A systematic search of PubMed, EMBASE, the Cochrane Central Register of Controlled Trials, and Web of Science databases was conducted to identify relevant studies on 10 December 2020 without date restrictions. Key search terms were (“vitamin D”) AND (“heart surgery” OR “cardiac surgery”). We searched the PubMed database using [vitamin D [MeSH Terms]) OR (vitamin D)] AND [(heart surgery [MeSH Terms]) OR (cardiac surgery) OR (heart surgery) OR (heart operation) OR (cardiosurgery)].

Two independent investigators undertook thorough literature searches, with discrepancies resolved by a third investigator in a blinded fashion. The retrieved results were de-duplicated and screened against the pre-specified eligibility criteria.

Two reviewers screened the titles and abstracts of studies independently. Then, they assessed the full text of the selected studies in detail for eligibility. We excluded studies if the duplicated data for all outcomes of interest were published elsewhere, and preserved the studies that provided comparative data if there was an overlap of data between studies.

We developed inclusion criteria based on the Population, Index prognostic factor, Comparator prognostic factor, Outcome, Timing, Settings (PICOTS) framework adapted from the guideline proposed by Riley et al. (15). We defined patients as having confirmed VitD deficiency if they had a VitD concentration in serum <20 ng/mL or <50 nnmol/L irrespective of clinical signs and symptoms (17). Eligible studies had to meet the following criteria: (1) population: patients having undergone cardiac surgery and VitD test without supplementation with VitD after the surgical procedure. (2) Index prognostic factor: VitD deficiency was defined as VitD concentration in serum <50 nnmol/L irrespective of clinical signs and symptoms (17). (3) outcome: major adverse cardiovascular events (the composite of myocardial infarction, stroke, or cardiovascular death), SYNTAX score >22, maximum VIS >20, and duration of stay in the ICU. (4) time: preoperative and postoperative VitD level should be measured within a week before and after surgery. (5) Setting: in-hospital.

Inter-rater reliability for the study selection was calculated using the kappa statistic. Studies meeting any of the following criteria were excluded: (1) patients with heart disease were not related the surgery (2) Time of VitD measurement were not reported (3) The outcomes didn't meet with our definition (4) conference articles, reviews, abstracts, other non-peer-reviewed literatures or those not based on original studies.

Two investigators (Yulin Zhang and Jiawen Li) assessed the eligibility of reports independently at the title and abstract level. A third reviewer (Yifei Li) determined divergence according to the inclusion or exclusion criteria and quality of the reports. Studies that met all of the inclusion criteria were selected for further analyses. The baseline data from the included studies were extracted and are shown in Table 1. The quality of the included studies in the meta-analysis was assessed using the Newcastle–Ottawa Scale (NOS). In addition, studies which scored ≥5 stars were considered to have moderate-to-high methodological quality.

First, we collected the VitD level of patients before, immediately after, and 24-h after undergoing cardiac surgery from correlational studies. Then, we compared the outcomes between the VitD-deficient group and VitD-sufficient group.

A publication bias was tested using Egger's regression and funnel graph by STATA 15.1 (State, College Station, TX, USA). Each dot represents a study in the meta-analysis and asymmetry of distribution of dots indicates a potential publication bias. A quantified result of P < 0.05 in Egger's test indicated that a publication bias might be present.

Heterogeneity in the pooling sensitivity and specificity was examined using the Q-test and was deemed to be significant if P < 0.10 in these qualitative tests. The I² test was also carried out in each pooling analysis to estimate quantitatively the proportion of total variation across studies that were due to heterogeneity rather than chance. I² can range from 0 to 100%, and I² >50% suggests significant heterogeneity.

Sensitivity analyses were conducted (using STATA 15.1) for each study to determine if a single study incurred undue weight in the meta-analysis fixed/random-effects estimates.

Analyses were undertaken for adjusted and unadjusted estimates. Adjusted estimates were utilized primarily for reporting and interpretation of results (38). The pooled effects of dichotomous outcomes were converted to risk ratios (RRs) along with their 95% confidence intervals (CIs). Quantitative synthesis was first conducted by comparing VitD-deficient patients with VitD-sufficient patients or the highest vs. lowest categories of VitD using the generic inverse variance method with the DerSimonian–Laird random-effects model (15). Continuous outcome variables are expressed as the mean ± SD. All pooled effects are presented as forest maps. Quality of evidence was assessed by the modified Grading of Recommendations Assessment, Development, and Evaluation system (GRADE) by consensus among the authors (39, 40). If pooled effect sizes with great heterogeneity comprised >5 studies, subgroup analyses were carried out based on the study design, study location, sample size, risk of a bias, type of effect size, and surgical method. Conversely, sensitivity analyses were conducted by leave-one-out analysis and the exclusion of studies with a high risk of a bias. GRADE was used to evaluate the overall quality of the evidence for each outcome, which ranged from high quality to very low quality, and was based on five domains: limitations of design; inconsistency of results; indirectness; imprecision; other factors (e.g., a publication bias) (Table 2).

Initially, 2,063 potentially relevant articles were retrieved by our search method. Of these, 31 articles were considered to be of interest after reading the title and abstract. However, 15 articles were excluded by reading the complete articles due to: article type (n = 1); absence of measurement of VitD levels at required time points or the outcomes described in the criteria of inclusion (n = 13); not using the standard definition of VitD deficiency (n = 1). Ultimately, 16 reports (1, 2, 18–31) were included in the meta-analysis (Figure 1). No reports from China met the inclusion criteria.

The characteristics of the involved patients in all included studies are presented in Table 1. The sixteen published reports (1, 2, 18–31) enrolled a total of 1,785 patients (children and adults) from 10 countries. Five studies (2, 18, 19, 25, 31) focused on children with congenital heart disease and 11 studies (1, 20–24, 26–30) were on adults, of which seven studies (1, 22–24, 26–28) reported patients undergoing coronary artery bypass surgery (CABG). Eleven studies (1, 2, 18, 19, 21, 23–26, 29, 31) reported a detailed method used to measure the serum concentration of VitD and 10 studies (1, 2, 18–20, 25, 26, 28, 30, 31) reported a clear definition of VitD deficiency as 25-(OH) VitD <20 ng/ml or <50 nmol/L. About half of the included studies yielded a high risk of a bias and the other half were at low risk according to the NOS (Table 2).

Table 3 presents the pooled effect sizes and 95%CIs for the synthesis analysis, and GRADE level of certainty.

Six studies (18, 19, 25, 29–31) provided data on the VitD level preoperatively, immediately after surgery, and 24 h after surgery, and one study (21) provided data only on the VitD level preoperatively and 24 h after surgery. Furthermore, four studies (18, 19, 25, 31) reported the VitD level in children with congenital heart disease and three studies (21, 29, 30) were in adults. However, two studies (29, 30) did not mention the way to measure the serum 25-(OH)-VitD concentration. Three studies in adults (21, 29, 30) reported on patients who underwent types of cardiac surgery other than CABG.

We demonstrated that the VitD level decreased significantly immediately after cardiac surgery [stand mean difference (SMD), 0.69; 95%CI (0.1, 1.28); P = 0.000] (Figure 2) or 24 h after cardiac surgery [SMD, 0.84; 95%CI (0.47, 1.21); P = 0.000] (Figure 3), though all models yielded substantial heterogeneity (I² = 92.2% for immediately after surgery; I² = 85.6% for 24 h after surgery). Subgroup analyses based on enrolled patients (adults or children) were partially consistent with the overall effects. Immediately after surgery, adult patients demonstrated a significant reduction in the 25-(OH)-VitD level [SMD 0.96, 95%CI (0.64, 1.28), I² = 17.7%], whereas child patients had a normal level [SMD 0.51, 95%CI (−0.36, 1.39), I² = 94.5%]. Among data from adults, there was no heterogeneity, and the pooled heterogeneity was considered to originate from child-based studies. At 24-h after cardiac surgery, results from children and adults remained consistent with an overall significant reduction. However, heterogeneities were also demonstrated upon analyses of both subgroups. Egger's test presented an absence of a publication bias with P = 0.713 and P = 0.849 for measurements immediately after and 24-h after cardiac surgery, respectively (Supplementary Figures 1A,B).

Moreover, three studies reported adjusted RRs of VitD deficiency before and 24 h after cardiac surgery. Pooled data demonstrated that cardiac surgery was negatively associated with the VitD level (RR 0.59; 95%CI (0.47, 0.73) (Figure 4) without significant heterogeneity (I² = 33.5%). Egger's test presented an absence of a publication bias with P = 0.235 (Supplementary Figure 1C).

Eight studies (1, 21–24, 26–28) in our meta-analysis reported postoperative severe events and VitD level. Severe events included postoperative atrial fibrillation (Po-AF) and high SYNTAX score. The latter is an angiographic grading tool to evaluate the complexity and extensity of coronary artery disease (CAD). All lesions causing ≥50% of stenosis in a coronary artery of diameter ≥1.5 mm were included in calculation of the SYNTAX score, and the latter was divided into two groups: high (≥23) and low (<23) (1). AF was confirmed by 12-lead electrocardiography. Three studies (23, 27, 28) described AF as irregular, fast oscillations or fibrillary waves instead of regular P waves at electrocardiography and a range of ventricular rates between 90 and 170 bpm. An AF episode longer than 5 min or which necessitated therapy for hemodynamic instability was accepted as Po-AF. However, two studies (21, 26) did not report the criterion of AF, and one study (24) did not report the standard of Po-AF clearly. Another study (22) defined neither AF nor Po-AF precisely. All these definitions were considered to represent AF or Po-AF in the analyses. In total, the VitD level was decreased significantly after surgery in the severe group [SMD, −0.8; 95%CI (−1.41, −0.19); P = 0.01] (Figure 5) with significant heterogeneity (I² = 94.7%). Egger's test revealed an absence of a publication bias with P = 0.737 (Supplementary Figure 1D).

Three studies (2, 19, 31) provided data on the maximum VIS and VitD level. The VIS was calculated by various equations in different studies. One study (31) calculated it using the following equation with the drug dose in μg/kg/min: (dopamine + dobutamine) + (milrinone × 10) + (epinephrine × 100) + (norepinephrine × 100). One study (19) used the following equation with the drug dose in ICU: dopamine (^*1) + dobutamine (×1) + amrinone (×1) + milrinone (×15) + epinephrine (×100) + norepinephrine (×100). Another study (2) did not report the specific calculation of VIS, and none of the studies reported the definition of high VIS and low VIS. Given this information and the limited number of related studies, we regarded these definitions as being equivalent. Overall, children undergoing cardiac surgery with a lower postoperative VitD level had a significantly higher VIS [SMD, −3.71; 95%CI (−6.32, −1.10); P = 0.005, I² = 97.1%] (Figure 6) than that of patients with normal VitD level. Egger's test revealed an absence of a publication bias with P = 0.368 (Supplementary Figure 1E).

Four studies (2, 19, 20, 31) provided data on the duration of ICU stay, and classified patients into two groups: VitD- sufficient and VitD-deficient. All studies defined VitD deficiency using this standard. In total, there was no significant difference in the duration of ICU stay [SMD, −0.53; 95%CI (−1.6, 0.09); P = 0.096] between the two groups (Figure 7). Egger's test revealed an absence of a publication bias with P = 0.160 (Supplementary Figure 1F).

We systematically and qualitatively analyzed the sensitivity across the included studies to determine the influence of individual studies on the results (Figure 8). We did not detect a significant impact from a single study. We confirmed the direction of the results except for one study (Ripple et al.) in analyses of the relationship between the duration of ICU stay and serum level of VitD (Figure 7). After excluding that study, the result was opposite and identified a significant difference in the duration of ICU stay [SMD, −0.76; 95%CI (−1.52, −0.012); P = 0.046] between groups.