In observational studies, type 2 diabetes is associated with two- to fourfold higher risk of cardiovascular diseases (CVD). Using data from the China Kadoorie Biobank (CKB), we examined associations of genetically predicted type 2 diabetes with CVD among ∼160,000 participants to assess whether these relationships are causal. A type 2 diabetes genetic risk score (comprising 48 established risk variants) was associated with the presence of carotid plaque (odds ratio 1.17 [95% CI 1.05, 1.29] per 1 unit higher log-odds of type 2 diabetes; n = 6,819) and elevated risk of ischemic stroke (IS) (1.08 [1.02, 1.14]; n = 17,097), nonlacunar IS (1.09 [1.03, 1.16]; n = 13,924), and major coronary event (1.12 [1.02, 1.23]; n = 5,081). There was no significant association with lacunar IS (1.03 [0.91, 1.16], n = 3,173) or intracerebral hemorrhage (ICH) (1.01 [0.94, 1.10], n = 6,973), although effect estimates were imprecise. These associations were consistent with observational associations of type 2 diabetes with CVD in CKB (P for heterogeneity >0.3) and with the associations of type 2 diabetes with IS, ICH, and coronary heart disease in two-sample Mendelian randomization analyses based on summary statistics from European population genome-wide association studies (P for heterogeneity >0.2). In conclusion, among Chinese adults, genetic predisposition to type 2 diabetes was associated with atherosclerotic CVD, consistent with a causal association.

Cardiovascular diseases (CVD) remain the major cause of death among individuals with diabetes (1). In observational studies, diabetes has been associated with two- to fourfold higher risks of CVD, in particular various types of atherosclerotic CVD (ASCVD), such as coronary heart disease (CHD), ischemic stroke (IS), and peripheral arterial disease (24). However, it remains unclear whether this reflects a causal effect or confounding by shared genetic or other risk factors. Furthermore, the associations of diabetes with subclinical atherosclerosis (57) and other CVD types, including hemorrhagic stroke (24,8), are less well understood. Although randomized controlled trials can assess causality, existing trial evidence for the effects of intensive glycemic control in diabetes (9), and of interventions to reduce or delay progression of prediabetes to diabetes (10,11), on risk of CVD is inconclusive.

Genome-wide association studies (GWAS) have identified multiple type 2 diabetes–associated genetic variants, which can be used as a genetic instrument for type 2 diabetes in so-called “Mendelian randomization” (MR) studies to help assess causal associations (12). An effect of type 2 diabetes on CVD, independent of environmental and other factors potentially confounding observational risk estimates, can then be inferred from an association of these variants with CVD, subject to the assumption that the effect of the variants is through type 2 diabetes aetiological pathways. Several studies have examined the association of genetically predicted type 2 diabetes with CVD (1318). However, these have mainly involved populations of European ancestry, and focused largely on CHD (13,14,16), with limited evidence on stroke (15), particularly hemorrhagic stroke (17), and subclinical atherosclerosis (18,19). The characteristically high rates of stroke, with a higher proportion of hemorrhagic stroke, in Chinese, compared with Western, populations (20) provide a valuable opportunity to examine the causal relevance of type 2 diabetes for CVD subtypes.

Using data from the prospective China Kadoorie Biobank (CKB), we aim to 1) examine the association of genetically predicted type 2 diabetes with different CVD and with subclinical atherosclerosis and 2) compare the associations of genetically predicted type 2 diabetes with different CVD in Chinese and European populations.

Study Population

Details of the CKB study design and population have previously been published (21). Briefly, 512,713 Chinese adults (210,205 men, 302,508 women), aged 30–79 years at enrollment, were recruited between 2004 and 2008 from 10 diverse areas (5 urban, 5 rural) across China. All participants provided written consent prior to participation, including permission for follow-up. Ethics approval was obtained from Oxford University, the Chinese Center for Disease Control and Prevention (China CDC), and the 10 study areas’ local CDCs.

Data Collection

Participants were interviewed by trained health workers using laptop-based questionnaires, collecting information on sociodemographic and lifestyle factors (e.g., smoking, alcohol intake, educational attainment), medical history, and current medication. Physical measurements were recorded (including blood pressure and anthropometry), using calibrated instruments and standard protocols. A 10-mL nonfasting venous blood sample was collected (with time since participants last ate recorded) for long-term storage and on-site testing of random plasma glucose (RPG) levels (SureStep Plus meter, LifeScan, Milpitas, CA). Individuals with an RPG level ≥7.8 and <11.1 mmol/L were invited to return for a fasting plasma glucose test the next day. Plasma concentrations of total, LDL, and HDL cholesterol and triglycerides were assayed (AU680 analyzer; Beckman-Coulter) (Wolfson Laboratories, Oxford Clinical Trial Service Unit and Epidemiological Studies Unit [CTSU], Oxford, U.K.) among 18,256 participants (including 10,434 incident stroke and 1,287 incident CHD cases) included in a nested case-control study of CVD, of whom 17,948 also had genotyping data. A 5-yearly resurvey of a 5% randomly selected sample of surviving CKB participants was undertaken, collecting the same data as at baseline plus certain enhancements. In the 2013–2014 resurvey (n = 24,822), carotid ultrasound measures (Panasonic CardioHealth Station), including carotid intima-media thickness (cIMT) and assessment for the presence of carotid artery plaque, were undertaken (22).

Follow-up for Morbidity and Mortality

The vital status of study participants was confirmed by active annual follow-up through local residential and administrative records and by linkage to death registries based at China’s Disease Surveillance Points (23). Cause of death information was supplemented by medical record review, and, for the small proportion of deaths (<5%) without recent prior medical attention, was determined through verbal autopsy. Information on disease diagnoses resulting in, and during, hospitalizations was obtained through linkage, via unique national identification number, to established disease registries (for CHD, stroke, diabetes, and cancer) and to the national health insurance system (>98% coverage). All events were ICD-10 coded by trained staff blinded to baseline information. Participants were followed up until 1 January 2017 (mean 9 years).

Diagnostic Criteria for Diabetes

Prevalent type 2 diabetes was defined as self-reported physician-diagnosed diabetes or screen-detected diabetes (no prior diabetes diagnosis with a plasma glucose concentration ≥7.8 mmol/L and fasting time ≥8 h, a plasma glucose concentration ≥11.1 mmol/L and fasting time <8 h, or a fasting plasma glucose concentration ≥7.0 mmol/L) at baseline (excluding individuals with possible type 1 diabetes, defined as diagnosis at <30 years of age and use of insulin [n = 35]) (24). Incident type 2 diabetes included diabetes diagnoses (ICD-10 E11–E14) recorded during follow-up in the disease surveillance system or health insurance databases or as underlying or contributing to death on death certificates among individuals without prevalent type 2 diabetes.

Diagnostic Criteria for CVD

The primary disease end points for the current study were IS (ICD-10 I63, I69.3) (further classified into lacunar and nonlacunar), intracerebral hemorrhage (ICH) (ICD-10: I61, I69.1), and major coronary event (MCE) (nonfatal myocardial infarction [MI] [ICD-10 I21, I23] or fatal CHD [ICD-10 I20–I25]). Presence of carotid artery plaque was defined as focal thickening or protrusion from the artery wall into the lumen with cIMT >1.5 mm (22,25,26). Additional analyses examined fatal total stroke (ICD-10 I60, I61, I63, I64) and cardiovascular mortality (ICD-10 I00–I25, I27–I88, I95–I99). By 1 January 2017, 37,289 (7.3%) participants had died and 4,875 (<1%) were lost to follow-up (CKB database version 13.0).

Genotyping and Genetic Instruments

A 384–single nucleotide polymorphism (SNP) array was used to genotype 95,680 participants on the Illumina Golden Gate platform (SNP panel). This was custom designed in October 2012 and included SNPs associated with CVD, their risk factors, and related phenotypes. In addition, 96,330 participants were genotyped using a custom-designed 800K-SNP Affymetrix Axiom array and imputed to 1000 Genomes Phase III (GWAS panel). Case and noncase samples were genotyped in the same batches, and assays were conducted blind to case status. After application of quality control criteria, 159,528 participants remained for inclusion in the current analyses (Supplementary Fig. 1), including a subset of 24,519 participants genotyped using both arrays. Concordance for type 2 diabetes–related variants between the two arrays was high (r ≥ 0.9); where discordant, SNP panel genotypes were used. Info scores for type 2 diabetes–related variants not directly genotyped on the GWAS panel (n = 7) were high (>0.94), and estimated allele dosages were used for imputed SNPs. The total genotyped population included a population-based genotyped sample (n = 148,512) randomly selected from the total CKB cohort and included in all genetic analyses and 11,016 additional stroke or CHD cases included only in CVD outcome analyses.

A total of 59 type 2 diabetes risk variants identified in GWAS at the time of SNP panel design were included on both the SNP and GWAS panels, including 5, 15, and 36 originally reported among South Asians, East Asians, and Europeans, respectively. After exclusion of monomorphic variants (n = 1) and variants with genotype calling failure (n = 3), with parent-of-origin–specific effects (n = 1), demonstrating heterogeneity in associations with type 2 diabetes between European and East Asian populations if first reported in Europeans (n = 2), located on the X-chromosome (n = 1), with low genotyping rate (n = 1), or acting primarily through obesity (n = 2), 48 independent variants remained for inclusion in the type 2 diabetes genetic risk score (GRS-T2D48) (Supplementary Tables 1 and 2). Additional analyses examined GRS comprising type 2 diabetes–associated variants with specific pathophysiological mechanisms: β-cell dysfunction (GRS-BC) (24 SNPs), insulin resistance (GRS-IR) (6 SNPs), or unclassified (18 SNPs) (27). Sensitivity analyses included 1) analyses excluding SNPs associated with plasma lipids, stroke, and CHD (Supplementary Table 2); 2) analyses using internal weights calculated through 1000-fold cross-validation; 3) analyses restricted to the population-based genotyped sample; 4) analyses using an 86-SNP GRS (GRS-T2D86) (comprising 86 of 101 SNPs associated with type 2 diabetes in GWAS studies by December 2016 selected using criteria described for GRS-T2D48) among individuals genotyped with the GWAS panel; 5) summary statistics–based analyses using SNP–type 2 diabetes effect estimates derived from transethnic type 2 diabetes GWAS (28,29) and SNP-CVD effect estimates derived from CKB; and 6) analyses using only those SNPs associated with type 2 diabetes at genome-wide significance level.

Statistical Analysis

Observational Analyses

Prevalence and mean values of baseline characteristics were calculated by type 2 diabetes status, standardized by 5-year age-group, sex, and study area. Observational analyses excluded individuals with prior CVD (CHD, stroke, or transient ischemic attack; n = 23,129) or missing BMI (n = 2). After application of these exclusions, 489,549 participants (200,118 men, 289,431 women) remained. Cox proportional hazards models, with time since entry into the study as the underlying time scale, were used to estimate hazard ratios (HRs) of CVD for prevalent type 2 diabetes (n = 26,381) versus not, stratified by age at risk, sex, and study area and adjusted for education, smoking, alcohol consumption, systolic blood pressure (SBP), physical activity, and BMI.

Genetic Analyses

In genetic analyses, type 2 diabetes was defined as combined prevalent and incident type 2 diabetes. Missing genotypes were imputed by assigning the participant’s mean study area genotype. An unweighted GRS-T2D (GRS-T2D48) was developed by summing the number of type 2 diabetes risk–increasing alleles. A weighted GRS-T2D (GRS-T2D48w) was constructed by weighting SNPs by the natural logarithm of the per-allele odds ratio (OR) derived from transethnic type 2 diabetes GWAS (28,29), which represented the best performing external weights (27).

An inverse-variance weighted two-stage regression approach, with weighted or unweighted GRS-T2D as the instrumental variable, was used to assess the causal role of genetically predicted type 2 diabetes in CVD, subclinical atherosclerosis, and cardiometabolic risk factors in CKB (12). The associations between GRS-T2D and type 2 diabetes were examined using logistic regression adjusted for age, sex, and study area. The associations of the resulting predicted values with CVD and with binary CVD risk factors were examined using logistic regression, with adjustment for the same variables. Linear regression was applied to the second stage in testing of associations with continuous traits. Further analyses additionally adjusted for known CVD risk factors (SBP and adiposity). For comparison with observational estimates, genetic estimates of the odds of CVD associated with type 2 diabetes were calculated using the following formula: where β is the regression coefficient of the GRS association with CVD as a function of the GRS association with type 2 diabetes and A is the prevalence of type 2 diabetes in CKB (16). Heterogeneity between observational and genetic risk estimates was assessed using the Cochran Q test. Two-sample MR was used to estimate the associations of genetically predicted type 2 diabetes with risk of CVD in individuals of European ancestry using 1000 Genomes–based GWAS summary statistics for SNPs included in GRS-T2D86 obtained from DIAbetes Genetics Replication And Meta-analysis (DIAGRAM) (26,676 type 2 diabetes case and 132,532 control subjects) (30), CARDIoGRAMplusC4D (Coronary ARtery DIsease Genome wide Replication and Meta-analysis (CARDIoGRAM) plus The Coronary Artery Disease (C4D) Genetics) (60,801 CHD case and 123,504 control subjects) (31), and the International Stroke Genetics Consortium (ISGC) (12,389 IS case and 62,004 control subjects, with 1,545 ICH case and 1,481 control subjects) (3234). Inverse-variance weighted analysis was performed by regression of the SNP-CVD associations on the SNP–type 2 diabetes associations. Two-sample MR analysis was performed using the R TwoSampleMR package (35). The Cochran Q test was used to assess heterogeneity between associations in CKB and European populations. Additional sensitivity analyses, assessing the robustness of the two-sample MR results, included 1) MR-Egger (36), 2) weighted median MR (37), and 3) weighted-mode MR (38).

Analyses were conducted using SAS, version 9.4, and R, version 3.0.2.

Data and Resource Availability

The data that support the findings of this study are from the CKB study, whose authors may be contacted at ckbiobank@ndph.ox.ac.uk, but restrictions apply to the availability of these data. No applicable resources were generated or analyzed during the current study.

Characteristics of Genotyped Participants

Among the 148,512 participants (60,073 men, 88,439 women) included in the population-based genotyped sample, mean (SD) baseline age was 52.0 (10.7) years and 9.2% (n = 13,713) had type 2 diabetes (8,886 prevalent and 4,827 new-onset during follow-up) (Table 1). Participants with type 2 diabetes were older, more likely to be residents of urban areas, and had higher levels of adiposity (P < 0.0001) and blood pressure (P < 0.0001), and higher plasma total cholesterol (P = 0.001) and triglyceride (P < 0.0001) concentrations, than those without type 2 diabetes. Individuals with type 2 diabetes were more likely to have prior CVD and a family history of diabetes or CVD (P < 0.0001). The characteristics of genotyped participants were similar to those of the whole CKB cohort (Supplementary Table 4).

Observational Associations of Type 2 Diabetes With CVD and Subclinical Atherosclerosis

During follow-up, 36,407 IS (including 4,562 lacunar IS), 8,487 ICH, 6,868 MCE, and 11,917 cardiovascular deaths were recorded among 489,549 participants without prior CVD. Type 2 diabetes at baseline was associated with significantly higher risks of incident IS (HR 1.56 [95% CI 1.51, 1.61]) and its subtypes, specifically, nonlacunar (1.62 [95% CI 1.56, 1.67]) and lacunar (1.18 [95% CI 1.07, 1.30]) IS, ICH (1.38 [95% CI 1.28, 1.49]), MCE (2.06 [95% CI 1.92, 2.20]), and cardiovascular mortality (1.94 [95% CI 1.84, 2.05]) (Fig. 1). Type 2 diabetes at baseline was associated with higher risk of carotid artery plaque (OR 1.74 [95% CI 1.50, 2.02]) and greater cIMT (β = 0.015 [95% CI 0.009, 0.025]).

Genetic Associations of Individual Variants and GRS-T2D With Type 2 Diabetes

Of the 48 SNPs included in GRS-T2D48, 14 were associated with type 2 diabetes at a genome-wide significance level (P < 5 × 10−8) in CKB, 35 showed statistically significant associations with type 2 diabetes after correction for multiple testing (using a false discovery rate method [39] with a cutoff of 0.05), and 47 showed directionally consistent associations with type 2 diabetes. There was no evidence of heterogeneity between CKB and European GWAS studies (Supplementary Table 1). Both GRS-T2D48 (unweighted GRS-T2D) and GRS-T2D48w (externally weighted GRS-T2D) were robustly associated with risk of type 2 diabetes (P = 8.48 × 10−217 and P = 4.64 × 10−295, respectively). Externally weighted (28,29) GRS-IR (P = 4.85 × 10−12) and GRS-BC (P = 3.04 × 10−235) were highly significantly associated with type 2 diabetes risk. GRS-T2D48w explained 1.4% of the type 2 diabetes liability scale variance (using Nagelkerke’s pseudo R2 [40]) (F-statistic 212), indicating it was a strong instrument.

Genetic Associations of GRS-T2D With CVD Risk Factors

In CKB, GRS-T2D48w was weakly positively associated with SBP (β = 0.33 [95% CI 0.02, 0.64] per 1-unit higher log-odds type 2 diabetes, P = 0.04) (Table 2). It was inversely associated with general adiposity (BMI β = −0.29 [95% CI −0.34, −0.24], P = 4.55 × 10−28; percentage body fat β = −0.44 [95% CI −0.54, −0.33], P = 1.72 × 10−16) and central adiposity (waist circumference β = −0.58 [95% CI −0.73, −0.43], P = 6.59 × 10−15) but positively associated with waist circumference adjusted for BMI (β = 0.10 [95% CI 0.03, 0.17], P = 3.98 × 10−3) and waist-to-hip ratio adjusted for BMI (β = 0.24 [95% CI 0.16, 0.32], P = 1.20 × 10−9). GRS-T2D48w was not associated with other CVD risk factors.

Genetic Associations of Individual Variants and GRS-T2D With CVD and Subclinical Atherosclerosis

There was modest genetic correlation between type 2 diabetes and ASCVD in CKB (Supplementary Table 5). GRS-T2D48w was associated with greater cIMT (β = 0.011 [95% CI 0.006, 0.016] per 1-unit higher log-odds of type 2 diabetes, P = 1.97 × 10−5) and with the presence of carotid plaque (OR 1.17 [95% CI 1.05, 1.29], P = 3.74 × 10−3) (Table 2). Likewise, GRS-T2D48w was associated with an elevated risk of MCE (1.12 [95% CI 1.02, 1.23]; n = 5,081), IS (1.08 [95% CI 1.02, 1.14]; n = 17,097), nonlacunar IS (1.09 [95% CI 1.03–1.16]; n = 13,924) but not lacunar IS (1.03 [95% CI 0.91, 1.16]; n = 3,173), fatal total stroke (1.01 [95% CI 0.91, 1.11]; n = 4,319), and cardiovascular mortality (1.03 [95% CI 0.96, 1.11]; n = 9,006) (Supplementary Table 6). There was no statistically significant association of GRS-T2D48w with risk of ICH (1.01 [95% CI 0.94–1.10]; n = 6,973). Similar effect estimates were observed in genetic analyses adjusting for known CVD risk factors (BMI, waist circumference, body fat percentage, and SBP) (Supplementary Table 6). There was a strong, highly significant association of genetically predicted type 2 diabetes with diabetic microvascular diseases (retinopathy, nephropathy, neuropathy) (OR 2.80 [95% CI 2.33–3.36]; n = 1,140), included as a positive control. For comparison with observational associations, genetic estimates of the odds of CVD associated with type 2 diabetes, versus with no type 2 diabetes, were estimated (Fig. 1); there was no evidence of heterogeneity between observational and genetic effect estimates (P for heterogeneity ≥0.3).

Sensitivity analyses using unweighted or internally weighted GRS-T2D, excluding variants with documented associations with lipids or CVD, or limited to the population-based sample, did not materially alter the associations (Supplementary Table 7). Similar results were found using weighted and unweighted GRS-T2D86 and a summary statistics–based two-sample approach (Supplementary Tables 8 and 10). Estimates of the associations of insulin resistance–related variants (GRS-IR) were nonsignificantly stronger than those of β-cell dysfunction–related variants (GRS-BC) for any and nonlacunar IS, ICH, MCE, presence of carotid plaque, and cIMT (P > 0.08) (Supplementary Table 11).

Comparison of Associations of Genetically Predicted T2D With CVD Among Chinese and European Populations

In two-sample MR analyses, based on summary statistics from European ancestry population GWAS consortia (3034), each 1-unit higher log-odds of genetically predicted type 2 diabetes was associated with 12% (OR 1.12 [95% CI 1.03, 1.22], P = 7.90 × 10−3), 24% (1.24 [95% CI 1.07, 1.44], P = 4.53 × 10−3), and 9% (1.09 [95% CI 1.04, 1.15], P = 9.32 × 10−4) higher odds of IS, large artery stroke, and CHD, respectively (Table 3 and Supplementary Table 12). There was no significant association with small-vessel stroke (1.15 [95% CI 0.97, 1.35], P = 0.11) or ICH (1.14 [95% CI 0.93, 1.40], P = 0.20), but risk estimates were imprecise. Sensitivity analyses using weighted median, weighted mode, and MR-Egger approaches showed similar findings, as did analyses based on GRS-T2D86, and there was no evidence of unbalanced pleiotropy (P for pleiotropy from MR-Egger ≥0.31) (Supplementary Fig. 2). There were no statistically significant differences between Chinese and European population MR effect estimates for IS, CHD, or ICH (P for heterogeneity 0.28, 0.90, and 0.21, respectively) (Table 3).

This large study in a Chinese population provides new evidence for a robust association of genetic predisposition to type 2 diabetes with subclinical and clinical ASCVD, including coronary and cerebral manifestations. This is consistent with a causal role for type 2 diabetes in ASCVD, although shared heritability and unidentified pleiotropic effects of type 2 diabetes–associated variants may also contribute to the identified associations. There was no significant association between genetic predisposition to type 2 diabetes and risk of ICH, but statistical power to reliably confirm (or refute) any modest association was inadequate (Supplementary Table 14).

Large prospective observational studies (3,4), including CKB (2), have reported two- to fourfold higher risks of CHD in diabetes. Two previous MR analyses, using summary-level data from the same or related GWAS consortia, have provided evidence supporting a causal role of type 2 diabetes in CHD (13,16). The current study provides further strong evidence for a causal role of type 2 diabetes in CHD in a Chinese population. Few previous studies have examined the genetic association of type 2 diabetes with stroke (15,17), and only one has investigated IS (15). Based on European population GWAS summary statistics, and with inclusion of a number of stroke cases similar to that of the current study (18,476 and 17,097 events, respectively), there was an elevated risk of IS associated with genetically predicted type 2 diabetes (OR 1.12 [95% CI 1.07, 1.17] per 1-unit higher log-odds of type 2 diabetes) (15). This is broadly consistent with effect estimates reported in observational analyses (3,4), including in CKB, and with associations of genetically predicted type 2 diabetes in CKB.

The associations of genetic predisposition to type 2 diabetes with CHD and IS provide evidence in support of a causal role of pathways leading to type 2 diabetes, or type 2 diabetes itself, in ASCVD. In the current study, in general, we found nonsignificantly stronger associations of genetically predicted, compared with observed, type 2 diabetes, possibly reflecting the lifelong influence of genetic variants. Previous observational (57) and genetic (18,19) epidemiological studies examining the association of type 2 diabetes with subclinical atherosclerosis, defined in various ways, have reported conflicting findings. A study including ∼12,000 individuals from the U.S. found no significant effect of a 62-SNP GRS-T2D on various measures of subclinical atherosclerosis (19). In contrast, in a population-based study of ∼11,000 Chinese adults, a 34-SNP GRS-T2D was associated with 24% (95% CI 6, 47) higher risk of increased arterial stiffness (18). These conflicting findings may reflect differences in subclinical atherosclerosis assessment methods, inadequate statistical power, or differences between ancestries. CKB included approximately the same number of participants as previous studies combined and provides the strongest evidence to date of a causal role of type 2 diabetes for subclinical atherosclerosis.

Previous observational study findings on the association of type 2 diabetes with risk of IS subtypes have been conflicting (41,42). One European population genetic study reported 15% (95% CI 4, 25) higher odds of imaging-confirmed lacunar IS (n = 2,191) associated with genetically predicted type 2 diabetes (17), while another, using summary estimates from the same type 2 diabetes and, for a proportion of events, stroke GWAS consortia, found 21% (95% CI 10, 33) higher odds of imaging-confirmed small-vessel (equivalent to lacunar [43]) IS (15). We found a weaker, nonsignificant, association of genetically predicted type 2 diabetes with lacunar IS (1.03). Widespread use of computed tomography and MRI in China, often resulting in detection of lacunar infarcts without apparent neurological deficit (44), may partly explain this difference. Moreover, although 18.6% of IS included in the present genetic analyses were lacunar IS, there was still limited statistical power (0.08) to detect a modest association (Supplementary Table 14). The estimated association of genetically predicted type 2 diabetes with nonlacunar IS in CKB (1.09) lies between previous estimates of the association of genetically predicted type 2 diabetes with large-vessel (atherosclerotic) IS (1.28 [95% CI 1.16, 1.40]) and cardioembolic IS (1.06 [95% CI 0.97, 1.15]) (15), the two major nonlacunar IS subtypes.

Large prospective observational studies have reported 30–60% higher risks of ICH in diabetes (3,4). With more ICH cases than previous studies combined, CKB provides the most robust observational evidence to date, showing 40% higher risk of ICH in type 2 diabetes. However, previous MR analyses, including ∼2,200 ICH cases from multiple GWAS, reported no significant association with genetically predicted type 2 diabetes (OR 1.07 [95% CI 0.95, 1.20] per 1-unit higher log-odds of type 2 diabetes) (17). Likewise, the current study found no clear evidence of a causal association between type 2 diabetes and ICH, suggesting the observational association might be due chiefly to residual confounding. However, there was limited power to detect an association (0.17), and these data do not completely rule out a modest causal effect.

We identified inverse associations of genetic predisposition to type 2 diabetes with measures of general and central adiposity. Several type 2 diabetes risk–increasing variants have been associated with lower adiposity (28,45,46), potentially reflecting the associations of insulin resistance with higher risk of type 2 diabetes and propensity to visceral, rather than peripheral, adiposity (47). However, caution is required in interpreting the association with some adiposity measures (e.g., BMI-adjusted waist-to-hip ratio), given the risk of collider bias (12). Additional adjustment for adiposity (and SBP) and exclusion of SNPs with known associations with lipids and CVD did not substantially alter the risk estimates, suggesting these factors do not explain the presented findings. However, use of measured phenotypes, rather than genetic associations, may underestimate their influence.

The strengths of the current study are several fold. In contrast with most previous studies (13,1517), the present analyses were based on individual participant data from a single population. Moreover, this is the first study to examine the association of genetically predicted type 2 diabetes with CVD in a non–European ancestry population. CKB includes large numbers of well-characterized stroke subtypes (∼90% of stroke cases have been confirmed by neuroimaging), and stroke adjudication shows high accuracy of diagnoses (∼90% verified through medical record review). Utilization of multiple data sources and outcome adjudication for phenotyping of incident CVD reduced potential limitations inherent in the use of ICD-10 coding (e.g., inadequate granularity of data, inconsistent or incomplete coding), while the passive follow-up approach reduced potential reporting, nonresponse, and loss to follow-up biases. Furthermore, as well as enabling assessment of pleiotropic effects of type 2 diabetes–associated variants, extensive phenotyping of CKB participants provided mechanistic insights through investigation of subclinical atherosclerosis. Finally, the large sample size and relatively high stroke incidence facilitated precise estimates of genetic associations of type 2 diabetes with major CVD types. However, the study also has limitations, including inadequate statistical power to reliably assess the associations of genetically predicted type 2 diabetes with ICH and lacunar IS (Supplementary Table 14). Furthermore, the GRS-T2D may have unidentified pleiotropic effects. However, inclusion of SNPs acting through different pathways should limit the impact of this on causal inferences. Type 2 diabetes genetic risk prediction could be impaired by limited portability of GRS-T2D across diverse populations (48), although this is less likely, since the GRS-T2D was based on GWAS-identified SNPs, rather than polygenetic score–based, and single variant effect sizes were consistent across Chinese and European populations (27). Finally, a proportion of diabetes cases will have remained undiagnosed in the study population due to the methods used to identify undiagnosed diabetes at baseline and incident diabetes; this would likely underestimate type 2 diabetes–associated CVD risks in the current study.

In summary, this large study of Chinese adults provides new evidence supporting a causal role for type 2 diabetes in clinical and subclinical ASCVD, consistent with observational epidemiological findings, and highlighting the importance of prevention and appropriate management of type 2 diabetes for reducing the burden of ASCVD. Genetic predisposition to type 2 diabetes was not strongly associated with risk of lacunar IS and ICH, in contrast to observational findings. However, further larger studies examining the associations of genetically predicted type 2 diabetes with stroke subtypes are needed to clarify these findings.

Acknowledgments. The chief acknowledgment is to the participants, the project staff, and the Chinese Center for Disease Control and Prevention (China CDC) and its regional offices for access to death and disease registries. The Chinese national health insurance scheme provides electronic linkage to all hospital admission data.

Funding. The baseline survey and the first resurvey were supported by a research grant from the Kadoorie Charitable Foundation in Hong Kong. The long-term continuation of the project is supported by program grants from the U.K. Wellcome Trust (088158/Z/09/Z, 104085/Z/14/Z, 212946/Z/18/Z), the Medical Research Council (Newton Fund MC_PC_13049, MC_PC_14135), the Chinese Ministry of Science and Technology (2011BAI09B01, 2012-14), the Chinese National Natural Science Foundation (81390540, 81390541, 81390544, 2014-18), and the National Key Research and Development Program of China (2016YFC0900500, 2016YFC0900501, 2016YFC0900504, 2016YFC1303904, 2016-2021). The British Heart Foundation, Medical Research Council, and Cancer Research UK provide core funding to the Oxford CTSU. M.I.M. is a Wellcome Senior Investigator and a National Institute for Health Research (NIHR) Senior Investigator. M.I.M. acknowledges support from Wellcome (090532, 098381, 203141), National Institutes of Health (U01-DK105535), and the NIHR (NF-SI-0617-10090). F.B. acknowledges support from the Oxford BHF Centre of Research Excellence. M.V.H. works in a unit that receives funding from the U.K. Medical Research Council and is supported by a British Heart Foundation Intermediate Clinical Research Fellowship (FS/18/23/33512) and the NIHR Oxford Biomedical Research Centre. This work was supported by the NIHR Oxford Biomedical Research Centre.

The views expressed in this article are those of the authors and not necessarily those of the U.K. National Health Service (NHS), the NIHR, or the Department of Health.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. All authors were involved in study design, conduct, long-term follow-up, analysis of data, interpretation of findings, or writing of the manuscript. W.G., F.B., L.L., and Z.C. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

1.
International Diabetes Federation
.
Diabetes Atlas
, 8th ed.
Brussels
,
Belgium
,
2017
2.
Bragg
F
,
Li
L
,
Yang
L
, et al.;
China Kadoorie Biobank (CKB) Collaborative Group
.
Risks and population burden of cardiovascular diseases associated with diabetes in China: a prospective study of 0.5 million adults
.
PLoS Med
2016
;
13
:
e1002026
3.
Sarwar
N
,
Gao
P
,
Seshasai
SR
, et al.;
Emerging Risk Factors Collaboration
.
Diabetes mellitus, fasting blood glucose concentration, and risk of vascular disease: a collaborative meta-analysis of 102 prospective studies
.
Lancet
2010
;
375
:
2215
2222
4.
Shah
AD
,
Langenberg
C
,
Rapsomaniki
E
, et al
.
Type 2 diabetes and incidence of cardiovascular diseases: a cohort study in 1·9 million people
.
Lancet Diabetes Endocrinol
2015
;
3
:
105
113
5.
Wagenknecht
LE
,
Zaccaro
D
,
Espeland
MA
,
Karter
AJ
,
O’Leary
DH
,
Haffner
SM
.
Diabetes and progression of carotid atherosclerosis: the insulin resistance atherosclerosis study
.
Arterioscler Thromb Vasc Biol
2003
;
23
:
1035
1041
6.
Mathiesen
EB
,
Joakimsen
O
,
Bønaa
KH
.
Prevalence of and risk factors associated with carotid artery stenosis: the Tromsø Study
.
Cerebrovasc Dis
2001
;
12
:
44
51
7.
Fine-Edelstein
JS
,
Wolf
PA
,
O’Leary
DH
, et al
.
Precursors of extracranial carotid atherosclerosis in the Framingham Study
.
Neurology
1994
;
44
:
1046
1050
8.
Boulanger
M
,
Poon
MTC
,
Wild
SH
,
Al-Shahi Salman
R
.
Association between diabetes mellitus and the occurrence and outcome of intracerebral hemorrhage
.
Neurology
2016
;
87
:
870
878
9.
Hemmingsen
B
,
Lund
SS
,
Gluud
C
, et al
.
Intensive glycaemic control for patients with type 2 diabetes: systematic review with meta-analysis and trial sequential analysis of randomised clinical trials
.
BMJ
2011
;
343
:
d6898
10.
Hopper
I
,
Billah
B
,
Skiba
M
,
Krum
H
.
Prevention of diabetes and reduction in major cardiovascular events in studies of subjects with prediabetes: meta-analysis of randomised controlled clinical trials
.
Eur J Cardiovasc Prev Rehabil
2011
;
18
:
813
823
11.
Li
G
,
Zhang
P
,
Wang
J
, et al
.
Cardiovascular mortality, all-cause mortality, and diabetes incidence after lifestyle intervention for people with impaired glucose tolerance in the Da Qing Diabetes Prevention Study: a 23-year follow-up study
.
Lancet Diabetes Endocrinol
2014
;
2
:
474
480
12.
Burgess
S
,
Thompson
S
.
Mendelian Randomization. Methods for Using Genetic Variants in Causal Estimation
.
Boca Raton, FL
,
CRC Press
,
2015
13.
Ahmad
OS
,
Morris
JA
,
Mujammami
M
, et al
.
A Mendelian randomization study of the effect of type-2 diabetes on coronary heart disease
.
Nat Commun
2015
;
6
:
7060
14.
Jansen
H
,
Loley
C
,
Lieb
W
, et al.;
CARDIoGRAM Consortium
.
Genetic variants primarily associated with type 2 diabetes are related to coronary artery disease risk
.
Atherosclerosis
2015
;
241
:
419
426
15.
Larsson
SC
,
Scott
RA
,
Traylor
M
, et al.;
METASTROKE Collaboration and NINDS Stroke Genetics Network (SiGN)
.
Type 2 diabetes, glucose, insulin, BMI, and ischemic stroke subtypes: Mendelian randomization study
.
Neurology
2017
;
89
:
454
460
16.
Ross
S
,
Gerstein
HC
,
Eikelboom
J
,
Anand
SS
,
Yusuf
S
,
Paré
G
.
Mendelian randomization analysis supports the causal role of dysglycaemia and diabetes in the risk of coronary artery disease
.
Eur Heart J
2015
;
36
:
1454
1462
17.
Liu
J
,
Rutten-Jacobs
L
,
Liu
M
,
Markus
HS
,
Traylor
M
.
Causal impact of type 2 diabetes mellitus on cerebral small vessel disease: a Mendelian randomization analysis
.
Stroke
2018
;
49
:
1325
1331
18.
Xu
M
,
Huang
Y
,
Xie
L
, et al
.
Diabetes and risk of arterial stiffness: a Mendelian randomization analysis
.
Diabetes
2016
;
65
:
1731
1740
19.
Dauriz
M
,
Porneala
BC
,
Guo
X
, et al
.
Association of a 62 variants type 2 diabetes genetic risk score with markers of subclinical atherosclerosis: a transethnic, multicenter study
.
Circ Cardiovasc Genet
2015
;
8
:
507
515
20.
Liu
M
,
Wu
B
,
Wang
WZ
,
Lee
LM
,
Zhang
SH
,
Kong
LZ
.
Stroke in China: epidemiology, prevention, and management strategies
.
Lancet Neurol
2007
;
6
:
456
464
21.
Chen
Z
,
Chen
J
,
Collins
R
, et al.;
China Kadoorie Biobank (CKB) Collaborative Group
.
China Kadoorie Biobank of 0.5 million people: survey methods, baseline characteristics and long-term follow-up
.
Int J Epidemiol
2011
;
40
:
1652
1666
22.
Clarke
R
,
Du
H
,
Kurmi
O
, et al.;
China Kadoorie Biobank Collaborative Group
.
Burden of carotid artery atherosclerosis in Chinese adults: implications for future risk of cardiovascular diseases
.
Eur J Prev Cardiol
2017
;
24
:
647
656
23.
Yang
G
,
Murray
C
,
Zhang
Z
.
Exploring Adult Mortality in China: Levels, Patterns, and Causes
.
Beijing
,
Hua Xia Press
,
1991
24.
Bragg
F
,
Li
L
,
Smith
M
, et al.;
China Kadoorie Biobank Collaborative Group
.
Associations of blood glucose and prevalent diabetes with risk of cardiovascular disease in 500 000 adult Chinese: the China Kadoorie Biobank
.
Diabet Med
2014
;
31
:
540
551
25.
Greenland
P
,
Alpert
JS
,
Beller
GA
, et al.;
American College of Cardiology Foundation/American Heart Association Task Force on Practice Guidelines
.
2010 ACCF/AHA guideline for assessment of cardiovascular risk in asymptomatic adults: executive summary: a report of the American College of Cardiology Foundation/American Heart Association Task Force on Practice Guidelines
.
Circulation
2010
;
122
:
2748
2764
26.
Stein
JH
,
Korcarz
CE
,
Hurst
RT
, et al.;
American Society of Echocardiography Carotid Intima-Media Thickness Task Force
.
Use of carotid ultrasound to identify subclinical vascular disease and evaluate cardiovascular disease risk: a consensus statement from the American Society of Echocardiography Carotid Intima-Media Thickness Task Force
.
J Am Soc Echocardiogr
2008
;
21
:
93
111
27.
Gan
W
,
Walters
RG
,
Holmes
MV
, et al.;
China Kadoorie Biobank Collaborative Group
.
Evaluation of type 2 diabetes genetic risk variants in Chinese adults: findings from 93,000 individuals from the China Kadoorie Biobank
.
Diabetologia
2016
;
59
:
1446
1457
28.
Mahajan
A
,
Go
MJ
,
Zhang
W
, et al.;
DIAbetes Genetics Replication And Meta-analysis (DIAGRAM) Consortium
;
Asian Genetic Epidemiology Network Type 2 Diabetes (AGEN-T2D) Consortium
;
South Asian Type 2 Diabetes (SAT2D) Consortium
;
Mexican American Type 2 Diabetes (MAT2D) Consortium
;
Type 2 Diabetes Genetic Exploration by Nex-generation sequencing in muylti-Ethnic Samples (T2D-GENES) Consortium
.
Genome-wide trans-ancestry meta-analysis provides insight into the genetic architecture of type 2 diabetes susceptibility
.
Nat Genet
2014
;
46
:
234
244
29.
Fuchsberger
C
,
Flannick
J
,
Teslovich
TM
, et al
.
The genetic architecture of type 2 diabetes
.
Nature
2016
;
536
:
41
47
30.
Scott
RA
,
Scott
LJ
,
Mägi
R
, et al.;
DIAbetes Genetics Replication And Meta-analysis (DIAGRAM) Consortium
.
An expanded genome-wide association study of type 2 diabetes in Europeans
.
Diabetes
2017
;
66
:
2888
2902
31.
Nikpay
M
,
Goel
A
,
Won
HH
, et al
.
A comprehensive 1,000 Genomes-based genome-wide association meta-analysis of coronary artery disease
.
Nat Genet
2015
;
47
:
1121
1130
32.
Traylor
M
,
Farrall
M
,
Holliday
EG
, et al.;
Australian Stroke Genetics Collaborative, Wellcome Trust Case Control Consortium 2 (WTCCC2)
;
International Stroke Genetics Consortium
.
Genetic risk factors for ischaemic stroke and its subtypes (the METASTROKE collaboration): a meta-analysis of genome-wide association studies
.
Lancet Neurol
2012
;
11
:
951
962
33.
Malik
R
,
Traylor
M
,
Pulit
SL
, et al.;
ISGC Analysis Group
;
METASTROKE collaboration
;
Wellcome Trust Case Control Consortium 2 (WTCCC2)
;
NINDS Stroke Genetics Network (SiGN)
.
Low-frequency and common genetic variation in ischemic stroke: The METASTROKE collaboration
.
Neurology
2016
;
86
:
1217
1226
34.
Woo
D
,
Falcone
GJ
,
Devan
WJ
, et al.;
International Stroke Genetics Consortium
.
Meta-analysis of genome-wide association studies identifies 1q22 as a susceptibility locus for intracerebral hemorrhage
.
Am J Hum Genet
2014
;
94
:
511
521
35.
Hemani
G
,
Zheng
J
,
Elsworth
B
, et al
.
The MR-Base platform supports systematic causal inference across the human phenome
.
eLife
2018
;
7
:
e34408
36.
Bowden
J
,
Davey Smith
G
,
Burgess
S
.
Mendelian randomization with invalid instruments: effect estimation and bias detection through Egger regression
.
Int J Epidemiol
2015
;
44
:
512
525
37.
Bowden
J
,
Davey Smith
G
,
Haycock
PC
,
Burgess
S
.
Consistent estimation in Mendelian randomization with some invalid instruments using a weighted median estimator
.
Genet Epidemiol
2016
;
40
:
304
314
38.
Hartwig
FP
,
Davey Smith
G
,
Bowden
J
.
Robust inference in summary data Mendelian randomization via the zero modal pleiotropy assumption
.
Int J Epidemiol
2017
;
46
:
1985
1998
39.
Holm
S
.
A simple sequentially rejective multiple test procedure
.
Scand J Stat
1979
;
6
:
65
70
40.
Nagelkerke
N
.
A note on a general definition of the coefficient of determination
.
Biometrika
1991
;
78
:
691
692
41.
Bezerra
DC
,
Sharrett
AR
,
Matsushita
K
, et al
.
Risk factors for lacune subtypes in the Atherosclerosis Risk in Communities (ARIC) Study
.
Neurology
2012
;
78
:
102
108
42.
van Harten
B
,
de Leeuw
FE
,
Weinstein
HC
,
Scheltens
P
,
Biessels
GJ
.
Brain imaging in patients with diabetes: a systematic review
.
Diabetes Care
2006
;
29
:
2539
2548
43.
Adams
HP
 Jr
,
Bendixen
BH
,
Kappelle
LJ
, et al
.
Classification of subtype of acute ischemic stroke. Definitions for use in a multicenter clinical trial. TOAST. Trial of Org 10172 in Acute Stroke Treatment
.
Stroke
1993
;
24
:
35
41
44.
Wu
B
,
Lin
S
,
Hao
Z
, et al
.
Proportion, risk factors and outcome of lacunar infarction: a hospital-based study in a Chinese population
.
Cerebrovasc Dis
2010
;
29
:
181
187
45.
Lu
Y
,
Day
FR
,
Gustafsson
S
, et al
.
New loci for body fat percentage reveal link between adiposity and cardiometabolic disease risk
.
Nat Commun
2016
;
7
:
10495
46.
Kilpeläinen
TO
,
Zillikens
MC
,
Stančákova
A
, et al
.
Genetic variation near IRS1 associates with reduced adiposity and an impaired metabolic profile
.
Nat Genet
2011
;
43
:
753
760
47.
Lotta
LA
,
Gulati
P
,
Day
FR
, et al.;
EPIC-InterAct Consortium
;
Cambridge FPLD1 Consortium
.
Integrative genomic analysis implicates limited peripheral adipose storage capacity in the pathogenesis of human insulin resistance
.
Nat Genet
2017
;
49
:
17
26
48.
Martin
AR
,
Gignoux
CR
,
Walters
RK
, et al
.
Human demographic history impacts genetic risk prediction across diverse populations
.
Am J Hum Genet
2017
;
100
:
635
649
Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. More information is available at http://www.diabetesjournals.org/content/license.

Supplementary data