A comprehensive search of literature was performed from December 31, 2020, to May 31, 2021, by using seven databases including PubMed, Medline, Embase, Cochrane Library, Google Scholar, Foreign Medical Journal Service, and Excerpta Medica. The following keywords were used in the search: (“microRNA”, “miRNA”, “microRNA-146a”, “miRNA-146a”, “miR-146a”, “rs2910164”, “microRNA-499a”, “miRNA-499a”, “miR-499a”, or “rs3746444”), (“polymorphism”, “mutation”, “variation”, “mutant”, “variant”, “SNP”, or “single nucleotide polymorphism”), and (“lipids”, “circulating lipids”, “blood lipids”, “plasma lipids”, “serum lipids”, “triglycerides”, “total cholesterol”, “low-density lipoprotein cholesterol”, “high-density lipoprotein cholesterol”, “TG”, “TC”, “LDL-C”, or “HDL-C”). Additionally, the reference lists of all eligible studies were manually retrieved to obtain more literature.

The specific inclusion criteria were listed as follows. (1) The articles investigated the effects of miRNA-146a rs2910164 (G > C) polymorphism or miRNA-499a rs3746444 (A > G) polymorphism on lipid levels. (2) The articles at least provided one parameter in the lipid profile (TG, TC, LDL-C, and HDL-C). (3) The articles provided the genotype frequencies of rs2910164 and rs3746444 polymorphisms. (4) The articles offered the mean lipid levels with standard deviation (SD) or standard errors (SE) by genotypes. (5) The interventional articles provided pre-intervention data. (6) The language of eligible articles was restricted to English or Chinese.

Two authors (SW and FL) extracted the data independently by using a standardized data extraction table. The discrepancy in data extracted was resolved by consensus or a discussion with the third author (ZL). If key data were absent, e-mail or telephone was used to contact the corresponding author to acquire these information.

One study conducted by Qiu et al. (2020) did not offer usable data in its original article; fortunately, they offered raw lipid data by the genotypes of rs2910164 and rs3746444 polymorphisms in its Supplementary Material. Therefore, we analysis and obtained these important data by performing SPSS software (version 23.0, Inc, Chicago, IL, USA).

The following data were extracted from each eligible article: the last name of the first author, year, country, gender, study population, ethnicity, genotype counts, genotyping methods, type of study, type of disease, total sample size, and mean lipid levels with SD or SE by genotypes.

The units of TG, TC, LDL-C, and HDL-C were converted into mmol/L. All extracted data were expressed as mean ± SD. The Hardy–Weinberg equilibrium (HWE) of the populations was tested by χ² test. Since most of the included studies presented data in a dominant model [(GC + CC) vs. GG for rs2910164; (AG + GG) vs. AA for rs3746444], a dominant model was adopted to ensure adequate statistical power. All the analyses were performed by STATA software (version 15.0, College Station, TX, USA). p < 0.05 was recognized as statistically significant. The standardized mean difference (SMD) and 95% confidence interval (CI) were used to evaluate the differences in lipid levels between the genotypes. The raw data downloaded from Qiu et al. (2020) Supplementary Material were analyzed using SPSS software (version 23.0, Inc, Chicago, IL, USA). If data follow normal distribution, ANOVA was adopted to calculate the results.

Inevitably, there were differences between the included articles in a meta-analysis. The differences among participants, interventions, or inner authenticity variations in those articles were defined as “heterogeneity” (Higgins and Green, 2006). Heterogeneity was tested by I² statistic and Cochran's χ2-based Q statistic. Galbraith plots were used to detect the potential sources of heterogeneity. If heterogeneity was significant (I² > 50%, p ≤ 0.05), the random-effect model (DerSimonian–Laird method) was used to calculate the results (DerSimonian and Kacker, 2007). Otherwise, the fixed-effect model (Mantel–Haenszel method) would be adopted (I² <50%, p > 0.05).

The risk of bias among the included studies was evaluated by the risk-of-bias plot (Savović et al., 2014), in which different colors represent different levels of risk of bias, e.g., green represents low risk of bias, while yellow refers to unclear risk of bias; however, red represents high risk of bias.

The publication bias among the included studies was evaluated by Begg's funnel plot and Egger's linear regression test (Begg and Mazumdar, 1994). The funnel plots were asymmetric when there were publication biases and symmetric in case of no publication bias.

Subgroup analysis was carried out by ethnicity, female subjects, and cardiovascular disease (CVD) patients due to the limited number of studies. The ethnicity was divided into Caucasian and Asian. The subjects with CAD, ischemic stroke (IS), or hypertension were considered as CVD patients. In some studies, the subjects were divided into more than one subpopulation (e.g., the subjects with different types of disease, the subjects originated from different races, case and control subjects). Each subpopulation was regarded as an independent comparison in this study.

Sensitivity analysis was conducted in this meta-analysis, in which the comparison was excluded one by one, and the analysis was performed again after omitting each comparison. If the synthetic results in any of the comparison changed substantially to alter the results from significant to non-significant or the other way around. The absence of such a phenomenon usually indicates the robustness and stableness of synthetic results.

By searching relevant databases, 4,106 articles were identified; after the screening, 3,811 articles were excluded by its title and abstract. Next, 123 articles were further estimated by its contents, in which 110 articles provided data of other polymorphisms, one article conducted by Mir et al. (2020) did not provide the specific lipid levels by the genotypes of rs2910164, one article conducted by Huang et al. (2015) provided lipid levels by the genotypes of rs2910164 but expressed odds ratio (OR) and 95% CI, and 1 article conducted by Ibrahim et al. (2020) provided lipid levels by the genotypes of rs2910164 but in a strange genetic model [(CG + GG) vs. CC]. Therefore, 113 articles were further excluded. Despite that we carefully and repeatedly reviewed all the relevant literatures and tried our best to obtain more studies, only 10 studies (4,087 subjects) were included due to the limited number of eligible studies, in which, six studies (3,470 subjects) and six studies (3,227 subjects) were respectively included for lipid association analysis in rs2910164 and rs3746444 (Figure 1).

The references of the included studies are listed in Supplementary Material. The characteristics of the studies included in the meta-analysis are presented in Supplementary Table S1. The plasma lipid levels by the genotypes of miRNA-146a rs2910164 polymorphism are presented in Supplementary Table S2. The plasma lipid levels by the genotypes of miRNA-499a rs3746444 polymorphism are presented in Supplementary Table S3. The forest plot of the meta-analysis between miRNA-499a rs3746444 polymorphism and plasma lipid levels is presented in Supplementary Figures S1–S4. The sensitivity analysis between miRNA-499a rs3746444 polymorphism and plasma lipid levels is presented in Supplementary Figures S5–S7, S11. The sensitivity analysis between miRNA-146a rs2910164 polymorphism and plasma lipid levels is presented in Supplementary Figures S8–S10, S12. The risk-of-bias plot of the meta-analysis between miRNA-499a rs3746444 polymorphism and plasma lipid levels is presented in Supplementary Figure S13. The Begg's funnel plot of the association analysis between miRNA-146a rs2910164 polymorphism and plasma TC, LDL-C, and HDL-C levels is presented in Supplementary Figures S14–S16, respectively. The Begg's funnel plot of the association analysis between miRNA-499a rs3746444 polymorphism and plasma lipid levels is presented in Supplementary Figures S17–S20.

The rs2910164 polymorphism largely increased the TG (SMD = 0.35, 95% CI = 0.15–0.54, p < 0.001), TC (SMD = 0.43, 95% CI = 0.16–0.70, p < 0.001), and LDL-C (SMD = 0.37, 95% CI = 0.11–0.63, p = 0.01) levels and decreased the HDL-C level (SMD = –0.27, 95% CI = −0.47−0.07, p = 0.01) (Table 1; Figures 2–5). When the analysis was limited to the studies in HWE, the significant associations of the rs2910164 polymorphism with TG, TC, LDL-C, and HDL-C levels were also detected (Table 1).

Then, the subgroup analysis by the characteristics of the subjects (Table 1) showed that the significant associations of the rs2910164 polymorphism with TG, TC, LDL-C, and HDL-C levels were only observed in Caucasians and females (Table 1). In addition, a marginal significance was observed in patients with CVD in the association analysis between the rs2910164 polymorphism and TG levels (Table 1).

The analysis that excluded the studies with heterogeneity was also carried out (Table 2), and the analysis results showed that a significant association of the rs2910164 polymorphism with TG levels was observed in Caucasians, Asians, and CVD patients, while a significant association of the rs2910164 polymorphism with TC levels was only observed in Asians. Moreover, a significant association of the rs2910164 polymorphism with HDL-C levels was observed in Caucasians and Asians. However, only a marginal significance was observed in the analysis of the correlation between the rs2910164 polymorphism and LDL-C levels.

No statistical significance was observed in the analysis of associations of the rs3746444 polymorphism with TG (Supplementary Figure S1), TC (Supplementary Figure S2), LDL-C (Supplementary Figure S3), and HDL-C (Supplementary Figure S4) levels (Table 3). However, after excluding the studies with heterogeneity, the correlation between the rs3746444 polymorphism and TG levels showed statistical significance in Asians and CVD patients (see Table 4 for more details).

In the analysis of correlations between the rs2910164 polymorphism and lipid levels, significant heterogeneity was detected in TG, TC, LDL-C, and HDL-C (Table 1). Two (Abo-Elmatty and Mehanna, 2019; Qiu et al., 2020), three (Mehanna et al., 2015; Chen et al., 2018; Abo-Elmatty and Mehanna, 2019), three (Mehanna et al., 2015; Chen et al., 2018; Abo-Elmatty and Mehanna, 2019), and three (Li et al., 2015a; Abo-Elmatty and Mehanna, 2019; Qiu et al., 2020) comparisons were recognized as the main contributors to TG, TC, LDL-C, and HDL-C heterogeneity, respectively. SMD values and 95% CIs of TG, TC, and HDL-C did not change substantially after excluding these comparisons (Table 2). However, the SMD value and 95% CI of LDL-C (SMD = 0.08, 95% CI = −0.01–0.17, p = 0.08) changed significantly after excluding these outlier comparisons.

In the correlation analysis between the rs3746444 polymorphism and lipid levels, significant heterogeneity was detected in TG and HDL-C (Table 3). One comparison (Qiu et al., 2020) and one comparison (Chen et al., 2017) were recognized as the main contributors to TG and HDL-C heterogeneity, respectively. The SMD value and 95% CI of HDL-C did not change substantially after excluding this comparison (Chen et al., 2017). However, the SMD value and 95% CI of TG (SMD = −0.09, 95% CI = −0.17−0.01, p = 0.03) changed significantly after excluding this comparison (Qiu et al., 2020).

Sensitivity analysis showed that no comparison may affect the associations of the rs3746444 polymorphism with the TC (Supplementary Figure S5), LDL-C (Supplementary Figure S6), and HDL-C (Supplementary Figure S7) levels. However, one comparison (Li et al., 2015a) may affect the significant associations of the rs2910164 polymorphism with the TC (Supplementary Figure S8), LDL-C (Supplementary Figure S9), and HDL-C (Supplementary Figure S10) levels as well as the significant correlation between the rs3746444 polymorphism and TG (Supplementary Figure S11) levels, while another comparison (Qiu et al., 2020) may affect the correlation between the rs2910164 polymorphism and TG (Supplementary Figure S12) levels.

The correlations between the rs2910164 and rs3746444 polymorphisms and lipid levels did not change substantially [rs2910164 and TG levels: (SMD = 0.41, 95% CI = 0.21–0.61, p <0.001); rs2910164 and TC levels: (SMD = 0.50, 95% CI = 0.17–0.84, p < 0.01); rs2910164 and LDL-C levels: (SMD = 0.45, 95% CI = 0.13–0.77, p = 0.01); rs2910164 on HDL-C levels: (SMD = −0.33, 95% CI = −0.55−0.10, p < 0.01); rs3746444 and TG levels: (SMD = 0.03, 95% CI = −0.07–0.13, p = 0.60)] after omitting above comparisons (Li et al., 2015a; Qiu et al., 2020). It indicated that the present analysis results were robust and stable.

In the association analysis between the rs3746444 polymorphism and lipid levels, some concerns were observed in the randomization process (33.3%) and measurement of the outcome (16.7%); however, the overall results showed a low risk of bias (66.7%) among the included studies (Supplementary Figure S13). It indicated that the studies included in the meta-analysis were of relatively high quality. In addition, there was no risk of bias in the analysis of association of the rs3746444 polymorphism with lipid levels in addition to some concerns (33.3%) in the randomization process (Figure 6).

In the present study, all the funnel plots were symmetric visually (Figure 7; Supplementary Figures S14–S20). Begg's test did not find any publication bias in the analysis for the rs2910164 and rs3746444 polymorphisms, which was confirmed by Egger's regression test [p = 0.22 for rs2910164 and TG (Figure 7), p = 0.06 for rs2910164 and TC (Supplementary Figure S14), p = 0.06 for rs2910164 and LDL-C (Supplementary Figure S15), p = 0.07 for rs2910164 and HDL-C (Supplementary Figure S16), p = 0.34 for rs3746444 and TG (Supplementary Figure S17), p = 0.27 for rs3746444 and TC (Supplementary Figure S18), p = 0.10 for rs3746444 and LDL-C (Supplementary Figure S19), p = 0.18 for rs3746444 and HDL-C (Supplementary Figure S20)].