Our objective was to estimate the causal association between 23 medication-taking traits and the incidence of stroke adopting genome-wide association study (GWAS) summary statistics by a two-sample MR design. This instrumental variable analysis in the random assignment of single-nucleotide polymorphisms (SNPs) is parallel to randomized controlled trials (RCTs) (the impact of confounders like age and gender was not incorporated). In addition, this study was performed on the condition that three key assumptions were met: (i) Genetic instruments have strong predictive power for the exposure of interest (p < 5 × 10⁻⁸); (ii) Genetic instruments are not related to potential confounding factors (no statistical inference); and (iii) Interdependency exists between genetic instruments and the results, taking into account exposure and confounders. (Only through the risk factors do genetic instruments exert impact on the results.) (13) All statistical analyses in this study were run in R software (4.2.1) via the TwoSampleMR packages. The study overview was presented in Figure 1.

After the exclusion of participants taking case drugs and similar drugs, a grand total of 23 case–control medication categories were included in this study. In any case, this should be explained. GWAS data for 23 medication-taking traits were derived from a case–control GWAS meta-analysis conducted among 318,177 United Kingdom Biobank (UKB) study participants from the UK Biobank (14). GWAS data of medication-taking traits were defined as the record of self-reported regular medication and health supplements taken weekly, monthly, or 3-monthly. Scholars coded these related drug and health supplement data in 6,745 categories. In total, 1,809 drug classes were taken by at least 10 participants among 6,745 medication categories. The active ingredients corresponding to these 1,809 categories were unearthed through online information, Drugs (https://www.drugs.com/), and NetDoctor (https://www.netdoctor.co.uk/). They were then categorized using the Anatomical Therapeutic Chemical (ATC) Classification System (15). More detailed information about the GWAS data can be obtained from the study of Wu et al. (14).

IVs associated with medication traits were identified by rigorous screening conditions from multiple perspectives. After a series of GWASs and post-GWAS analysis, SNPs were reported significantly associated as the candidate IVs across 23 medication traits ultimately [p < 5  ×  10⁻⁸, linkage disequilibrium (LD) r² < 0.01]. The F-statistics of the selected IVs were all in excess of the conventional value of weak instruments of F-statistic/10, indicating that the IVs had a strong potential to predict 23 medication categories (16). Next, we extracted medication category-associated SNPs from the outcome and discarded SNPs associated with the outcome (p < 5  ×  10⁻⁸). We further harmonized SNPs for exposure and outcome, and palindromic effects and allelic inconsistent SNPs were removed (e.g., A/G vs. A/C). Details of the 910 SNPs are presented in Supplementary Table S1.

Stroke, an acute cerebrovascular disease, is a rapidly progressing sign of focal (or global) disturbance of cerebral function, which mostly presents with sudden onset, rapid onset of limited, or diffuse brain dysfunction, with the clinical symptoms lasting more than 24 h (17). It is usually divided into two categories based on clinical and radiographic standard: ischemic and hemorrhagic. All summary-level data for stroke and ischemic stroke (all subtypes) in primary analysis were acquired from the publicly available Medical Research Council Integrative Epidemiology Unit (IEU) Open GWAS database (18). More detailed information about the sample characterization, statistical analysis, and genotyping can be found in the original publication (19). In brief, we gathered relevant stroke data from a large GWAS meta-analysis that included 29 genome-wide association studies with solely European individuals, containing a total of 40,585 cases of stroke and 34,217 cases of IS, with 40,611 controls. In accordance with the TOAST classification of stroke, IS participants were subdivided into three types in focus of infarction: large artery atherosclerotic (n = 4,373 cases/40,611 controls), cardioembolic (n = 7,193 cases/40,611 controls), and small vessel (n = 5386 cases/192,662 controls) in this meta-analysis, represented by LAS, CES, and SVS, respectively (19, 20). For hemorrhagic stroke, according to the FinnGen consortium (21), four hemorrhagic stroke outcomes were exhibited, including SAH (1,961 cases, and 281,638 controls), intracerebral hemorrhage (ICH, 2,643 cases and 281,373 controls), nontraumatic intracranial hemorrhage (nITH, 4,283 cases and 284,164 controls), and a combination of SAH, unruptured cerebral aneurysm, and aneurysm operations SAH (AOS) (3,201 cases and 284,164 controls) (21). Aneurysm operations in AOS are endovascular or surgical procedures to intracerebral aneurysms. AOS was involved in hemorrhagic stroke outcomes due to the end-point attribute of SAH in AOS. According to how FinnGen defines the end point, nITH can be viewed as equivalent to hemorrhagic stroke based on the exclusion of other brain hemorrhages (21). As a result, we extracted summary statistics from the R7 release of the FinnGen consortium. In addition, the participants in our MR analyses were all of European descent. GWASs employed in the study are specified in Table 1.

We employed some types of MR approaches [inverse variance weighted (IVW), MR-Egger, and WM] to determine MR estimates of medication-taking for stroke after harmonizing the effect alleles across the GWASs of 23 medication categories and stroke was completed. As a method for meta-summarization of effects at multiple loci, IVW was used as a primary analytical method in a relatively ideal state and its application assumes that all SNPs are valid and completely independent of each other (22, 23). Moreover, several other MR approaches including MR-Egger and WM have been utilized for the detection of the robustness of study results in a wider range of scenarios. As the median of the distribution function obtained by sorting all individual SNP effect values by weight, WM permits fewer than 50% of SNPs to be invalid, yet MR-Egger permits pleiotropy of all SNPs, but it assumes that the effect of genetic variation pleiotropy on the outcome is independent of the effect of genetic variation on exposure factors. In other words, MR-Egger was able to offer a test for heterogeneity and pleiotropy for all SNPs in the existence of horizontal heterogeneity (24, 25). p < 0.002 (Bonferroni correction p = 0.05/23 sets of exposures) is considered to show a remarkable causal relationship. Taking inconsistency into account, the study adopted a tightened instrument p-value threshold (26).

In addition, specific sensitivity analyses were used to evaluate and adjust any potential heterogeneity and pleiotropy to satisfy relevant key assumptions of our MR study design. Cochran's Q statistic was employed to detect heterogeneity. With an aim to guarantee no directional pleiotropy for the linkage between medical traits and stroke of all subcategories, the p-value of the Cochran Q test should exceed 0.05 (27). Furthermore, we conducted the MR-Egger intercept tests to appraise the influence level of directional pleiotropy on risk estimates (25). The intercept value obtained from MR-Egger regression deviating considerably from null (p < 0.05) was employed for indicating the presence of horizontal pleiotropy (28, 29). For further interpretation, we performed the leave-one-out (LOO) analysis, which assesses whether the results are strongly influenced by each exposure-associated SNP and then repeat the IVW analysis in turn by discarding a SNP (28).

To demonstrate the mediating effect of certain medication categories on stroke via some risk factors, we calculated some specific diseases of drug indications representative as potential mediators for further MR analysis, namely, hypertension, type 2 diabetes, gastroesophageal reflux disease (GORD), and total cholesterol. GWASs employed in the study are also specified in Table 1.

Among the medication categories tested, by using medication-related SNPs, IVW initially screened out 10 medication categories (HMG CoA reductase inhibitors, antihypertensives, diuretics, drugs for peptic ulcer and GORD, antithrombotic agents, calcium channel blockers, agents acting on the renin-angiotensin system, drugs used in diabetes, beta blocking agents, and salicylic acid and its derivatives) significantly associated with stroke (Supplementary Tables S2, S3 and Figure 2A), implying a high-risk function on stroke. Meanwhile, a consistent direction in MR-Egger and weighted median supported the robustness of the result.

In addition, the MR-Egger intercept testing suggested no sign of a possible horizontal pleiotropy effect (Supplementary Table S4). However, with the exception of a small number of medication categories, defined as antihypertensives and drugs for peptic ulcer and GORD, heterogeneity was examined in Cochran's Q test analysis between categories mentioned above and stroke, including HMG CoA reductase inhibitors (p for Cochran's Q = 1.94 × 10⁻⁶), diuretics (p for Cochran's Q = 1.78 × 10⁻¹²), antithrombotic agents (p for Cochran's Q = 9.70 × 10⁻⁵), calcium channel blockers (p for Cochran's Q = 1.25 × 10⁻⁵), agents acting on the renin-angiotensin system (p for Cochran's Q = 2.69 × 10⁻¹⁴), drugs used in diabetes (p for Cochran's Q = 0.01), beta blocking agents (p for Cochran's Q = 2.36 × 10⁻⁶), and salicylic acid and its derivatives (p for Cochran's Q = 7.55 × 10⁻⁵) (Supplementary Table S5). Despite the existence of heterogeneity in some outcomes, MR estimates were not invalidated as random-effect IVW in this research, which could act as a counterbalance to pooled heterogeneity.

For IS, IVW yielded the estimates of the effect of nine medication categories (HMG CoA reductase inhibitors, antihypertensives, diuretics, drugs for peptic ulcer and GORD, antithrombotic agents, calcium channel blockers, agents acting on the renin-angiotensin system, drugs used in diabetes, and beta blocking agents) on IS (Supplementary Tables S6, S7 and Figure 2B), as shown in the figure. Similar risk estimates were gained using the MR-Egger regression and WM approaches, though sometimes the association was of no statistical significance. p values of the MR-Egger intercept testing all exceeded 0.05, suggesting the absence of horizontal pleiotropy (Supplementary Table S8). At the same time, we detected the heterogeneity among some medication categories, including HMG CoA reductase inhibitors (p for Cochran's Q = 2.91 × 10⁻⁵), diuretics (p for Cochran's Q = 6.08 × 10⁻¹⁰), calcium channel blockers (p for Cochran's Q = 4.89 × 10⁻⁵), agents acting on the renin-angiotensin system (p for Cochran's Q = 4.05 × 10⁻¹⁵), and beta blocking agents (p for Cochran's Q = 1.39 × 10⁻⁶) (Supplementary Table S9). Since the random-effects IVW were employed as the primary outcome, heterogeneity was at an acceptable level (30).

As for subtypes of IS, IVW preliminarily identified six categories significantly associated with SVS, which were beta blocking agents, drugs used in diabetes, agents acting on the renin-angiotensin system, diuretics, HMG CoA reductase inhibitors, and calcium channel blockers. Nevertheless, pleiotropy was observed in the MR-Egger intercept analysis between HMG CoA reductase inhibitors and SVS (p = 0.002) (Supplementary Table S12). Meanwhile, the result of MR-Egger suggested that no causal relation existed between HMG CoA reductase inhibitors and SVS [odds ratio (OR) = 1.00, 95% CI = 0.839–1.192, p = 0.999]. Therefore, we would regard this as an insignificant estimate. Eventually, only five of them met the criteria involved in the development of SVS (Supplementary Tables S10, S11 and Figure 3A), including beta blocking agents, drugs used in diabetes, agents acting on the renin-angiotensin system, diuretics, and calcium channel blockers. In other words, these five categories were linked to a high SVS risk. The Q test outcome demonstrated heterogeneity between beta blocking agents and SVS (p for Cochran's Q = 0.0001), agents acting on the renin-angiotensin system and SVS (p for Cochran's Q = 4.12 × 10⁻⁵), diuretics and SVS (p for Cochran's Q = 8.25 × 10⁻⁵), and calcium channel blockers and SVS (p for Cochran's Q = 0.0004) (Supplementary Table S13).

For LAS, nine categories (calcium channel blockers, HMG CoA reductase inhibitors, agents acting on the renin-angiotensin system, antithrombotic agents, diuretics, beta blocking agents, salicylic acid and its derivatives, drugs used in diabetes, and vasodilators used in cardiac diseases) were associated with the advance of LAS in the IVW analysis (Supplementary Tables S14, S15 and Figure 3B). Other two MR methods yielded a consistent direction, supporting the robustness of the result. Pleiotropy was observed in the MR-Egger intercept analysis between beta blocking agents and LAS (p = 0.02); the result of MR-Egger suggested that beta blocking agents had evidence of causality on LAS (OR = 2.98, 95% CI = 1.67–5.30, p = 0.0005) (Supplementary Table S16). The Q test result indicated heterogeneity between calcium channel blockers and LAS (p for Cochran's Q = 0.03), HMG CoA reductase inhibitors and LAS (p for Cochran's Q = 0.0009), agents acting on the renin-angiotensin system and SVS (p for Cochran's Q = 1.10 × 10⁻⁸), antithrombotic agents and LAS (p for Cochran's Q = 0.002), diuretics and LAS (p for Cochran's Q = 0.02), beta blocking agents and LAS (p for Cochran's Q = 0.0001), and salicylic acid and its derivatives and LAS (p for Cochran's Q = 0.02) (Supplementary Table S17).

For CES, IVW identified three categories significantly associated with a high risk of CES (Supplementary Tables S18, S19 and Figure 3C), including beta blocking agents, agents acting on the renin-angiotensin system, and calcium channel blockers. Pleiotropy was also observed in the MR-Egger intercept analysis between agents acting on the renin-angiotensin system and CES (p = 0.006); the result of MR-Egger estimated that agents acting on the renin-angiotensin system had the causality on CES (OR 1.65, 95% CI 1.27–2.13, p = 0.0002) (Supplementary Table S20). The Q test outcome showed heterogeneity between agents acting on the renin-angiotensin system and CES (p for Cochran's Q = 8.28 × 10⁻⁷), calcium channel blockers, and CES (p for Cochran's Q = 0.0008) (Supplementary Table S21).

For ICH, IVW identified four categories (agents acting on the renin-angiotensin system, vasodilators used in cardiac diseases, diuretics, and calcium channel blockers) significantly associated with ICH (Supplementary Tables S22, S23 and Figure 4A), as the figure shows. The MR-Egger intercept testing showed no signs of horizontal pleiotropy effect (Supplementary Table S24). Heterogeneity was also detected between agents acting on the renin-angiotensin system and ICH (p for Cochran's Q = 0.03), and diuretics and ICH (p for Cochran's Q = 0.04) (Supplementary Table S25).

For nITH, four categories were linked to a high nITH risk in the IVW analysis (Supplementary Tables S26, S27 and Figure 4B), including beta blocking agents, agents acting on the renin-angiotensin system, diuretics, and calcium channel blockers. Notably, we did not detect evidence of a causal association in two categories [beta blocking agents (p = 0.02) and agents acting on the renin-angiotensin system (p = 0.17)] via the MR-Egger analysis, but a consistent direction made the outcome convincing. Meanwhile, the WM yielded a consistent direction and magnitude, with all p values of four categories greater than 0.05, supporting the robustness of the result. Although we observed the presence of horizontal pleiotropy between calcium channel blockers and nITH in IVW estimates, the significant causal association was verified by MR-Egger regression (Supplementary Table S28). In addition, we can observe heterogeneity in the Q test analysis between beta blocking agents and nITH (p for Cochran's Q = 9.92 × 10⁻⁶), and agents acting on the renin-angiotensin system and nITH (p for Cochran's Q = 0.009) (Supplementary Table S29). Nevertheless, MR estimates were not invalidated as random-effect IVW in this research by the presence of heterogeneity.

For AOS, IVW identified four categories (diuretics, beta blocking agents, calcium channel blockers, and agents acting on the renin-angiotensin system) significantly associated with this combination (Supplementary Tables S30, S31 and Figure 4C). The direction was consistent in MR-Egger and weighted median, supporting the robustness of the result. The p-value for the MR-Egger intercept is >0.05 except for calcium channel blockers, indicating that no horizontal pleiotropy existed (Supplementary Table S32). The positive causal estimates between genetic predisposition toward calcium channel blockers and AOS were still verified using MR-Egger analysis. In addition, despite the detection of heterogeneity in calcium channel blockers (p for Cochran's Q = 0.007) and agents acting on the renin-angiotensin system (p for Cochran's Q = 0.016) with AOS (Supplementary Table S33), the causal relations were still valid due to acceptable heterogeneity (30).

For SAH, IVW identified three categories significantly associated with SAH (Supplementary Tables S34, S35 and Figure 4D), including agents acting on the renin-angiotensin system, diuretics, and beta blocking agents, as can be seen from the figure. Other two MR methods yielded a consistent direction, supporting the robustness of the result. Beyond this, no trace of remarkable intercept indicated that there was no possible pleiotropy observed (Supplementary Table S36). No heterogeneity was detected in Cochran's Q test analysis (Supplementary Table S37).

To explore if the risk factors disturb the causal relationship between medical categories and stroke, we used MR approaches to evaluate the causality between stroke and relevant risk factors such as hypertension, type 2 diabetes, gastroesophageal reflux disease, and total cholesterol. As shown in Table 2, we found that total cholesterol level was significantly associated with stroke (p = 0.004), IS (p = 0.003), and LAS (p = 0.003). Meanwhile, the existence of causal effect GORD on stroke (4.58 × 10⁻⁷) and IS (1.72 × 10⁻⁵) was verified in IVW analysis. For hypertension and type 2 diabetes, no causal effects were observed from these two diseases on stroke and all subtypes, implying that the causality of certain medication categories on stroke phenotypes was not biased by potential risk factors.