Abstract
Introduction
There is a growing body of evidence suggesting a complex relationship between cholesterol metabolism and bleeding events. Cholesterol serves three fundamental biological functions 1 : (1) maintaining membrane fluidity and permeability through its amphipathic properties, (2) serving as the precursor for all steroid hormones (cortisol, aldosterone, and sex hormones), and (3) forming the backbone of bile acids essential for lipid digestion. Cholesterol is a vital component of cell membranes and is involved in the synthesis of hormones and bile acids. Low-density lipoprotein cholesterol (LDL-C), commonly referred to as “bad cholesterol,” plays a pivotal role in the development of atherosclerotic cardiovascular diseases by contributing to plaque formation within arterial walls.2,3 This physiological duality extends to hemostasis, where cholesterol 4 : (1) modulates platelet membrane stability and aggregation potential, (2) maintains endothelial barrier function, and (3) influences clotting factor activity. These mechanisms may collectively affect bleeding risk.5,6
The use of lipid-lowering drugs, such as HMGCR inhibitors (statins), PCSK9 inhibitors, and NPC1L1 inhibitors, has been widespread for the prevention and treatment of cardiovascular diseases. These medications are effective in reducing LDL-C levels, but there has been a concern that they might also increase the risk of bleeding complications. This concern arises from the fact that these drugs can affect not only cholesterol levels but also other pathways that could potentially impact the balance between hemostasis and bleeding.7,8
Although observational studies have suggested a potential association between LDL-C levels and bleeding events, the causal nature of this relationship remains uncertain. Observational studies are limited by potential biases and confounding factors, which can obscure the true nature of the relationship between LDL-C levels and bleeding outcomes. To overcome these limitations, Mendelian randomization (MR) is increasingly being used as a method to establish causality in observational epidemiology. MR leverages genetic variants that are associated with modifiable risk factors (in this case, LDL-C levels) as instrumental variables to infer causality without the confounding effects of environmental factors. While recent MR studies have demonstrated the cardiovascular benefits of lipid-lowering therapies 9 —including their protective effect against aortic stenosis—their potential bleeding risks remain less well characterized.
Given the potential implications for clinical practice, particularly in the management of patients with dyslipidemia, there is a growing interest in understanding whether LDL-C or lipid-lowering drugs have a causal effect on bleeding outcomes. This study aims to fill this gap by conducting two-sample MR analyses and drug target MR to evaluate the associations of LDL-C levels with risks for three bleeding outcomes: intracerebral hemorrhage (ICH), gastrointestinal bleeding (GIbleeding), and hemorrhage from respiratory passages (HRP). Additionally, the relationship between LDL-C and these bleeding outcomes will be re-examined in a retrospective cohort study involving patients with coronary artery disease (CAD), providing a comprehensive evaluation of the causal effect of LDL-C on bleeding risks. By combining the strengths of MR analyses with the real-world evidence from a retrospective cohort study, this comprehensive approach aims to provide robust evidence on the causal effect of LDL-C and lipid-lowering drugs on bleeding outcomes.
Methods
The study design is shown in Figure 1. To investigate the causal relationship between LDL-C levels and the risk of bleeding outcomes, we employed a two-pronged approach utilizing both MR and a retrospective cohort study.
Study flowchart. BARC: Bleeding Academic Research Consortium; CAD: coronary artery disease; GIBleeding: gastrointestinal bleeding; HRP: hemorrhage from respiratory passage; ICH: intracerebral hemorrhage; IVW: inverse-variance weighted; LDL-C: low-density lipoprotein cholesterol; MR: Mendelian randomization; SMR: summary-data-based Mendelian randomization.
Mendelian randomization analyses
Genetic instrument selection
In the two-sample MR approach, we utilized summary statistical data for LDL-C from the Global Lipids Genetics Consortium (Supplemental file 1—Table 1), comprising a sample size of 173,082 individuals. 10 This dataset was employed to identify single nucleotide polymorphisms (SNPs) associated with LDL-C levels, with a focus on common variants having a minor allele frequency greater than 1%. We selected genetic variants associated with LDL-C at a genome-wide significance level (P < 5 × 10−8) as instrumental variables. To ensure independence among these variants, we applied a linkage disequilibrium (LD) threshold of r2 < 0.001 using PLINK and the phase 3 version 5 of the 1000 Genomes Project as the reference panel. We estimated the F statistics of instruments variables (using the square of the β divided by square of the standard error 11 ) with an F greater than 10 being suggestive of adequate instrument strength. 12
For the drug target MR approach, we investigated three classes of lipid-lowering drugs: HMGCR inhibitors, PCSK9 inhibitors, and the NPC1L1 inhibitor. Two complementary drug-target MR approaches were employed: a proximal approach using Summary Data-Based Mendelian Randomization (SMR) and a distal approach using two-sample Mendelian randomization. Initially, we leveraged established expression quantitative trait loci (eQTLs) for the target genes of these drugs—HMGCR, PCSK9, and NPC1L1—to serve as proxies for drug exposure. For drug target MR, HMGCR eQTLs were obtained from the eQTLGen Consortium (https://www.eqtlgen.org/), and cis-eQTLs for PCSK9 and NPC1L1 from GTEx Consortium V8 (https://gtexportal.org/), the details of which are presented in Supplemental file 1—Table 1. We identified common eQTL SNPs with a minor allele frequency greater than 1% that were significantly associated (P < 5.0 × 10^−8) with the expression levels of HMGCR or PCSK9 in blood and NPC1L1 in adipose subcutaneous tissue. It is noteworthy that for NPC1L1, no eQTLs in blood or other tissues were available at the required significance level. We exclusively considered cis-eQTLs, defined as those located within 1 Mb of the gene they regulate, in constructing the genetic instruments for this study. SMR was conducted using SMRinR v0.5.0 R package with HEIDI test (P < 0.01 to exclude linkage-driven associations). Subsequently, to corroborate the association observed through eQTLs, we developed an additional instrument by selecting SNPs within 100 kb windows of each drug's target gene that were associated with LDL cholesterol levels at a genome-wide significance level (P < 5.0 × 10^−8). This approach allowed us to proxy the exposure to lipid-lowering drugs.
Outcome sources
The bleeding outcomes under investigation included ICH, GIbleeding, and HRP. The corresponding GWAS data for these outcomes were sourced from the UK Biobank. 13
Retrospective cohort study
Study population and data collection
The Real-World Study of Coronary Heart Disease Diagnosis and Treatment in Tianjin City (ClinicalTrials.gov Identifier: ChiCTR2400094021) is a large prospective cohort study containing patients diagnosed with coronary heart disease from Tianjin. The data provider, Tianjin Health and Medical Big Data Co., Ltd, is authorized to be responsible for the data collection, governance, and application of the Platform. The Platform collects and aggregates clinical diagnosis and treatment data from 43 tertiary and 39 secondary hospitals in the Tianjin area, as well as data from the public health system. After normalization and de-identification on the Platform, the above data are transformed into a mutually accessible and available scientific research application database. The reporting of this study conforms to the STROBE guidelines. 14 The CAD specialized database includes patients who were hospitalized at least once between 1 January 2010 and 31 March 2024, with discharge diagnoses including CAD. To further validate the MR findings, we enrolled patients from this CAD cohort. Exclusion criteria included: (1) less than three available lipid test profiles during 2-year follow up and (2) the interval between the first and the last blood lipid measurements < 3 months. Patients’ demographics, medical history, laboratory test results, echocardiographic, and angiographic evaluation results were collected and verified using an electronic medical recording system. The outcomes of bleeding including ICH and major bleeding event were collected and recorded during 2-year follow-up visits. ICH is defined by the rupture of a cerebral blood vessel and the entry of blood into the brain parenchyma. Major bleeding event is defined by the Bleeding Academic Research Consortium (BARC) 15 grades 3 to 5.
Data sources and quality control
For the MR analyses, we extracted summary statistics from publicly available GWAS datasets, ensuring that the data met quality control standards, including genotyping quality, sample size, and imputation procedures. For the cohort study, data were extracted from a well-maintained electronic health record system with regular audits to ensure data accuracy and completeness.
Ethical considerations
This study was conducted in accordance with the ethical principles of the Helsinki Declaration of 1975, as revised in 2024. Ethical approval for the retrospective cohort study was obtained from the institutional review board, and informed consent was waived due to the use of de-identified data. For the MR analyses, ethics approval was not required as the data were derived from publicly available GWAS datasets.
Statistical analysis
Primary MR analysis
Primary causal estimates were derived via inverse-variance weighted (IVW) regression using the TwoSampleMR v0.6.5 R package. Sensitivity analyses included MR-Egger regression (testing for directional pleiotropy; intercept P > 0.05 indicated no bias), weighted median (robust to ≤50% invalid instruments), simple mode, and weighted mode (both resistant to outlier variants). Heterogeneity was evaluated via Cochran's Q test (P < 0.05 deemed significant). For further investigation of highly heterogeneous SNPs, we applied radial MR analysis to identify and remove potential “outliers,” thereby yielding more robust and reliable results.
The proximal drug target MR
The proximal drug target MR approach, implemented using SMR, was utilized to generate effect estimates when eQTLs as instrumental variables. This approach examines the association between gene expression levels and the outcomes of interest using summary-level data from GWAS and eQTL investigations. 16 The SMRinR package (version 0.5.0) was employed for allele harmonization and analysis. This package facilitates rapid analysis across six major proteomic cohorts, namely deCODE, Fenland, UKB-PPP, INTERVAL, FinnGen_Olink, and FinnGen_Somascan. Data for these cohorts were directly accessed online via the SMRinR package, thereby eliminating the need for additional data processing fees. To evaluate whether the observed association between gene expression and the outcome was driven by linkage disequilibrium rather than a direct causal effect, we employed the heterogeneity in dependent instruments (HEIDI) test. A HEIDI test result with a P-value <0.01 suggests that the association is likely attributable to linkage rather than a true causal relationship.
The distal drug target MR
We selected genetic variants associated with LDL-C to serve as instrumental variables, consistent with the approach used in the primary Mendelian randomization analysis. Specifically, we identified variants within ±100 kb of the HMGCR gene (build GRCh37/hg19: chromosome 5: 74632154–74657929) to instrument statins, within ±100 kb of the NPC1L1 gene (chromosome 7: 44552134-44580914) to instrument ezetimibe, and within ±100 kb of the PCSK9 gene (chromosome 1: 55505221-55530525) to instrument PCSK9 inhibitors, such as alirocumab or evolocumab. To maximize the strength of the instrument for each drug, SNPs used as instruments were allowed to be in low weak linkage disequilibrium (r2< 0.10) with each other. We instrumented LDL using genome-wide significant variants throughout the genome that had pairwise correlations of r2 < 0.001, excluding variants within the three drug target gene regions described previously. We also employed co-localization analysis to assess whether genetic variants identified as significant in both exposure and outcome datasets were likely to be driven by the same underlying genetic signal.
Retrospective cohort study
Continuous variables were reported as mean ± SD or median and interquartile range. Student's t-test is used for normal distribution, and the Mann–Whitney test is used for abnormal distribution.
First, we analyzed the relationship between the mean, SD, and coefficient of variation (CV; defined by
All data analyses were conducted using R version 4.4.0. Specifically, allele harmonization and subsequent analyses were performed utilizing the TwoSampleMR v0.6.5 and SMRinR v0.5.0 packages within the R environment. Statistical significance was determined by a P-value threshold of <0.05.
Results
Primary MR analysis
For the primary analysis that used LDL-C data from the GLGC (Supplemental file 1—Tables 1 and 2). The IVW method was used as the primary approach to estimate the causal effect of LDL-C levels on bleeding outcomes including ICH, GIbleeding, and HRP. As shown in Figure 2, the IVW analysis found no significant causal effect in these bleeding outcomes (P > 0.05). To assess the robustness of our findings, we conducted several sensitivity analyses using alternative MR methods, including MR–Egger regression, weighted median, simple mode, and weighted mode analyses. These analyses provided consistent results with the primary IVW analysis, supporting the robustness of our findings. Heterogeneity and Pleiotropy of MR analysis were also performed as shown in Supplemental file 1—Tables 3 and 4.
Associations between LDL-C and risk of bleeding outcomes including ICH, GIbleeding, and HRP. CI: confidence interval; GIBleeding: gastrointestinal bleeding; HRP: hemorrhage from respiratory passage; ICH: intracerebral hemorrhage; LDL-C: low-density lipoprotein cholesterol; OR: odds ratio.
The proximal drug target MR
We identified a total of 10 cis-eQTLs from the eQTLGen and GTEx Consortium for the target genes of the drugs HMGCR, PCSK9, and NPC1L1, respectively. The most significant cis-eQTL SNP was selected as the genetic instrument for each target gene (Figure 3, Supplemental file 1—Table 5). In Figure 3, the SMR analysis revealed no significant associations between the expression levels of HMGCR, PCSK9, and NPC1L1 and bleeding outcomes, including ICH, GIbleeding, and HRP.
SMR association between expression of gene HMGCR, PCSK9, or NPC1L1 and bleeding outcomes, including ICH, GIBleeding, and HRP. SMR method was used to assess the association. CI, confidence interval; GIBleeding: gastrointestinal bleeding; HRP: hemorrhage from respiratory passage; ICH: intracerebral hemorrhage; OR: odds ratio; SMR: summary-data-based Mendelian randomization.
The distal drug target MR
As depicted in Figure 4 and Supplemental file 1—Table 6, the IVW-MR analysis revealed suggestive evidence for an association between HMGCR-mediated LDL-C and the risk of ICH, with an odds ratio (OR) of 1.0018 (95% confidence interval (CI), 1.0001–1.0035; P = 0.043). Sensitivity analyses showed consistent estimates, with no statistical evidence for bias from horizontal pleiotropy. No heterogeneity or horizontal pleiotropy was detected. The posterior probability of co-localization was calculated for each variant, and no significant evidence of co-localization was found (< 0.5). This suggests that the genetic variants influencing LDL-C in the HMGCR gene and ICH are likely independent and not driven by a shared causal variant. Estimates using all LDL-associated SNPs outside the HMGCR, PCSK9, and NPC1L1 gene regions are also presented. Notably, LDL-C-associated SNPs outside the HMGCR regions yielded nonsignificant estimates for the causal effect of LDL-C on ICH (OR = 1.0000, 95% CI = 0.9996–1.0005; P = 0.87). Furthermore, IVW-MR analysis provided no evidence for associations between PCSK9-mediated LDL-C, NPC1L1-mediated LDL-C, and bleeding outcomes, including ICH, GIbleeding, and HRP.
Two-sample drug target MR association between genetically proxied lipid-lowering drugs and bleeding outcomes, including ICH, GIBleeding, and HRP. CI: confidence interval; GIBleeding: gastrointestinal bleeding; HRP: hemorrhage from respiratory passage; ICH: intracerebral hemorrhage; MR: Mendelian randomization; OR: odds ratio.
Retrospective cohort study
The clinical baseline information is described in Supplemental file 1—Tables 7 and 8. A total of 81,565 cases were included in the study, among which 189 cases developed ICH, and 162 patients experienced BARC criteria 3 to 5 grade bleeding. Supplemental Figure 1 presents the results of the restricted cubic spline regression analysis, which evaluated the relationship between LDL-C levels and the risk of ICH and BARC criteria 3 to 5 grade bleeding within the CAD cohort. Neither the mean, SD, nor the CV of LDL-C levels exhibited a significant association with the risk of ICH or BARC criteria 3 to 5 grade bleeding. However, as depicted in Figure 5, lower LDL-C levels within the initial 3 months were significantly associated with an increased risk of BARC type 3 to 5 bleeding (P < 0.001), and this association remained statistically significant after multivariable adjustment (adjusted P < 0.001) (Supplemental Figure 2). While univariate analysis suggested a trend toward higher ICH risk with LDL-C reduction (P = 0.051), this relationship was attenuated after adjustment for covariates. Importantly, previous ICH history emerged as a strong independent predictor for subsequent hemorrhage events.
Unadjusted association between LDL-C levels during the final 3-month follow-up period and risks of ICH and major bleeding (BARC types 3 to 5) in patients. BARC: Bleeding Academic Research Consortium; ICH: intracerebral hemorrhage.
Discussions
This study comprehensively evaluated the causal relationship between LDL-C levels and bleeding outcomes, including ICH, GIbleeding, and HRP, using multiple analytical approaches. We found that reduced LDL-C during the initial 3-month period could potentially increase BARC type 3 to 5 bleeding risk. Furthermore, statin therapy may confer an additional risk of ICH independent of its LDL-lowering effects.
In a prospective study, individuals with LDL-C levels below 70 mg/dL had a significantly higher risk of developing ICH compared to those with LDL-C levels between 70 and 99 mg/dL.5 This association was observed across different subgroups (e.g. age, sex, and hypertension status), suggesting that low LDL-C levels may be an independent risk factor for ICH. Another analysis also found that middle-aged individuals with LDL-C levels below 70 mg/dL had a higher risk of ICH. 17 Furthermore, Xu et al. 18 and Cheng et al. 19 observed that among patients who had an ischemic stroke, the low LDL-C level at baseline is associated with increased risk of ICH and other bleeding events during early stage. Nevertheless, a comprehensive prospective cohort study involving 267,500 Chinese participants in 2019 revealed that LDL-C exhibited no significant association with the incidence of hemorrhagic stroke. 20 Similarly, in the ODYSSEY 21 and FOURIER 22 clinical trials, despite the substantial reduction in LDL-C levels achieved through the administration of PCSK-9 inhibitors, there was no observed increase in intracranial hemorrhage events. Our study builds upon these findings, displaying that lower LDL-C levels may increase the risk of intracranial hemorrhage.
While LDL-C is widely recognized as a major risk factor for atherosclerosis, its association with bleeding may involve multiple underlying mechanisms. Lower LDL-C levels may increase the fragility of red blood cell membranes, thereby affecting vascular integrity. 23 Additionally, lower LDL-C levels may be associated with impaired coagulation function, which could increase the risk of bleeding. 24 Furthermore, lower LDL-C levels have been linked to an increased number of cerebral microbleeds, which are known risk factors for ICH. 25 These mechanisms suggest that excessively low LDL-C levels may increase the risk of ICH through various pathways.
Several important external factors warrant consideration when interpreting our findings. While our study design controlled for major covariates, factors such as: drug–drug interactions (particularly with antiplatelet/anticoagulant medications), dietary patterns (e.g. Mediterranean vs. Western diets), environmental exposures (including air pollution and heavy metals), lifestyle factors (physical activity, smoking, alcohol use), may influence both lipid metabolism and bleeding risk. These factors could potentially modify the observed relationships, though their systematic evaluation would require specifically designed studies with more detailed phenotyping. The development of comprehensive risk prediction models incorporating these multidimensional factors represents an important direction for future research.
Given the significant reduction in atherosclerotic cardiovascular disease risk associated with lowering LDL-C, current guidelines strongly advocate for achieving very low LDL-C levels. In our study, concerning the relationship between LDL-C levels and bleeding events, the primary MR analysis showed that LDL-C levels are not associated with bleeding outcomes. SMR results also suggested that lipid-lowering drugs are not related to bleeding. However, the distal drug target MR analysis indicated that statins may independently increase the risk of ICH apart from lowering LDL-C levels. Additionally, the clinical study also suggested a significant association between low LDL-C levels and bleeding. Despite growing concerns raised by numerous studies that lower LDL-C levels and lipid-lowering drugs, such as statins, may potentially increase the risk of bleeding events, particularly ICH, it is also important to note that the overall event rate remains relatively low (1% for BARC criteria 3 to 5 grade bleeding and 0.3% for ICH). Therefore, it is essential to weight the clinical relevance of this small risk and the large benefit of lipid-lowering therapies. For individuals with a high risk of bleeding, especially ICH, the use of statins should be considered with caution.
Several limitations should be noted. First, although our integrated approach strengthens causal inference, the MR findings remain hypothesis-generating and require further validation through randomized controlled trials. Additionally, while the cohort study is large in scale, its observational nature means results may be confounded by unmeasured factors. Second, since the GWAS data used in our MR analyses were derived exclusively from European populations, the generalizability of these findings to other ethnic groups should be considered with caution. Third, in our clinical CAD cohort used for validation, the majority of patients were receiving antiplatelet therapy, which may independently influence bleeding risk and thus potentially confound the observed associations with LDL-C. Furthermore, while our cohort spans 2010 to 2024, recent data were excluded due to ongoing quality control procedures; future studies should incorporate these updated records once curation is complete. Finally, subsequent in vivo studies are warranted to elucidate the precise biological mechanisms underlying LDL-C's potential role in vascular integrity.
Conclusions
Our MR analyses found no evidence for a causal relationship between LDL-C levels and bleeding outcomes. However, the retrospective cohort data demonstrated an association between recent LDL-C levels and major bleeding risk in patients with CAD. The study suggests statins may modestly increase ICH risk through mechanisms beyond LDL-C reduction. These findings support careful clinical evaluation when prescribing statins to patients at high bleeding risk, while emphasizing the need for further research to clarify these associations.
Supplemental Material
sj-xlsx-1-sci-10.1177_00368504251375876 - Supplemental material for Lipid, lipid-lowering drugs, and bleedings: A Mendelian randomization and retrospective study
Supplemental material, sj-xlsx-1-sci-10.1177_00368504251375876 for Lipid, lipid-lowering drugs, and bleedings: A Mendelian randomization and retrospective study by Shaohua Guo, Sutao Hu, Hui Chen and Kang-Yin Chen in Science Progress
Footnotes
Abbreviations
ORCID iDs
Authors’ contributions
Funding
Declaration of conflicting interests
Data availability statement
Supplemental Material
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
