目的：鉴于既往降糖药物与缺血性卒中风险关联的研究结果尚不一致，本研究采用孟德尔随机化 （Mendelian randomization，MR）方法探索主要降糖药物对缺血性卒中及其亚型风险的潜在因果关联，并利 用独立欧洲及东亚人群数据进行验证。方法：从已发表研究中提取磺酰脲类（ABCC8/KCNJ11）、GLP-1类 似物（GLP1R）、噻唑烷二酮类（PPARG）、胰岛素及类似物（INSR）、二甲双胍作用靶点相关的遗传变异作为 工具变量（所有工具变量F值＞10）。原始分析基于MEGASTROKE欧洲人群数据，采用逆方差加权法（inverse variance weighted method，IVW）或 Wald ratio 法评估药物靶点对急性缺血性卒中（acute ischemic stroke，AIS）、大动脉粥样硬化型卒中（large artery atherosclerosis stroke，LAS）、心源性栓塞型卒中（cardioembolic stroke，CES）和小血管卒中（small vessel stroke，SVS）的因果效应。外部验证采用：①独立欧洲人群 FinnGen数据验证AIS总体；②GIGASTROKE东亚人群数据对卒中亚型进行探索性分析。本研究为探索性 分析，未对多重比较进行校正，以P＜0.05作为名义关联的统计学标准。结果：原始分析（MEGASTROKE） 显示：磺酰脲类（rs757110）与AIS风险降低存在名义关联（OR=0.05，95% CI 0.01-0.29，P=0.001），但效应极 端；二甲双胍与 LAS 存在名义关联（OR=0.47，P=0.047）；磺酰脲类多个单核苷酸多态性（single nucleotide polymorphism，SNP）工具（KCNJ11/ABCC8）与CES存在名义关联（OR=0.09，P=0.046）。外部验证：在FinnGen中，所有药物与AIS的关联均无统计学显著性（rs757110: OR=0.53，P=0.601；其余P＞0.05）。在东亚人 群探索性分析中，原始亚型阳性信号均无名义关联（二甲双胍与LAS: OR=1.22，P=0.715；磺酰脲类（ABCC8/ KCNJ11）与CES: OR=1.68，P=0.871）的关联均未重复，且置信区间极宽，提示统计功效不足。结论：原始分 析中降糖药物与缺血性卒中的阳性信号在独立外部验证中均未得到支持。磺酰脲类SNP rs757110的极端 保护效应判定为假阳性；二甲双胍与LAS、磺酰脲类与CES的名义关联因外部验证未重复，仅作为探索性假 说。当前遗传学证据不支持任何降糖药物对缺血性卒中及其亚型的临床应用决策。

Previous studies evaluating the association between antidiabetic drugs and the risk of ischemic stroke have yielded inconsistent results. This study employed a Mendelian randomization (MR) approach to preliminarily explore the potential causal associations of major antidiabetic drugs with the risk of ischemic stroke and its subtypes, and performed validation using independent European and Asian population data. Methods: Genetic variants associated with the targets of sulfonylureas (ABCC8/KCNJ11), GLP-1 analogs (GLP1R), thiazolidinediones (PPARG), insulin and its analogs (INSR), and metformin were extracted from published studies as instrumental variables (all instrument F-statistics>10). The primary analysis was conducted using MEGASTROKE European population data. Inverse variance weighting (IVW) or the Wald ratio method was used to estimate the causal effects of drug targets on acute ischemic stroke (AIS), large artery atherosclerosis stroke (LAS), cardioembolic stroke (CES), and small vessel stroke (SVS). External validation consisted of: (1) validation of overall AIS using independent European population data from FinnGen; (2) an exploratory analysis of stroke subtypes using GIGASTROKE East Asian population data. This study was exploratory in nature; no corrections were made for multiple comparisons, and a P-value < 0.05 was used as the statistical criterion for nominal association. Results: The primary analysis (MEGASTROKE) showed that sulfonylureas (rs757110) had a nominal association with reduced AIS risk (OR=0.05, 95% CI 0.01-0.29, P=0.001), although the effect was extreme; metformin showed a suggestive association with LAS (OR=0.47, P=0.047); and the multi-single nucleotide polymorphism (SNP) instrument for sulfonylureas (KCNJ11/ABCC8) showed a suggestive association with CES (OR=0.09, P=0.046). External validation: In FinnGen, none of the associations between the drugs and AIS were statistically significant (rs757110: OR=0.53, P=0.601; all others P>0.05). In the exploratory analysis of the East Asian population, none of the original positive signals for stroke subtypes showed a nominal association [metformin with LAS: OR=1.22, P=0.715; sulfonylureas (ABCC8/KCNJ11) with CES: OR=1.68, P= 0.871]; the associations were not replicated, and the confidence intervals were extremely wide, suggesting insufficient statistical power. Conclusion: The positive signals for antidiabetic drugs on ischemic stroke observed in the primary analysis were not supported by independent external validation. The extreme protective effect of the sulfonylurea SNP-rs757110 was judged as a false positive. The suggestive associations of metformin with LAS and sulfonylureas with CES were not replicated in external validation and should be considered only as exploratory hypotheses. The current genetic evidence does not support the clinical application of any antidiabetic drug for ischemic stroke or its subtypes.