Characteristics of eligible studies and samples
A total of 80 studies met the inclusion criteria (see Fig. 1 for PRISMA flowchart). As described in the online supplementary materials (Additional file 4), 116 samples and 56,161 subjects were involved. Approximately 44.27% were women (24,860/56,161), and less than half were Asians (for Asian: 42.41%, samples = 75, n = 23,820 [34–80]; for Caucasian: 53.36%, samples = 37, n = 29,967 [9, 16, 18, 81–101]; and one African [102] and two mixed populations [103, 104] were defined as the mixed group: 4.23%, samples = 4, n = 2,374). The samples could also be categorised into patients (23.96%, samples = 45, n = 13,456), random (37.14%, samples = 44, n = 21,861) and mixed (38.9%, samples = 27, n = 21,844) groups according to their clinical information. More than 12 kinds of diseases were included in collected studies’ samples.
Relationship between R219K and HDLC level
Overall pooled effect of R219K on HDLC
Pooled data revealed a significantly lower HDLC level in the RR genotype group than that in the K allele carrier group (SMD = -0.28 mmol/L, 95%CI: -0.36 ~ -0.20, z = -6.85, P < 0.001) in a random model. Extreme heterogeneity among all studies was observed, I2 = 94.62% (Q = 2138.42, df = 115, P < 0.001) among these eligible studies. Four of the included samples (Abellán [93] data1, Sun [45] data 2, Ya [105] data1 and data 2) were identified as outliers with low-quality data and extreme effect size (see online supplementary materials, Fig. S5-1 and Fig. S5-2). The trend of the RR genotype population having a significantly lower HDLC level (SMD = -0.17 mmol/L, 95%CI: -0.22 ~ -0.12, z =-6.39, P < 0.001, and I2 = 85.43%, Q = 761.80, df = 111, P < 0.001) than that of the K allele carriers in a random model was still observed after removing the four outliers.
Effect of R219K on HDLC in Asian and Caucasian subgroups
For Asian populations (samples = 72, n = 22,959), the RR genotype group had significantly lower HDLC level than the K allele carriers (SMD = -0.25 mmol/L, 95%CI: -0.32 ~ -0.17, z = 6.60, P < 0.001), and I2 = 82.34% (Q = 401.98, df = 71, P < 0.001) in the heterogeneity test (Fig. 2). For Caucasian populations (samples = 36, n = 29,358), the difference in HDLC level between the RR genotype and K allele carriers did not reach statistical significance, SMD = -0.04 mmol/L, 95%CI: -0.11 ~ 0.02, z = -1.27, P = 0.20 in the random model, and I 2 = 85.00% (Q = 233.33, df = 35, P < 0.001) in the heterogeneity test. The relationship between different populations (e.g., Asian and Caucasian) was also significantly different (Q = 5.20, df = 1, P = 0.02), as estimated by the metafor package.
Effect of R219K on HDLC in different health-condition subgroups
Figure.3 shows that for clinical patients (samples = 43, n = 13,235), individuals with RR genotype had significantly lower HDLC level than the K allele carriers (SMD = -0.16 mmol/L, 95%CI: -0.23 ~ -0.09, z = -4.32, P < 0.001) in the random model, and I2 = 68.78% (Q = 131.32, df = 41, P < 0.001) was obtained in the heterogeneity test. For the random population (samples = 43, n = 19,812), the RR genotype carriers also had significantly lower HDLC level than the K allele carriers (SMD = -0.15 mmol/L, 95%CI: -0.25 ~ -0.05, z = -2.89, P = 0.004), and I2 = 89.30% (Q = 392.65, df = 42, P < 0.001) was acquired in the heterogeneity test. The same trend was observed in the mixed population (samples = 27, n = 21,844, SMD = -0.20 mmol/L, 95%CI: -0.29 ~ -0.11, z = -4.28, P < 0.001) with extreme heterogeneity of I2 = 88.77% (Q = 231.59, df = 26, P < 0.001). Additionally, no any significant difference of the estimated effect of R219K on HDLC level was observed among the three subgroups.
Heterogeneity sources
Meta-regression analysis was also performed to explore all potential moderators including categorical (e.g., ethnicity and healthcondition) and continuous (publication data, gender, and age) variables. The racial factor which is responsible for more than 13% of the sample variance (for Asian population, β = -0.32, 95% CI: -0.72 to 0.08, R2 = 0.15; for Caucasian, β = -0.12, 95% CI: -0.52 to 0.28, R2 = 0.13) was significantly associated with the effect of R219K (Q = 17.00, df = 2, P < 0.001. Table 1). Meanwhile, the other four variables including publication date, mean sample age, percentage of females in each sample, and clinical status were not associated with the pooled effect (all Ps > 0.05).
Table 1
The meta-regressions of moderators for the estimated effect of R219K
Moderator | No. of studies | β | 95%CI | z | P | R-square |
Intercept | | 15.68 | -7.54 to 38.90 | 1.32 | 0.186 | 0.00 |
Publication data | 108 | -0.01 | -0.02 to 0.00 | -1.31 | 0.189 | 0.00 |
Racea | | | | | | |
Asian | 72 | -0.32 | -0.72 to 0.08 | -1.56 | 0.119 | 0.15 |
Caucasian | 36 | -0.12 | -0.52 to 0.28 | -0.61 | 0.543 | 0.13 |
| | Q = 17.00, df = 2, P < 0.001 |
Health condition b | | | | | | |
Patients | 42 | 0.05 | -0.08 to 0.18 | 0.78 | 0.435 | 0.10 |
Random | 43 | -0.03 | -0.18 to 0.12 | -0.40 | 0.688 | 0.08 |
| | Q = 1.09, df = 2, P = 0.581 |
Mean sample age | 100 | 0.10 | -0.10 to 0.29 | 0.96 | 0.339 | 0.06 |
Percentage of females | 101 | 0.00 | -0.01 to 0.00 | -0.61 | 0.545 | 0.06 |
a Samples which consisted multiple ethnic subjects set as reference; |
b Samples including patients and controls set as reference; |
c Bold type denotes P < 0.05. |
Publication bias analysis
A publication bias analysis was performed and is available in the online supplementary materials (Additional file 5). Both Begg’s rank correlation (tau = -0.24, z = 3.83, P < 0.001) and Egger’s weighted regression (t110 = 3.82, P < 0.001) detected a significant publication selection bias in this meta-analysis. The funnel plot (Fig. S5-3) also showed a considerable asymmetry distribution among the included studies. Meanwhile, the trim-and-fill test estimated approximately 23 missing studies on the left side of the mean effect, and the overall initial effect was significantly changed (SMDadj = -0.26 mmol/L, 95%CI: -0.31 ~ -0.20; t245 = 2.34, P = 0.02) after the adjustment for the missing data.
Meta-analysis for genetic variant R219K and LDLC
Relationship between R219K and LDLC levels
The pooled effect of the genetic variant R219K on LDLC levels was estimated with 65 eligible studies (samples = 94) including 34,901 participants. The meta-analysis showed that the RR genotype population had significantly higher LDLC level than K allele carriers (SMD = 0.12 mmol/L, 95%CI: 0.04 ~ 0.20, z = 2.83, P = 0.005) in the random model. Extreme heterogeneity among all studies was observed with I2 = 91.56% (Q = 1101.80, df = 93,P < 0.001). The same correlation trend (SMD = 0.05 mmol/L, 95%CI: 0.01 ~ 0.10, z = 2.25, P = 0.02, and I2 = 72.08%,Q = 329.46, df = 92, P < 0.001 for heterogeneity test) between R219K and LDLC levels was observed after removing the outliers of Abellán’s sample [93] identified by the metafor software package.
In Asian populations (samples = 66, n = 17,180), the LDLC level was clearly higher in the RR genotype group than in the K allele carrier group (SMD = 0.06 mmol/L, 95%CI:-0.01 ~ 0.12, and I2 = 69.45%, Q = 212.76, df = 65, P < 0.001 for the heterogeneity test. Fig. S6-2), but the difference did not reach statistical significance (z = 1.81, P = 0.07). For Caucasian populations (samples = 25, n = 16,015), the difference also disappeared (SMD = 0.06 mmol/L, 95%CI: -0.02 ~ 0.13, z = 1.43, P = 0.15, and I2 = 76.37%, Q = 101.55, df = 24, P < 0.001 for heterogeneity test; Fig. S6-3). No significant difference was found in the comparison of the effects of R219K estimated in Asian and Caucasian populations. The same result trend was observed when analysing other subgroups categorised by subjects’ health condition (Fig. S6-4). Furthermore, the meta-regression analysis indicated that no moderators were significantly associated with the relationship between ABCA1 R219K polymorphism and individual LDLC levels (Table S6-1).
Publication bias analysis
For the publication bias analysis, Begg’s rank correlation suggested no significant publication selection bias (z = 1.12, P = 0.26) and Egger’s weighted regression showed a weak bias (t91 = 2.16, P = 0.033). The funnel plot also showed no considerable asymmetry distribution of the effect of R219K of each eligible study, with 10 missing studies estimated by the trim-and-fill method (Fig. S6-5).
Meta-analysis for genetic variant R219K and TC
The effect size of the genetic variant R219K on TC levels was pooled from 66 studies (samples = 96, n = 34,814). The meta-analysis showed no significant difference in the TC level between the RR genotype population and the K allele carriers in the random model, and significant heterogeneity was found among all studies. After four outliers (Abellán [93] data1, Ya [105] data1, 2 and 3) detected by the meta and metafor packages were removed (Fig. S7-1 and 2), a consistent result was obtained (SMD = 0.01 mmol/L, 95%CI:-0.06 ~ 0.08, z = 0.20, P = 0.846; I2 = 88.37%, Q = 782.18, df = 91, P < 0.001. Fig. S7-3). Hierarchical and meta-regression analyses were also performed to explore the heterogeneity among samples, but no variable was identified. Furthermore, no significant publication bias was found among the current selected studies (Table S7-1).
Meta-analysis for genetic variant R219K and TG
Sixty-five eligible studies (samples = 95, n = 34,478) were collected in this study to explore the relationship between R219K polymorphism and individual TG levels. Pooled results showed that the RR genotype population had significantly higher TG level than the K allele carriers (SMD = 0.15 mmol/L, 95%CI: 0.05 ~ 0.25, z = 2.92, P = 0.003) in the random model, and I 2 = 94.54% (Q = 1721.66, df = 94, P < 0.001) for the heterogeneity test. However, this significant effect disappeared when four outlier data (Delgado-Lista [19], Sun [45] data1 and data 2, Ya [105] data1) were removed from all samples (Fig. S8-1 and 2).
Additionally, no significant effect between R219K genotype and TG level was observed in the inter- and intra-population of Asians and Caucasians (Fig. S8-3). Neither the following subgroup analysis (Fig. S8-4) nor meta-analysis (Table S8-1) revealed that the variable was associated with the effect of R219K or responsible for the extreme heterogeneity of the current study.
Begg’s rank correlation and Egger’s weighted regression methods showed no publication bias in all studies. However, looking through the effect distribution of each study in the funnel plot and considering the adjustment for missing studies estimated by trim-and-fill method, the initial effect of R219K on TG level was significantly changed (the estimatedright missing studies n = 21, SMDadj = 0.14, 95%CI: 0.08 ~ 0.21; t = 2.37, P = 0.019).