Abstract
Background Statin-associated muscle symptoms (SAMS) are a major cause of treatment discontinuation. Adjusting statin dosages for solute carrier organic anion transporter family member 1B1 (SLCO1B1) genotype has been proposed to reduce SAMS. We hypothesized that the association between SLCO1B1 genotype and SAMS is misestimated because of publication bias.
Methods We searched for published systematic reviews evaluating the association between SLCO1B1 genotype and SAMS. We collected the odds ratio (OR) of this association in each clinical study. We assessed the presence of publication bias using the visual inspection of a funnel plot and Egger’s test and used the Bayes Factor (BFPublication-bias) of the Robust Bayesian Meta-Analysis (RoBMA) as a sensitivity analysis. We evaluated the effect of publication bias by comparing qualitatively and quantitatively (ratio of OR [ROR]) OR of the meta-analysis i) uncorrected for potential publication bias (ORUncorrected) and ii) corrected using the trim-and-fill (ORTrim&Fill). We also used the RoBMA (ORRoBMA) for corrected OR as a sensitivity analysis. Our primary analysis covered the associations between any SLCO1B1 genotype and any statin drug. Secondary analysis focused on SLCO1B1 genotypes and statin drug subgroups.
Results We included 8 cohort and 11 case-control studies, totaling 62 OR of three SLCO1B1 genotypes and five statin drugs plus one ‘mixed’ statin treatment. All controls were statin-tolerant patients. In the primary analysis, the funnel plot was suggestive of publication bias, confirmed by Egger’s test (p=0.001) and RoBMA (BFPublication-bias=18). Correcting the estimate for publication bias resulted in loss of the association, from a significant ORUncorrected (1.31 95% CI [1.13– 1.53]) to corrected ORs suggesting no difference: i) ORTrim&Fill (1.07 95% CI [0.89–1.30]) and ii) ORRoBMA (1.02 95% CI [1.00–1.33]). The RORTrim&Fill and the RORRoBMA suggested that publication bias overestimated the association by 18% and 23%, respectively. The results were similar for the most studied SLCO1B1 genotype, as for simvastatin and atorvastatin.
Conclusion The effect of the SLCO1B1 genotype on the risk of developing SAMS is overestimated in the published literature. This could lead prescribers to incorrectly decreasing statin doses or even avoiding statin use, leading to a loss of the potential cardiovascular benefit of statins.
Clinical perspective What is new?
There is significant publication bias in the available literature regarding the association between SLCO1B1 genotype and statin-associated muscle symptoms.
The available literature overestimates the importance of the SLCO1B1 genotype on statin-associated muscle symptoms.
What are the clinical implications?
The cardiovascular benefit of statins might be wrongly lost when adjusting statin therapy with the SLCO1B1 genotype.
The effect of publication bias should be considered when writing guidelines.
INTRODUCTION
Statins decrease low-density lipoprotein cholesterol (LDL-C). Several randomized controlled trials (RCTs) have confirmed the cardiovascular (CV) benefits of several statins in different primary and secondary prevention populations [1–3]. The most common reported adverse events (AEs) are statin-associated muscle symptoms (SAMS). SAMS range from myalgia to rhabdomyolysis, with or without elevated creatine kinase (CK); they decrease or stop after treatment discontinuation [4]. The exact mechanism of statin muscle toxicity has yet to be fully clarified [5]. SAMS occurrence is likely to limit patient adherence to statins. The prevalence of muscle symptoms in treated patients has been estimated to be around 10% [range 5%–25%]. More specifically, muscular symptoms only attributable to statins have been estimated around 1-2% [ranging from 0.5% to 4%] [4]. Several genotypes of the solute carrier organic anion transporter family member 1B1 (SLCO1B1) gene have been reported to be associated with an increased risk of SAMS [6–11]. The SLCO1B1 gene encodes for the OATP1B1 protein, which is involved in the hepatic transport of statins. The European Medicines Agency (EMA) and the Food and Drug Administration (FDA) have listed the SLCO1B1 gene mutation as a potential marker for SAMS. The Clinical Pharmacogenetics Implementation Consortium (CPIC) recently updated their guidelines and recommend dose adjustment for statin treatment according to SLCO1B1 genotype “if pharmacogenetic test results are available” to reduce the risk of SAMS [12]. The impact of genetic information on prescription changes according to CPIC guidelines was evaluated in a recent study [13]. In 64% of cases, decision aids were used by prescribers, resulting in: a reduced dose of simvastatin in 74% and change in the statin in 26%. Although CPIC does not endorse genetic testing to determine the SLCO1B1 genotype, their estimates suggest that 11-36% of individuals have an intermediate activity phenotype and 0-6% have a low activity phenotype [12].
However, the supposed association between SAMS and SLCO1B1 genotype might be affected by publication bias [14]. Such publication bias has been documented in drug safety assessment [15], genetic epidemiology [16] and more recently in pharmacogenetics [17]. Regarding the SLCO1B1—SAMS association, the available systematic reviews suggest a potential significant publication bias but failed to take it into account. The meta-analysis of Jiang et al. reported a significant association between a SLCO1B1 genotype and an increased risk of AE for various statins [18]. However, they also reported significant Egger’s and Begg’s tests for one association, but without including this particular risk of bias in their conclusion or in the abstract. Moreover, they acknowledged in the discussion section that their review might lack power to detect publication bias. More recently, Xiang et al. reported a significant increased risk of SAMS with specific SLCO1B1 genotype [19]. However, among the eight Egger’s and Begg’s tests they reported, four were significant (p value <5%). Moreover, they used the trim- and-fill (T&F) correction to take into account the publication bias, and reported a loss of the association for one out the three reported corrected ORs [20]. However, they did not state this publication bias and its impact in the conclusion or in the abstract. The presence of this uncertainty poses a significant concern, considering the fact that guidelines are recommending the use of SLCO1B1 genotype to reduce the incidence of SAMS [12]. Consequently, accurately estimating the association between the SLCO1B1 genotype and SAMS is imperative for making well-informed clinical decisions.
Therefore, we aimed to systematically assess the presence and the effect of publication bias on the association between SLCO1B1 genotype and SAMS.
METHODS
Protocol and registration
The protocol was registered a priori on the Open Science Framework (OSF) website: https://osf.io/36jq8/. Our methodology complied with the Cochrane Handbook for Systematic Reviews of Interventions [21]. We reported our study in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) checklist (Supplemental material [S1]) [22]. The current study follows the methodology already used in a similar work we previously published [17].
Search strategy and inclusion criteria
We conducted an umbrella review of the literature up to November 2022. We used PubMed, Embase and Cochrane Central. Our search terms were: “[polymorphism genetic OR solute carrier organic anion transporter 1B1 OR SLCO1B1 OR organic anion transporter polypeptide 1B1 OR OATP1B1] AND [statin OR rosuvastatin OR fluvastatin OR pravastatin OR simvastatin OR cerivastatin OR lovastatin OR atorvastatin OR pitavastatin] AND [myopathy OR myalgia OR myositis OR rhabdomyolysis OR creatine kinase OR CK]” (see S2 for details). We restricted the search to articles published in English. In addition, we performed a manual search in the references of articles obtained by the automatic search. Three reviewers (AG, IA, SG) independently selected the articles based on title and abstract then on full text. Any disagreement over article selection was resolved by consensus. Our PICOt criteria were as follow: (i) patient: humans, (ii) intervention: any statin, (iii) comparator: any control, (iv) outcome: point estimate of SLCO1B1-SAMS association, (v) type of publication: systematic reviews and meta-analyses. Our inclusion criteria were: (i) systematic reviews published in peer-reviewed journals [23], (ii) examining the association between SLCO1B1 genotype and SAMS, (iii) including clinical studies, (iv) published in English, and (v) calculation of point estimate using meta-analyses. We focused on English-published studies as they are more likely to be used in guidelines. Indeed, the CPIC guideline on the topic was restricted to English articles for the SLCO1B1 assessment [12]. In this way, our results are representative of the literature that is used for such guideline elaboration.
Data extraction and quality assessment of included studies
Two independent reviewers (AG, IA) extracted the data using a standardized form. Extracted data included study characteristics (e.g. authors, year of publication), information on the statins studied (e.g. type, dose), genetic variants of the SLCO1B1 gene, muscle pain criteria, measures of association‒‒odds ratios (OR) and confidence intervals (CI), and adjustments for potential covariates. When some data of a clinical study were missing in the included systematic reviews, we extracted them from the original clinical study report. The risk of bias in the clinical studies was the one provided by the included systematic review, following the Cochrane Handbook recommendations for umbrella review [21].
Dataset
Our dataset included all point estimates as OR and its 95% CI of the association between SAMS and SLCO1B1 genotype collected across the included systematic reviews. Clinical studies included in more than one systematic review were used only once (S3). The dataset also included characteristics associated with the point estimates: type of statin, type of polymorphism and genetic model (homozygous comparison, heterozygous comparison, dominant, allelic) used for subgroup analyses.
Primary, secondary and exploratory analyses
Our primary analysis integrated all point estimates: any SLCO1B1 single nucleotide Polymorphism (SNPs), any statin. In the primary analysis, we prioritized the homozygous (CC vs TT) and heterozygous (TC vs TT) comparisons to avoid overlapping measurements for the same clinical study.
Our secondary analysis assessed publication bias in different subgroups: each SLCO1B1 SNP, focus each statin type and each genetic model (heterozygous comparison, homozygous comparison, allelic model - C vs T - or dominant model - TC + CC vs TT - ) when possible (≥ 10 point estimates needed for Egger’s test) [24]. We also performed an analysis focused on SNP rs4149056 because it is the most incriminating SNP in SAMS.
We also conducted exploratory analyses to look for a source of publication bias according to three factors: clinical study sponsors, statin drug names, and ethnic groups. For these three potential sources, we plotted the main funnel plot but distinguishing private versus public sponsors for the first one, statin drug names for the second one, and according to the ethnic groups as reported in the included systematic reviews for the third one, although the reporting of population differences in clinical research is inconsistent and plagued with methodological issues [25].
Statistical analysis
We first used the funnel plot to assess the presence of publication bias. Two independent researchers (AG, IA) visually analyzed the funnel plots and classified them as: “suggestive of a publication bias”, “not suggestive of a publication bias”, or “not assessable”. Agreement was estimated using Free-marginal kappa estimator [26]. A third researcher (GG), blinded to the previous classifications, helped resolve disagreements. Second, we used Egger’s test for associations with at least 10 point estimates [24] . We did not use Begg’s test [27]. When the Egger’s test was significant, we added a sensitivity analysis using the Robust Bayesian Meta- Analysis (RoBMA) approach (see “sensitivity analysis” section) [28].
We estimated the SLCO1B1-SAMS association at the meta-analysis level i) without and ii) with a correction for publication bias. We used an inverse variance weighting random-effect meta- analysis to estimate the uncorrected OR (ORUncorrected) and its 95% confidence interval (95% CI). Inter-study heterogeneity was assessed using the Cochran’s Q test, with significance set at p < 0.05, and the I2 was calculated [29]. When publication bias was confirmed by a significant p-value of the Egger’s test, we estimated a corrected OR (ORTrim&Fill) and its 95% CI, using the T&F method [20]. We then compared the estimate of the SLCO1B1-SAMS association i) qualitatively (significant or not significant association) and ii) quantitatively by calculating the ratio of the OR (ORTrim&Fill / ORUncorrected ) and its 95% CI (RORTrim&Fill) (see for details S4).
Sensitivity analysis
We conducted sensitivity analyses using the recently proposed Robust Bayesian Meta-Analysis (RoBMA) approach [28]. RoBMA integrates several publication bias detection and correction methods, notably selection models that are less sensitive to heterogeneity [30]. RoBMA results use Bayes factors (BF), a continuous measure of evidence in favor of the presence or absence of effect, heterogeneity and publication bias. Bayes factor values above 10 indicate very strong evidence, from 3 to 10 moderate evidence, and from 1 to 3 weak evidence. In several simulation studies, RoBMA was found to be superior to other bias correction methods [28]. RoBMA has been used notably in the field of psychology [31–33]. We used RoBMA to provide a sensitivity analysis of i) the publication bias detection, when the Egger’s test was significant and ii) the evaluation of the publication bias effect, by providing a corrected ORRoBMA and comparing it to the ORUncorrected as previously described with the T&F method. The ratio ORRoBMA divided by ORUncorrected being RORRoBMA.
We conducted the analyses using the R 4.2.2 software [34] and the package meta (6.2-1) [35] and RoBMA (2.3.1). All p-values were considered significant at 0.05 without multiple testing adjustment.
RESULTS
Study selection and characteristics
Out of 436 identified references, we included 6 systematic reviews, totaling 19 original clinical studies, from which we extracted 62 unique point estimates of the association between the SLCO1B1 genotype and SAMS (Figure 1).
All clinical studies used wild-type (WT) statin-tolerant patients as a control group, leading to an observational design for the assessment of this association for all studies (losing the benefit of randomization when they were part of a randomized trial). They evaluated one or more statins, mainly simvastatin and atorvastatin; 6 did not specify the statin drugs (statin mixed). Eighteen clinical studies evaluated rs4149056 single nucleotide polymorphisms (SNPs), 7 evaluated rs2306283 and 3 evaluated rs4363657 of the SLCO1B1 gene (Table 1). The criteria for assessing myopathy varied between studies, with the main criterion being muscle symptoms, irrespective of elevated CK levels. However, the reporting of the muscle adverse event was highly heterogeneous, precluding any subgroup constitution. The study population was predominantly of European descent. One clinical study was sponsored by a pharmaceutical company.
The risk of bias of the clinical studies was assessed using the Newcastle-Ottawa Scale (NOS) in 5 reviews and the Risk of Bias Score for Genetic Association Studies in 1 review (i.e. Turongkaravee, 2021). The overall risk of bias was low as all the point estimates of the SLCO1B1-SAMS associations were assessed using non-randomized comparisons (for details see S3,S5).
Primary analysis
Detection of the presence of publication bias
We included 62 point estimates in the primary analysis. The visual inspection of the funnel plots was suggestive of publication bias (kappa agreement between the two reviewers: 64%). The visual inspection also revealed a possible clustering of the estimates around the significant p- value thresholds (Figure 2). The asymmetry was confirmed by Egger’s test (p=0.001) (Table 2). The sensitivity analysis with RoBMA showed strong evidence for publication bias with a BFPublication-bias of 18 (Table 2, S6).
Evaluation of the impact of publication bias on the association estimate
The uncorrected meta-analysis yielded an estimate with a 31% significant increased risk of SAMS with SLCO1B1 genotype (ORUncorrected = 1.31 CI 95% 1.13–1.53), with substantial heterogeneity (Cochran’s Q test p<0.05, I2=64%) (S7). Conversely, the corrected meta-analysis suggested no effect (ORTrim&Fill = 1.07, CI 95% [0.89–1.30]). Correcting for publication bias led to a loss of the association and a 18% nonsignificant change in the association estimate (RORTrim&Fill= 0.82, CI 95% 0.64–1.04) (Table 3).
The corrected meta-analysis using RoBMA also suggested no effect (ORRoBMA=1.02, CI 95% 1.00–1.33) in the sensitivity analysis. Correcting for publication bias resulted in a loss of the observed association, with a 23% significant change in the association estimate (RORRoBMA=0.77, CI 95% 0.55–0.86) (Table 3).
Secondary analysis
Detection of the presence of publication bias by statin type
We examined subgroups by statin type by incorporating 16 point estimates for simvastatin, 15 for atorvastatin, and 12 for rosuvastatin. The visual inspection of funnel plots and Egger’s tests suggested the presence of publication bias for simvastatin and atorvastatin (Figure 3 and Table 3). However, the visual inspection for atorvastatin revealed an asymmetry caused by non- significant point estimates. For rosuvastatin, the funnel plot and Egger’s test were not suggestive of publication bias.
The sensitivity analysis with RoBMA showed weak evidence of publication bias for simvastatin BFPublication-bias = 2.73 and atorvastatin BFPublication-bias =3.13. We did not perform this analysis for rosuvastatin because of the result of the Egger’s test.
Evaluation of the impact of publication bias in secondary analysis by statin type
Only simvastatin and atorvastatin uncorrected meta-analyses were statistically significant (Table 4), with substantial heterogeneity for simvastatin (Cochran’s Q test p=0.02, I2=77%) and atorvastatin (Cochran’s Q test p=0.02, I2=5%) (S7). Correction for publication bias with the T&F’s method resulted in the loss of association for simvastatin (ORTrim&Fill=1.00, CI 95% 0.58–1.72) and atorvastatin (ORTrim&Fill=1.13, CI 95% 0.89–1.43). Correcting for publication bias led to a loss of the association for simvastatin and atorvastatin with a 40% and 16% nonsignificant change in the association estimate, respectively (Table 3).
The corrected meta-analysis for simvastatin (ORRoBMA=1.11, CI 95% 0.91–2.08) and atorvastatin (ORRoBMA=1.02 CI 95% 0.93–1.36) using RoBMA was also inconclusive. Correcting for publication bias led to a loss of the association, with a 34% and 24% nonsignificant change in the association estimate, respectively (Table 3).
Detection of the presence of publication bias by SNP type
When subgrouping according to SNP type, we included 42 point estimates for rs4149056, 14 point estimates for rs2306283, and 6 point estimates for rs4363657 with substantial heterogeneity for rs4149056 (Cochran’s Q test p<0.001, I2=65%) and moderate heterogeneity for rs2306283 (Cochran’s Q test p=0.5, I2=35%) and rs4363657 (Cochran’s Q test p=0.1, I2=0%) (S7). Visual inspection of the funnel plot for rs4149056 was suggestive of publication bias (Figure 4), consistent with Egger’s test (Table 2). Conversely, it was not suggestive for SNP rs2306283 with nonsignificant Egger’s test (p=0.8) and undeterminable for rs4363657.
The sensitivity analysis with RoBMA showed moderate evidence of publication bias (BFPublication bias=6) for rs4149056. We did not perform this analysis for SNP rs2306283 and rs4363657.
Evaluation of the impact of publication bias by SNP type
In the uncorrected meta-analysis, we observed a significant association only for rs4149056 (ORUncorerrected=1.63, CI 95% 1.34–1.99). Correction for publication bias with the T&F’s method resulted in loss of the association (ORTrim&Fill=1.23, CI 95% 0.96–1.58) and a 25% nonsignificant change in the association estimate (RORTrim&Fill= 0.76, CI 95% 0.55–1.04) (Table 3).
The correction using RoBMA also resulted in loss of the significance of the association (ORRoBMA=1.11, CI 95% 1.00–1.77) and a 32% significant change in the association estimate (RORRoBMA=0.68, CI 95% 0.55–0.85) (Table 3).
Other secondary analyses
For secondary analyses focusing on rs4149056, visual inspection of the funnel plots suggested the presence of publication bias for simvastatin, atorvastatin, mixed statin, heterozygous, and homozygous comparisons (S8-9). Egger’s tests were significant only for simvastatin and heterozygous comparisons, and RoBMA was in favor of publication bias. The corrected meta- analysis for simvastatin (ORTrim&Fill=1.04, CI 95% 0.55-1.96) and heterozygous comparison (ORTrim&Fill=1.04, CI 95% 0.75-1.44) was inconclusive, as was the sensitivity analysis using RoBMA (S10).
Exploratory analyses
The results of the exploratory analysis do not suggest the presence of any notable publication bias, either in terms of type of statin used or study sponsor and ethnic group (S11–13).
DISCUSSION
Summary of evidence and implication
In our meta-epidemiologic study, we demonstrated the presence of publication bias in the pharmacogenetics of SAMS. This publication bias was sufficient to explain the suggested association. Our findings are corroborated by the sensitivity analyses using state of the art methods. To our knowledge, our meta-epidemiologic study is the first to specifically assess publication bias in pharmacogenetics of SAMS.
Previous meta-analyses generally used a small number of estimates (i.e., <10) or did not assess publication bias [36–41]. This may explain why this publication bias was not brought to light earlier. Nevertheless, Jiang et al. reported potential publication bias for the heterozygote comparison and the dominant model, but did not use a method to correct this bias [18]. We also detected publication bias for the heterozygote comparison, and correction of this resulted in a loss of the association (Table 2, Table 3). However, we were not able to assess publication bias for the dominant model, due to the low number of estimates (k = 5). Xiang et al. highlighted potential publication bias for the heterozygote comparison, the dominant and allelic model, and corrected this bias with the trim and fill method, which resulted in a loss of association for the heterozygote comparison [19]. As for Jiang et al., we found publication bias for the homozygous comparison and the correction also resulted in the loss of the association. However, we did not perform the test for the allelic model due to the low number of associations (k = 8), like the dominant model (k = 5). Their studies may have lacked the power to detect global publication bias in pharmacogenetics of SAMS, but remain consistent with our results.
Consequently, our results cast doubt on the validity of adjusting statin dose or type according to SLCO1B1 genotype in patients in daily clinical practice. The association between SLCO1B1 polymorphisms and SAMS needs to be specifically evaluated in a methodologically appropriate study before further investigating its clinical use, as it may be a false-positive. The pharmacogenetic guidelines should take the publication bias into account when formulating guidance. Furthermore, all pharmacogenetic clinical studies should also be a priori registered and systematically published, independent of their results.
Strengths of the study
In our study, we assessed any publication bias using many point estimates (i.e. between 10 and 62), with different levels of heterogeneity (i.e. I2 between 0 % and 79 %), with complementary methods such as funnel plots, Egger’s test, T&F’s method. To make our publication bias assessment more reliable, we extended our analyses to the RoBMA method. This recent approach has several advantages: first, the Bayesian framework allows us to provide evidence for or against publication bias without having to make all-or-nothing decisions of frequentist publication bias tests with BF [28]. Secondly, this method uses multiple selection models, which are considered the most reliable methods for detecting publication bias when there is significant heterogeneity [30]. RoBMA also provides a more insightful interpretation of results with the use of BF, which provides a continuous measure of evidence for the presence or absence of effect, heterogeneity and publication bias. This approach gave supportive evidence for publication bias, reinforcing our previous findings using conventional techniques. We corrected publication bias using T&F method and RoBMA, which resulted in loss of the association in the main analysis and in the subgroups studied. These results raise questions about the strength of the association originally reported and underscore the critical importance of assessing publication bias in meta-analyses. The observed loss of association may be explained by the fact that some negative or less pronounced estimates in favor of an increased risk of SAMS may have been underrepresented or less frequently published, leading to an overestimation of the association in the literature. Other factors, such as publication practices or P-hacking, may have exacerbated publication bias too [42]. The source of publication bias was also investigated. We examined the asymmetry of the funnel plots showing the main association by statin type or sponsor of the original clinical trials. We observed that simvastatin contributed to the asymmetry of the funnel plot. Analysis by funding type showed that estimates from industry- funded trials had a non-significative trend (S12).
Limitations
Our study is subject to several limitations. Regarding the limitations of the evidence included in the review, all trials have an observational design for testing the association between a potential biomarker and the treatment effect. Therefore, the potential treatment effect modification is estimated with the corresponding high risk of bias, not protected by randomization. Considering the limits of the review process itself, first, none of the publication bias correction methods can be considered fully reliable, as the actual data generation process remains unknown. Factors, such as the degree of P-hacking, the degree of publication bias and heterogeneity may influence the results. Therefore, despite our efforts to correct publication bias, it is important to remain cautious when interpreting the results and to recognize the inherent limitations of any statistical correction for this type of bias. Secondly, use of many point estimates has resulted in significant heterogeneity, which may have influenced the detection of publication bias using classical symmetry methods – funnel plots, Egger’s test and T&F –, but we obtained consistent results with RoBMA, which strengthens the credibility of our conclusions. Thirdly, we did not use the T&F and RoBMA methods if Egger’s test was not significant. A negative regression test does not guarantee the absence of publication bias [43]. We could have used a 10% significance level to compensate for its low power [24]. This would have made the analyses of several subgroups significant (Table 3, S10), but we preferred to minimize the risk of false positives. In addition, searching for interactions by ROR calculation is associated with a lack of power [44]. Nevertheless, the results of our study are all the stronger for having underestimated publication bias in most of our analyses. In addition, the effect of publication bias was more pronounced for simvastatin, which is known to be a more important substrate of the OATP transporter (SLCO1B1). It would have been interesting to also evaluate the BCRP (ABCG2) transporter with other statins less affected by publication bias that are also substrates of this transporter. This could be an important addition to a future study. We chose to focus on SCLO1B1 because it is the focus of the recent CPIC recommendations.
Conclusion
In this meta-epidemiological study, we provide a cluster of consistent evidence of publication bias in the SLCO1B1- SAMS association. Our analysis revealed the presence of publication bias, which after correction, resulted in the loss of significance of the association. Given the limitations of publication bias adjustment methods, these results do not definitively rule out this association, but they call it into question. This publication bias should be considered in further research. Indeed, failing to consider the risk of publication bias in this example might lead to failure to treat patients who stand to benefit from statins.
Data Availability
The protocol was registered a priori on the Open Science Framework
Sources of Funding
AG was funded by RCTs through a ‘CIFRE’ PhD at the UMR 5558, Université Lyon 1, CNRS.
Disclosures
AG, IA, SG, JCL, JMW, CV, AL, FG, GG declare that they have no competing interest.
Author Contributions
Conceptualization
AG, GG, FG, AL, IA. Data curation: AG, IA, SG. Formal analysis: AG Investigation: AG, GG, FG, AL, JCL. Methodology: AG, GG, FG, AL. Writing– original draft: AG, GG. Writing– review & editing: AG, GG, FG, AL, IA, SG, JCL, JMW, CV
Non-standard Abbreviations and Acronyms
- LDL-C
- Low-density lipoprotein cholesterol
- RCT
- Randomized controlled trial
- CV
- Cardiovascular
- AE
- Adverse event
- SAMS
- Statin-associated muscle symptoms
- CK
- Creatine kinase
- SLCO1B1
- Solute carrier organic anion transporter family member 1B1
- EMA
- European Medicines Agency
- FDA
- Food and Drug Administration
- CPIC
- Clinical Pharmacogenetics Implementation Consortium
- T&F
- Trim-and-Fill
- NOS
- Newcastle-Ottawa scale
- SNP
- Single nucleotide polymorphism
- PRISMA
- Preferred Reporting Items for Systematic Reviews and Meta-Analyses
- OR
- Odds ratio
- CI
- Confidence interval
- ORTrim&Fill
- OR corrected for publication bias by Trim and Fill method
- RORTrim&Fill
- Ratio of odds ratio ( OR corrected by Trim and Fill method)
- ORUncorrected
- OR uncorrected for publication bias
- RoBMA
- Robust Bayesian Meta-Analysis
- BF
- Bayes factor
- ORRoBMA
- OR corrected for publication bias by RoBMA approach
- RORRoBMA
- Ratio of odds ratio ( OR corrected by RoBMA approach)
- SNP
- Single Nucleotide Polymorphism
- WT
- Wild type
Supplemental Material
Supplementary material 1 (S1): PRISMA check list
Supplementary material (S2): Search strategy
Pharmacogenomics keywords : (”polymorphism, genetic”[MeSH Terms] OR (”polymorphism”[All Fields] AND “genetic”[All Fields]) OR “genetic polymorphism”[All Fields] OR (”polymorphism”[All Fields] AND “genetic”[All Fields]) OR “polymorphism genetic”[All Fields] OR ((”organic anion transporters”[MeSH Terms] OR (”organic”[All Fields] AND “anion”[All Fields] AND “transporters”[All Fields]) OR “organic anion transporters”[All Fields] OR (”solute”[All Fields] AND “carrier”[All Fields] AND “organic”[All Fields] AND “anion”[All Fields]) OR “solute carrier organic anion”[All Fields]) AND (”biological transport”[MeSH Terms] OR (”biological”[All Fields] AND “transport”[All Fields]) OR “biological transport”[All Fields] OR “transport”[All Fields] OR “membrane transport proteins”[MeSH Terms] OR (”membrane”[All Fields] AND “transport”[All Fields] AND “proteins”[All Fields]) OR “membrane transport proteins”[All Fields] OR “transporter”[All Fields] OR “transporters”[All Fields] OR “transportable”[All Fields] OR “transportation”[MeSH Terms] OR “transportation”[All Fields] OR “transportations”[All Fields] OR “transported”[All Fields] OR “transporter s”[All Fields] OR “transporting”[All Fields] OR “transports”[All Fields]) AND “1B1”[All Fields]) OR “SLCO1B1”[All Fields] OR ((”organic anion transporters”[MeSH Terms] OR (”organic”[All Fields] AND “anion”[All Fields] AND “transporters”[All Fields]) OR “organic anion transporters”[All Fields] OR (”organic”[All Fields] AND “anion”[All Fields] AND “transporter”[All Fields]) OR “organic anion transporter”[All Fields]) AND (”peptides”[MeSH Terms] OR “peptides”[All Fields] OR “polypeptide”[All Fields] OR “polypeptides”[All Fields] OR “polypeptid”[All Fields] OR “polypeptide s”[All Fields] OR “polypeptidic”[All Fields]) AND “1B1”[All Fields]) OR “OATP1B1”[All Fields])
AND
Statin keywords : statin OR rosuvastatin OR fluvastatin OR pravastatin (”hydroxymethylglutaryl coa reductase inhibitors”[Pharmacological Action] OR “hydroxymethylglutaryl coa reductase inhibitors”[MeSH Terms] OR (”hydroxymethylglutaryl coa”[All Fields] AND “reductase”[All Fields] AND “inhibitors”[All Fields]) OR “hydroxymethylglutaryl coa reductase inhibitors”[All Fields] OR “statin”[All Fields] OR “statins”[All Fields] OR “statin s”[All Fields] OR “statine”[Supplementary Concept] OR “statine”[All Fields] OR “statines”[All Fields] OR (”rosuvastatin calcium”[MeSH Terms] OR (”rosuvastatin”[All Fields] AND “calcium”[All Fields]) OR “rosuvastatin calcium”[All Fields] OR “rosuvastatin”[All Fields]) OR (”fluvastatin”[MeSH Terms] OR “fluvastatin”[All Fields]) OR (”pravastatin”[MeSH Terms] OR “pravastatin”[All Fields] OR “pravastatin s”[All Fields]) OR (”simvastatin”[MeSH Terms] OR “simvastatin”[All Fields] OR “simvastatin s”[All Fields] OR “simvastatins”[All Fields]) OR (”cerivastatin”[Supplementary Concept] OR “cerivastatin”[All Fields] OR “cerivastatin s”[All Fields]) OR (”lovastatin”[MeSH Terms] OR “lovastatin”[All Fields] OR “lovastatine”[All Fields] OR “lovastatin s”[All Fields]) OR (”atorvastatin”[MeSH Terms] OR “atorvastatin”[All Fields] OR “atorvastatine”[All Fields] OR “atorvastatin s”[All Fields]) OR (”pitavastatin”[Supplementary Concept] OR “pitavastatin”[All Fields]))
AND
Adverse event keywords : (”muscular diseases”[MeSH Terms] OR (”muscular”[All Fields] AND “diseases”[All Fields]) OR “muscular diseases”[All Fields] OR “myopathies”[All Fields] OR “myopathy”[All Fields] OR (”myalgia”[MeSH Terms] OR “myalgia”[All Fields] OR “myalgias”[All Fields]) OR (”myositis”[MeSH Terms] OR “myositis”[All Fields] OR “myositides”[All Fields]) OR (”rhabdomyolysis”[MeSH Terms] OR “rhabdomyolysis”[All Fields] OR “rhabdomyolyses”[All Fields]) OR (”creatine kinase”[MeSH Terms] OR (”creatine”[All Fields] AND “kinase”[All Fields]) OR “creatine kinase”[All Fields]) OR “CK”[All Fields])
Search strategy PubMed details
Pharmacogenomics keywords : (’polymorphism genetic’/exp OR ’polymorphism genetic’ OR ((’polymorphism’/exp OR polymorphism) AND (’genetic’/exp OR genetic)) OR ’solute carrier organic anion transporter 1b1’/exp OR ’solute carrier organic anion transporter 1b1’ OR ((’solute’/exp OR solute) AND (’carrier’/exp OR carrier) AND organic AND (’anion’/exp OR anion) AND (’transporter’/exp OR transporter) AND 1b1) OR slco1b1 OR ’organic anion transporter polypeptide 1b1’ OR (organic AND (’anion’/exp OR anion) AND (’transporter’/exp OR transporter) AND (’polypeptide’/exp OR polypeptide) AND 1b1) OR oatp1b1)
AND
Statin keywords: (’statin’/exp OR statin OR ’rosuvastatin’/exp OR rosuvastatin OR ’fluvastatin’/exp OR fluvastatin OR ’pravastatin’/exp OR pravastatin OR ’simvastatin’/exp OR simvastatin OR ’cerivastatin’/exp OR cerivastatin OR ’lovastatin’/exp OR lovastatin OR ’atorvastatin’/exp OR atorvastatin OR ’pitavastatin’/exp OR pitavastatin)
AND
Adverse event keywords: (’myopathy’/exp OR myopathy OR ’myalgia’/exp OR myalgia OR ’myositis’/exp OR myositis OR ’rhabdomyolysis’/exp OR rhabdomyolysis OR ’creatine kinase’/exp OR ’creatine kinase’ OR ((’creatine’/exp OR creatine) AND (’kinase’/exp OR kinase)) OR ck)
Search strategy Embase details
ID Search Hits
#1 polymorphism genetic
#2 solute carrier organic anion transporter 1B1 #3 SLCO1B1
#4 organic anion transporter polypeptide 1B1 OR OATP1B1 #5 #1 OR #2 OR #3 OR #4
#6 statin
#7 rosuvastatin
#8 pravastatin
#9 simvastatin
#10 cerivastatin
#11 lovastatin
#12 atorvastatin
#13 pitavastatin
#14 #6 OR #7 OR #8 OR #9 OR #10 OR #11 OR #12 OR #13
#15 myopathy
#16 myalgia
#17 myositis
#18 rhabdomyolysis
#19 creatine kinase
#20 CK
#21 #15 OR #16 OR #17 OR #18 OR #19 OR #20
#22 #5 AND #14 #21
Search strategy Cochrane Central details
Supplementary material 3 (S3): Overlap primary studies in systematic reviews and risk of bias assessment in selected systematic review using the Newcastle-Ottawa scale (NOS)
Supplementary material 4 (S4): Formula of the ratio of odds ratio and its 95% confidence interval
Supplementary 5 (S5): Assessment of the risk of bias (Score for Genetic Association Studies) in original clinical studies included by Turongkaravee et al. 2021
Supplementary material (S6): Bayes factor for the overall association and subgroups
Supplementary material 7 (S7): Proportion of heterogeneity in primary analysis and secondary analysis
Supplementary 8 (S8): Funnel plots of rs4149056 by type of statin
Supplementary 9 (S9): Funnel plot of rs4149056 by type of genetic model
Supplementary material 10 (S10) : Table of supplementary results for secondary analysis focused on rs4149056
Supplementary material 11 (S11): Funnel plot of the primary analysis [SLCO1B1 SNPs – Statins – SAMS] based on statin type
Supplementary material 12 (S12): Funnel plot of the association [SLCO1B1 SNPs – Statins – SAMS] based on study funding
Supplementary material 13 (S13): Funnel plot of the association [SLCO1B1 SNPs – Statins – SAMS] based on ethnic group as reported in systematic reviews
Acknowledgment
We would like to thank Gaëlle Siméon for her help in reviewing this article.