleeding; perinatal mortality; maternal bleeding incidence (primary outcomes); still birth; neonatal death; neonatal bleeding; maternal VK deficiency; anemia; and adverse side effects; and healthcare use e.g. special care/admission to intensive care unit and days of hospitalisation (mother-newborn). See protocol 33 for full outcome list. We excluded: (1) cross-over trials; (2) trials assessing VK effect in women undergoing imminent preterm delivery to avert newborn PVH (assessed earlier) 28 . Our intervention group was pregnant women who have received antenatal VK only or combined with micronutrients, without restrictions in dose, intake frequency, and length and timing of birth. Pregnant women were compared in three arms: (1) any VK dosage versus none; (2) VK versus placebo or none; (3) micronutrients containing VK versus micronutrients without VK.
Data extraction and quality assessment. A data extraction form 31 was developed which was used by two of the review authors (SS, KS or NH). We extracted data as per our protocol 33 and PICOS specifications including funding source information. We sought discussion or consultation from an arbitrator (EO or RM), if discrepancies occurred. Using Review Manager program 34 , we entered data and validated data accuracy. We contacted authors of the original papers or content experts for clarity regarding pertinent information. We contacted one research group in Japan (identified from included studies) who provided further information towards our data extraction and analysis plan. Using the Cochrane Risk of Bias (RoB) tool 31 , two of the review authors (SS, KS or NH) constructed independent assessments for each of the trial's methodological quality. We sought discussion or consultation from a third assessor (EO or RM) to resolve disagreement. The trials were assessed in 6 domains: Sequence generation; Concealment of allocation; Blinding of subjects and personnel; Blinding of end point assessment; Incompleteness in outcome data; Selective outcome reporting; Other bias; and Overall bias risk. Following the Cochrane RoB criteria 31 , we developed explicit judgements, whether high, low or unclear bias risk was a concern for each of the trials. For the above domains, we also judged the likely extent and trend of the risk of bias and suggested whether these could affect the findings. Finally, the evidence quality for the following end points was evaluated using the GRADE ranking 35 : perinatal death; newborn bleeding; maternal bleeding; neonatal mortality; preterm birth; neonatal jaundice; maternal plasma VK. We used the GRADEpro program 36 to import findings from RevMan 5.3 program 34 and generated 'Summary of findings' table. Summary intervention effect and quality evaluation for each outcome was created following the GRADE criteria with five methodological consideration: study limitations, imprecision, indirectness, consistency of effect, and publication bias. The evidence level was put down a level lower from 'high quality' since serious (or by two levels for very serious) limitation was found, and was guided by assessment of bias risk, evidence indirectness, serious discrepancy, inaccuracy of effect estimates or probability of publication bias.
Data synthesis and statistical analysis. Review Manager software 34 was used to carry out statistical analysis. We meta-analysed data for neonatal bleeding and maternal plasma VK1. For the rest of the outcomes, we either generated effect estimates as per our protocol 33 or followed a structured data synthesis approach, and presented them in a consistent manner as suggested 31 . We used Tau 2 , I 2 and Chi 2 statistics to assess statistical heterogeneity in each meta-analysis and heterogeneity was regarded substantial if an I 2 reached above 30% and either a Tau 2 went above zero, or the Chi 2 test showed P value of <0.10 which is low 31 . Using the GRADE criteria, we assessed the quality of the evidence (as high, moderate, low, very low) 35 by examining the design and execution of the clinical trials, evidence indirectness, and some additional domains through GRADEpro software. Dichotomous data findings were presented as summary risk ratio and 95% confidence intervals. For continuous data, we presented findings with mean ± standard deviation (SD) as reported by the study authors, or mean difference. If possible, we analysed cluster-randomised trials, heterogeneity and sensitivity analysis, including subgroup analyses and sensitivity analyses on the pooled data to investigate effects of bias risk, multi-arm trials, and publication bias, as recommended 31 . We addressed missing data in the included trials through levels of attrition. If possible, we performed a sensitivity analysis to assess whether large proportions of missing data in the included trials had any impact on the overall evaluation of the intervention. Wherever possible, we analysed outcomes adopting an intention-to-treat approach, i.e. all randomised participants in each arm were included in the analyses, and despite whether the allocated intervention was received or not, all participants were analysed based on their corresponding group of allocation. We derived the denominator of each end point in individual trial by the count of pregnant women randomised minus any women whose end points were identified as missing.
Data availability. All data in this review (and its Supplementary Information document) are available for public use without any restriction.

Results
Our search in the Cochrane PCG Trials Register identified nine papers from seven studies, of which 6 (three RCTs, 3 quasi-randomised trials) met our inclusion criteria [36][37][38][39][40][41] (Fig. 1). By including pregnant women, the trials assessed VK supplementation effect in comparison to no VK or placebo. The trials were published between 1951 and 1993. All trials were tertiary hospital based and from high-income countries: the US, the United Kingdom (UK), the Netherlands, and Japan 36-41 , except one from South Africa 42 . The South African trial 42 was excluded because pregnant women in the study groups were not stratified by mature/premature cases. We included 21,493 participants with varying gestational ages: from 34th gestational week 41 to labour 37,38 , the majority of which were healthy and without a medication use history 36,[39][40][41] . Trials reported oral VK1 with varying dosages and duration e.g. 1 mg/d until delivery and four treatment weeks 41 and intramuscular VK 50 mg between 4 and 12 hours before delivery 37 . Two each of the trials assessed oral or intramuscular VK and compared with no VK, placebo, or no VK shots. Neonatal bleeding (1/3 review primary outcomes) and surrogate VK deficiency markers e.g. serum Osteocalcin (irOC) and hydroxylapatite binding capacity (HBC), plasma VK1, VK2 and protein-induced VK-absence (PIVKA-II) were assessed (added as non-prespecified outcomes). None of the trials investigated the combined effect of multi-vitamins. See Table 1.

Risk of bias in included studies.
We judged the included trials [36][37][38][39][40][41] to be at high or unclear bias risk in overall. A summary risk of bias investigation is presented (Figs 2 and 3). See Supplementary Information for detailed RoB assessment.
Intervention effects. The study of antenatal VK supplementation effect versus no VK or placebo was based on 21,493 pregnant women and newborns from six clinical trials (GRADE evidence summary: Table 2). Wherever possible, we meta-analysed data, presented summary effects, or findings as was reported in the trials.
Low Apgar score at five minutes. Healthy Apgar scores were reported by three studies: from 9 to 10 39 and 8 or higher 36 . Since the measurement time was not specified by these two trials, data could not be pooled together. Apgar scores ≥8 at 1 minute were reported in one 41 . Hypoprothrombinemia. Protein-induced VK absence-II (PIVKA-II) or hypoprothrombinemia, a sensitive marker of VK deficiency, was measured 39 : a significantly less number of PIVKA-II newborns in both VK1 (n = 11) and K2 (n = 12) treated groups were found vs. the controls (n = 10); P = <0.05. Positive PIVKA-II was reported in 9/10 controls. These individual data were pointed as "not detectable. " Non-prespecified outcomes (newborns). Noncarboxylated prothrombin. Noncarboxylated prothrombin (a prothrombin form that is usually found in VK deficient population or those on VK antagonist drugs) 36        Adverse outcomes. Newborns (5/79 treated; 4/190 controls) were reported for admission to the NICU for LBW, asphyxiation, hyperbilirubinaemia, or heart murmurs 36 ; however, they were not separated by diagnosis. Three newborns were reported to have different conditions: septicemia and HDN (placebo arm), and bleeding dyscrasia (treatment arm) 38 . Eight babies were diagnosed with severe jaundice symptom (≥20 mg% bilirubin) and were at risk of Kernicterus development (VK arm) 38 . One trial reported finding no evidence of side effects for VK administration 40 .

Secondary outcomes (mothers).
Hypoprothrombinemia. Oral VK1 and K2 supplementation effect on vitamin K dependent factor PIVKA II was assessed for women 39 , but the data range weren't detectable (original study, Tables 1 and 2).

Non-prespecified outcomes (maternal). Maternal plasma vitamin K.
We converted data from pg/ml to ng/ml unit and pooled two trials 39

Discussion
According to this review, VK supplementation showed no benefits for the reduction of neonatal bleeding but for maternal plasma vitamin K1 status. Antenatal VK exerted beneficial effects for numerous other outcomes e.g. cord serum VK and HBC (mother-newborn), maternal-newborn VK-dependent factors, breast milk VK. Limited comments 38,40,42 on VK induced adverse outcomes were presented by authors.
Overall comprehensiveness and relevance of evidence. Our findings are largely dependent on two trials and the pooled estimates were dominated by single large studies. None of the included trials assessed perinatal death or maternal bleeding, including a set of secondary outcomes. Most of the trials were quite old, i.e. between years 1951 and 1993 and included women at varying gestational ages, mostly at >20 weeks. The interventions ran from minimum half an hour up to 2 weeks prior delivery and used both oral and intramuscular route. Oral route was commonly used. Different vitamin forms i.e. K, K1, and K2 were used and varied in dosages e.g. 1 mg daily were more common. All the vitamin forms and routes detected some level of significance for some of the reported outcomes. Therefore, it is difficult to precisely suggest which particular VK form, administration route, dosage, and intervention duration is most beneficial for women at a particular gestational age. We could only meta-analyse data for neonatal bleeding and maternal plasma VK1. We would pool data 39,41 also for neonatal cord sera; however VK1 levels were undetectable in the placebo arm (<20 pg/ml) 41 . We did not detect any significant effect of VK administration on the reduction of neonatal bleeding (Fig. 4, meta-analysis). In one trial 37 , groups were limited in comparability because of the wide variability in HDN incidence (bleeding from the stomach or the bowel was recorded only and IH was not considered); consequently, authors were unsure whether the findings were due to prothrombin deficiency or other undetermined reasons. Additionally, VK was not received by "many" (not specified) of the treated mothers before delivery, as a result, hurried labour prior to or right after hospitalization were documented by the authors. Maternal plasma VK1 was significantly improved in two small sized RCTs 39,41 (Fig. 5).
Although both the trials detected a significant intervention effect, VK1 levels in the treated mothers were much lower (2.51 ± 37.43 ng/ml) 41 compared to the treated mothers (11.36 ± 13.60 ng/ml) in another study 39 . The lower values could be due to varying dose administration i.e. VK1 (Konakion ® , 1 mg/day) orally until delivery 41 vs. oral VK1 (20 mg/day, Eisai Ltd., Japan) once only for 7 to 10 consecutive mornings until birth 39 . A significant correlation in VK1 concentrations between the maternal and cord sera was reported 39 , with a large gradient (mostly < one-tenth). Very low VK1 in the treated cord sera was also found 41 i.e. 2.5 times below the lower normal adult range, 60 times below the corresponding maternal values. Although high VK1 in maternal sera was found, it remained undetectable (<20 pg/ml) for the placebo mothers and cord samples, for which authors concluded that the placental barrier is truly a large factor for vitamin K transportation to the newborns. For VK1 and VK2, a significantly higher concentration in the treated maternal and cord sera (VK1 only in neonatal cord sera) were reported than controls 39 . Vitamin K1 showed statistical significance for elevating VK-dependent coagulant factors for the intervention babies 36 , where neonates (5 treated, 4 controls) were excluded for NICU admission due to LBW and other diagnosis e.g. hyperbilirubinemia. Both trials concluded that though in lower values, VK1 and VK2 crossed the placenta and continued activating the VK-dependent coagulant factors (minimum until day 5), which increased neonatal VK status, supported by another included trial 39 . Although the treated cord serum HBC (% absorbed) 41 was significantly higher i.e. 53.2 ± 2.20 vs. 35.6 ± 2.70 (placebo); P = 0.00001, the values were markedly lower in all cord samples compared to the mothers, which further indicated newborns usual VK deficiency at birth due to the large placental block. The extremely suboptimal HBC values in all control newborns were comparable to the population who generally use oral anticoagulants 39 . Newborns' immature carboxylase system has less capacity to maintain the creation of complete carboxylated osteocalcin and was indicated among the underlying mechanisms for suboptimal response of fetal osteocalcin to maternal VK consumption, further supporting the need for higher VK supplementation for fetal and neonatal development. Although VK1 benefitted newborns' usual prothrombin (two-stage, 55 hours of life) shortage, bilirubin was indifferent for groups and, due to various reasons, some 50% of the babies could not be considered in analysis 40 . The HBC (% absorbed) in the treated mothers was also significantly higher 41 ; however, in 11/20 control mothers, the values found lower than the normal adult range, for which some of the mothers were thought to have subclinical VK deficiency. In addition, the mothers and newborns' HBC in both groups were significantly correlated (r = +0.59; P = <0.0003), supporting the fact that maternal optimal VK boosts newborns' VK. Furthermore, breast milk from treated mothers showed significantly higher levels of vitamin K1 and K2 levels 39 , with these two groups showing a marked correlation. As a result, greater VK storage in maternal liver was suggested to be beneficial, which is transferred to the breast milk from maternal liver and consequently, help improve newborns' VK deposit.
Evidence quality. Overall, the trials were judged either as methodologically unclear or at high bias risk. The merit of the evidence (GRADE summary of findings Table 2) was very low for neonatal bleeding and neonatal jaundice, which both ranked two levels lower mainly due to high bias risk for participant selection technique in quasi RCTs and for wide 95% CI. Maternal sera VK1 was downgraded by two levels, because risk of bias domains were mostly unclear with wide 95% CI, including small sample size.
Present study findings in light of existing literature. According to the present study, the effectiveness of antenatal VK was not statistically significant for neonatal bleeding, except for maternal sera VK1. In cases other than frequent liver injury, neonatal bleeding including hypoprothrombinaemic condition was previously improved 43 . At delivery, maternal plasma VK was significantly higher, and since women's VK levels normally drop during the last trimester (e.g. 40 to 41 gestational weeks) 39 , VK supplement was suggested to improve maternal osteocalcin carboxylation, including newborns. Antenatal VK showed benefits for other outcomes. Consistent with previous studies 40, [44][45][46] , a large significant effect of oral VK1 for improved blood coagulation factors was detected in mother-newborn pair 36,39 , including cord blood VK 39 and neonatal serum VK 41 . However, possibility of VK dilution in the breast milk of the treated women or VK storage in the newborns liver were indicated by the authors as influencing factors. Neonatal VK deficiency at birth was pointed out as a common phenomenon due to the well-established fact of obstructed placental transfer of VK to the newborns 11 , including low HBC values which possibly appear from poor VK deposit in the newborns 41 . Ensuring an optimal deposit of maternal VK was suggested as a plausible remedy by the trialists to improve newborns' VK status. Nonetheless, administration of VK previously did not show marked benefit for the newborns with declining VK deficiency 12 . Furthermore, significantly elevated maternal breast milk VK1 for VK1 supplementation was observed at delivery 39 , consistent with previous research reporting a 7-fold (after 12 hours) and 2-fold (after 48 hours) increased VK1 in the breast milk for a single oral VK1 dose 47 . Breast-fed newborns who are non-supplemented with VK may develop severe bleeding complication 48 ; however, none of our included trials provided such assessment. Newborns are required to be supplemented because mothers carry extremely low amount of breast milk VK; therefore, greater maternal VK intake has been suggested to significantly enrich breast milk nutrients for offspring's optimal nutrient supply.

Limitations.
Our review evidence is weak on the following points. First, our review is limited to 6 trials which were quite old; and no new eligible trials were identified through our update search. Second, an equal number of papers (3 each) contributed data from RCT and quasi-RCT and therefore, design-wise, the strength of the evidence is not entirely strong. Third, most of the trials (4/6) included small samples and short follow-up time (during labour to 6 weeks prior delivery). Fourth, the available evidence is only from trials conducted in high-income countries and cannot be generalized globally. The only trial identified from a low-income country i.e. South Africa; (quasi-RCT), did not qualify for inclusion in this review. Fourth, information in the reports was often unclear, limited or unavailable. Finally, despite applying an exhaustive search method, it is possible there were additional trials (published or unpublished) in this field that we could have included.

Conclusions
Based on this review evidence, the effect of antenatal VK was not statistically significant for neonatal bleeding reduction except for maternal plasma VK1, including other outcomes e.g. neonatal plasma VK, maternal-newborn PIVKA-II factor and breast milk VK. This review was mainly driven by the need for compilation of systematic evidence on VK deficiency during pregnancy, its beneficial dosages and related morbidities in healthy women including those with epilepsy, malabsorption and other health conditions, as these evidence to date is insufficient. Although VK supplement is not necessary in normal pregnancy, deficiency may occur 1,6 in epilepsy and other impaired conditions 5,10 . We planned to assess these important women subgroups 33 however, none of the included trials provided such observations but either included mostly healthy women with uncomplicated pregnancy or excluded critical cases from the analyses, e.g. stillbirths, congenital abnormalities and LBW babies 40 . We believe that closing these knowledge gaps is particularly essential because epileptic pregnant women tend to experience troubled metabolism or have higher chances of giving birth to newborns with early bleeding and neural tube defects 32,33 . Furthermore, during the last century, the human diet has changed substantially, resulting in extremely low levels of VK deposit in human body 49 . Research cautioned of experiencing adverse outcomes if women's diet was altered during pregnancy 29,30,45 . Moreover, outcomes like neonatal bleeding and neonatal death are rare in high-income settings and could be difficult to detect due to various underlying factors such as healthcare seeking behaviour, socio-economic background, diet availability, and seasonal food variation 42 . It is essential that these factors are discussed in future studies as these might influence the intervention effect, especially in certain minority and low-income women. Future studies should assess outcomes that were not investigated or reported inadequately e.g. maternal bleeding and adverse outcomes, preferably using long follow-up, large sample size, and inclusion of critical subgroups.