This article was reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) statement (22). From the period of their creation until August 2022, PubMed, Web of Science, the Cochrane Library, and Embase databases were used in our literature searches. The following search phrases were used: “vascular calcification or Calcification, Vascular or Calcifications, Vascular or Vascular Calcifications or Vascular Calcinosis or Calcinoses, Vascular or Calcinosis, Vascular or Vascular Calcinoses” and “Vitamin k1 or Phytonadione or Vitamin K1 or Phytomenadione or Phylloquinone or Phyllohydroquinone or Aquamephyton or Konakion or Vitamin k2 or Menaquinones or Vitamin K2 or Menaquinone or Vitamin K Quinone” and “Randomized Controlled Trial or random^*.” For more details regarding the search strategy used for each database, please refer to Supplementary Document 1. According to PICOS criteria to develop a retrieval strategy summarized in Table 1.

The primary outcome was artery calcification: the change in CAC scores, artery volume, and others. We identified the following as suitable methods for evaluating VC: ① plain lateral abdominal X-ray, ② computed tomography (CT) measuring vessels and valve calcification, volume or mass calcification scores, and ③ ¹⁸F-Sodium Fluoride Positron Emission Tomography (¹⁸F-NaF PET) imaging. The secondary outcomes were dp-ucMGP and adverse events.

Two unbiased reviewers Te Li and Yun Wang used a study selection sheet to assess the articles acquired through the search strategy. Studies that did not fit the inclusion criteria were excluded after the titles and abstracts were initially screened. After a preliminary examination, the studies that had not been excluded were retrieved for full-text screening. It was decided whether the study should be considered for inclusion in our analysis based on the inclusion criteria. During the whole process, if there is a dispute, the final decision for inclusion was made by consensus by Weiping Tu.

The data were extracted by two unbiased reviewers Te Li and Yun Wang, including study characteristics (authors, publication year, country, and the number of centers), study participants (such as eligibility criteria and baseline characteristics), and study design traits. Others including study interventions, controls, period of treatment, length of follow-up, and study results were also extracted.

The intervention quality of the studies was evaluated for the risk of bias using the Cochrane Risk of Bias tool for RCTs (23). The risk was estimated using the six criteria listed below: adequate sequence generation, allocation concealment, blinding, incomplete data outcome, and selective reporting. According to these criteria, the risk was classified as low high, or unclear risk of bias. Sensitivity analyses and subgroup analyses were conducted to minimize inter-study heterogeneity. Sensitivity analyses were performed by eliminating one study at a time. Based on participants (hemodialysis vs. no-hemodialysis) and the date of publication (before 2015 vs. after 2015), subgroup analyses were undertaken.

Our article was produced utilizing the Review Manager version 5.4, Stata version 17, and SPSS version 26.0 (IBM Corp, Armonk, NY, United States) for analysis. For dichotomous or polytomous outcomes (adverse events), risk ratios (RRs) were calculated. The use of mean differences (MDs) and 95% confidence intervals (CIs) was evaluated for continuous variables. From the treatment and control groups, MD and standard deviation (SD) for VC and dp-ucMGP were extracted. If MD and SD for outcome indicators of interest were not reported, other available data were used for calculation. The MD and SD in VC were calculated from the median, interquartile range, and sample size. SD of VC or dp-ucMGP was calculated using standard methods based on the mean and 95% CIs or mean and p-value. Cochrane's Q-test and I² statistic were applied to examine the statistical heterogeneity (≥75%, high heterogeneity; 51%−75%, moderate heterogeneity; 26%−50%, low heterogeneity; and ≤25%, insignificant heterogeneity). It is worth noting that the I² estimates with 95% CIs can be used to assess heterogeneity, but it is possibly associated with fluctuation in meta-analysis with <15 trials (24). For the data analysis, we employed a random effect model if there was significant heterogeneity. If there was no significant heterogeneity, we chose a fixed effect model. Statistical significance was defined as a p-value of <0.05.

We conducted a comprehensive search across multiple databases and identified 332 articles. After removing duplicates, we excluded 242 studies based on title and abstract screening, leaving 19 studies for full-text review. After excluding three studies that lacked a control group, one article with duplicate data, and one study that used incorrect interventions, we included 14 RCTs in the final analysis. Figure 1 provides a summary of the included studies.

Our study included 14 RCTs with 1,842 participants initially enrolled. However, due to reasons, such as automatic withdrawal, adverse reactions, loss to follow-up, and other factors, the final number of participants included in the analysis was 1,533. Among them, four studies used VK1, nine studies used VK2 (MK-7), and one study only specified the use of VK without specifying the type of supplementation. Four of the included studies focused on type 2 diabetes mellitus (T2DM), while five studies focused on CKD patients (including both hemodialysis and non-hemodialysis patients). Two studies were conducted on postmenopausal women, one study focused on kidney transplant recipients (KTR), and the final two studies involved healthy individuals. Table 2 presents the main research characteristics.

Overall, seven studies presented a low risk of bias between all parameters. The research design methods in the Knapen et al. (25) study were poorly described. Sequence generation and allocation concealment were rated as ambiguous and highly biased in Oikonomaki et al.'s study (26). One trial used an open-label study design, which places it at high risk of bias in terms of participants and staff blinding (27). Additionally, it is unclear whether outcome data are completed for the study conducted by Zwakenberg et al. (28) with Bartstra et al. (29). Similarly, two studies made no mention of allocation concealment or blinding of outcome assessors for all outcomes (30, 31). Finally, Figure 2 presents the quality evaluation of the included RCTs.

We conducted a sensitivity analysis to investigate the effectiveness and safety of VK supplementation on VC and assess the stability and reliability of the meta-analysis results. After successively omitting each study, we found that the overall results had no significant impact, indicating that the conclusions drawn were stable and reliable. Figure 4 provides a summary of the results.

Seven studies were compared for evaluating the impact of VK supplementation on dp-ucMGP. However, the results exhibited a high degree of heterogeneity (I² = 71%). We performed a meta-regression with the MD in dp-ucMGP as the dependent variable and the sources of heterogeneity. Meta-regression revealed that there was no significant effect on VC of VK supplementation forms or doses, year, and duration of follow-up on univariate analysis (Table 4). There were insufficient observations on the country and population. Subgroup analyses were performed based on population (hemodialysis vs. no-hemodialysis) and country (the Netherlands vs. other countries) of dp-ucMGP. Similarly, a subgroup analysis showed that country and population are not the sources of heterogeneity (Figure 5). Due to the large differences in study outcomes and the small number of included articles, we did not test for publication bias.

In this article, we testify that VK supplementation slows the development of CAC. Belling et al. found that 3 months of VK1 supplementation decreased the development of newly calcifying lesions in the aorta and the coronary arteries, as detected by using ¹⁸F-NaF PET (12). Similarly, the study by Shea et al. put forward that VK supplementation delayed the course of pre-existing CAC in healthy women and elderly men (13). However, eight other trials that measured major arterial or valve calcification reported no appreciable benefits of VK intake. The role of VK in CAC seems to remain controversial.

Coronary artery calcification (CAC) means the existence of coronary artery disease that is accompanied by the progression of advanced atherosclerosis and is a well-established predictive indicator of future CVD events (35). It is easily identified by radiography as well as by CT and intravascular imaging (35, 36). At the same time, the CT scan-based Agatston score can serve as a quantitative measure of CAC (37). Their application has been expanded to examine other blood vessels. However, the Agatston score fails to capture information about the regional distribution of calcified plaque, and large variability tends to occur in baseline differences of CAC scores, which contribute to the negative results (38, 39). As a non-invasive and quantitative imaging technique, ¹⁸F-NaF PET provides a biomarker of calcification activity and detects new calcium generation outside the resolution of CT (40). In Zwakenberg et al.'s (28) study, ¹⁸F-NaF activity tended to rise in the MK-7 group compared with the placebo group, which is contrasting with what we found. The discrepancy in the baseline calcification mass and different calcification areas detected by ¹⁸F-NaF uptake may have played a role in the disappointing trial outcome (41). In addition, a related study evaluated the uptake of ¹⁸F-NaF PET in relation to gender and the number of cardiovascular risk factors (42). In the recent presentation, invasive imaging techniques, such as angiography, with or without optical coherence tomography, provide a higher spatial resolution of mineralized sites inside and along the artery walls (43). Although these are more sensitive for the identification of VC, their applicability is restricted as an invasive procedure. Overall, limitations in image resolution and other factors may bias the evaluation of calcification scores when using traditional imaging modalities such as CT scans. However, despite these limitations, we have found conclusive evidence that VK intake may impede VC, particularly CAC. Another way to assess the status of VK is by measuring dp-ucMGP levels in the blood, which has been linked to surrogated biomarkers of VC, vascular stiffness, and cardiovascular outcomes (44).

Matrix Gla proteins (MGP) are one of the most effective naturally occurring inhibitors of VC and require VK as a cofactor to become bioactive through post-translational γ-carboxylation and phosphorylation (45). The functional status of VK related to specific tissues can be reflected indirectly by measuring un-carboxylation or carboxylation MGP in vivo. However, other types of MGP (except for dp-ucMGP) have a high affinity for precipitated calcium salts and hydroxyapatite, preventing them from freely entering the bloodstream (46). In addition, determining VK content in the plasma is challenging due to the low circulating level of VK, the non-polar nature of VK, and the interference from lipids (5). As a result, measuring dp-ucMGP levels in the blood could be a useful alternative way to assess the status of VK, which is associated with a surrogate biomarker of vascular health (44).

A total of seven trials (578 participants) were compared, and there was a considerable impact on the decline in dp-ucMGP after VK supplementation. Holden et al. (39) conducted parallel RCTs in hemodialysis patients in 2022, which indicated that dp-ucMGP levels declined by 86% in the VK1 group. Levy-Schousboe et al.'s study showed that despite continuing VK supplementation within hemodialysis patients, dp-ucMGP decreased from the baseline in year 1 and increased again in year 2. It seems to demonstrate that VK supplementation significantly decreases dp-ucMGP levels but cannot prevent a subsequent increase (21). Moreover, another study of hemodialysis patients revealed that lengthy VK2 intake at therapeutic doses and VK antagonism withdrawals do not normalize systemic dp-ucMGP levels, but rather just reduce them (27). The results of the three studies are highly consistent, but they also raise questions worth considering. Specifically, the findings prompt us to inquire why VK supplementation does not restore dp-ucMGP to normal levels in hemodialysis patients.

There are two reasons to explain why dp-ucMGP levels cannot be normalized in hemodialysis patients after VK supplementation. On the one hand, multiple studies have confirmed that dp-ucMGP levels are significantly elevated in hemodialysis patients, indicating VK deficiency. The causes of VK insufficiency in hemodialysis are multifaceted and include inadequate intake, uremic suppression of the VK cycle, and possibly interfering with VK absorption by phosphate binders. On the other hand, dp-ucMGP concentrations are decided by VK status and total MGP levels which have been shown to increase with age and in CVD (47, 48). Furthermore, changes in dp-ucMGP levels were not only seen in hemodialysis patients but also in other patients. Fulton et al., Knapen et al., and Kurnatowska et al. demonstrated that dp-ucMGP levels fell significantly after VK supplementation in other patients. The result is consistent with Zwakenberg et al.'s study. Numerous studies focused on dp-ucMGP alterations in reaction to VK therapy, and it has not been mentioned whether decreasing the level of dp-ucMGP after VK supplementation could well inhibit or alter the advancement of VC. However, we have confirmed that VK supplementation reduces the absolute level of dp-ucMGP.