Introduction

Type 2 diabetes mellitus (T2DM) and coronary artery disease (CAD) are the leading causes of death worldwide.[1] The coexistence of these two common cardiometabolic conditions has long been documented. A large meta-analysis of prospective studies involving 698,782 individuals has confirmed a doubled risk of CAD among T2DM patients compared to those without T2DM.[2] Post-analysis of genome-wide association studies (GWASs) has identified a strong overall T2DM–CAD genetic correlation (rg = 0.40), indicating a considerable proportion of shared phenotypic variance due to genetics.[3,4] Multiple pleiotropic loci (e.g., CDKN2B-AS1,[5,6]TCF7L2,[6–8]HNF1A,[6,7]ALDH2,[5]FTO,[6]CCDC92,[7]APOE,[7] and MASP1[8]) are also reported to independently affect both diseases. Mendelian randomization (MR), which uses genetic variants as instrumental variables (IVs) to estimate exposure–outcome causal relationships, has consistently identified a causal association (effect sizes ranging from 1.09 to 1.63) between genetically determined T2DM and CAD[3,7,9–14] (a detailed summarization is shown in Supplementary Figure 1, https://links.lww.com/CM9/B771).

Although knowledge gained from genetic analyses has advanced our understanding of a shared etiology between T2DM and CAD, a few major gaps remain. First, previous studies used GWAS data from populations of admixed ancestry[6,7,14] or small sample sizes,[5–8,14] substantially limiting the statistical power. Second, previous MRs did not assess reverse causation,[3,7,9–14] an effect of genetically determined CAD on subsequent T2DM, or the pathway of the causation. A longitudinal study demonstrated that increased severity of coronary atherosclerosis accurately predicted the future development of T2DM,[15] suggesting a complex bidirectional relationship to which integrated care might be targeted. Third, overweight status and obesity, as reflected by the body mass index (BMI), contribute to both conditions and satisfactorily predict the development of CAD in T2DM patients.[16] However, this important confounder was not considered in previous genetic studies.[3–8,14] Finally, limited functional annotation was performed to relate statistical evidence with biological pathways.[3] To our knowledge, no large-scale observational or genetic analysis has been conducted to systematically investigate the extent and nature of the shared etiology underlying T2DM and CAD, taking into account the important confounder, obesity.

Therefore, the current study aimed to comprehensively interrogate the T2DM–CAD relationship, leveraging the most comprehensive observational and genetic data. We first investigated the bidirectional phenotypic association using more than 470,000 participants available in the United Kingdom Biobank (UKB). We next performed a genome-wide cross-trait analysis to quantify the overall and local genetic correlations, identify pleiotropic loci, detect tissue enrichment, and make causal inferences. The overarching goal of our study was to provide genetic insight into the observed associations underlying T2DM and CAD, which may facilitate progress in prevention strategies for common cardiometabolic diseases. The overall study design is shown in Figure 1.

Figure 1:

Flowchart of overall study design. BMI: Body mass index; CAD: Coronary artery disease; T2DM: Type 2 diabetes mellitus; T2DMadjBMI: Type 2 diabetes mellitus adjusted for BMI.

Methods Ethical approval

The participants from the UKB provided written informed consent, and ethical approval was granted by the National Health Service North West Multi-Centre Research Ethics Committee (No. 11/NW/0382). All GWAS summary statistics are publicly available, and the corresponding studies obtained proper institutional review board approval and participant consent.

Data sources UKB data

The UKB constitutes a large prospective cohort study involving approximately 500,000 individuals aged 40–69 years at baseline, recruited from England, Wales, and Scotland between the years 2006 and 2010.[17] All participants provided written informed consent. We only considered 472,050 participants of white descent. A diagnosis of T2DM was defined according to the International Classification of Diseases, Tenth Revision (ICD-10) code E11, and a diagnosis of CAD was made according to ICD-9 codes 410–414 and ICD-10 codes I20–I25. Participant selection for the current study is shown in detail in the Supplementary methods, https://links.lww.com/CM9/B771.

GWAS summary statistics

The largest GWAS of T2DM in people of European ancestry was conducted by the DIAbetes Genetics Replication And Meta-analysis (DIAGRAM) consortium, aggregating 32 participating studies, totaling 74,124 cases and 824,006 controls.[18] A fixed-effect meta-analysis was conducted to combine the effect sizes for each variant across all studies. Independent T2DM-associated single-nucleotide polymorphisms (SNPs) that reached genome-wide significance (P <5 × 10-8) were identified. Conditionally independent variants that reached locus-wide significance (P <10-5) were considered index SNPs. A GWAS of T2DM was performed for 50,409 T2DM cases and 523,897 controls, and the effect sizes of variants were further adjusted for BMI (T2DMadjBMI).

A total of 386 independent T2DM-associated SNPs and 143 independent T2DMadjBMI-associated SNPs were identified and used as IVs. We extracted the effect size and relevant information on these IVs [Supplementary Tables 1 and 2, https://links.lww.com/CM9/B771] and downloaded the full set of GWAS summary statistics.

An additional 338 independent SNPs—identified in the hitherto-largest multiancestry GWAS of T2DM, totaling 180,834 cases and 1,159,055 controls (48.9% non-European descent)—were obtained as IVs to increase the generalizability and robustness of the findings.[19]

The hitherto-largest GWAS of CAD was meta-analyzing data from 11 studies, totaling 181,522 cases and 984,168 controls of predominantly European ancestry (>95%). This meta-GWAS identified 241 significantly independent SNPs (P <5 × 10-8).[20] We extracted the effect size and relevant information on these SNPs [Supplementary Tables 3 and 4, https://links.lww.com/CM9/B771] and downloaded the full set of GWAS summary statistics. A table detailing the relevant information of all included GWASs is provided [Supplementary Table 5, https://links.lww.com/CM9/B771].

Statistical analysis Observational analysis

The baseline characteristics of the UKB participants are presented as the mean ± standard deviation or median (interquartile range) for continuous variables and frequencies for categorical variables.

We first assessed the association between T2DM and the risk of subsequent CAD. Person-years at risk for the T2DM-free category (unexposed) were calculated from baseline until T2DM diagnosis, CAD diagnosis, death, loss to follow-up, or the end of follow-up, whichever came first. Person-years at risk for the T2DM category (exposed) were calculated from baseline or T2DM diagnosis during follow-up until CAD diagnosis, death, loss to follow-up, or the end of follow-up, whichever came first. We constructed a Cox proportional-hazards regression model with exposure modeled as a time-dependent variable. We used three sets of adjustments. Estimates in Model 1 (basic model) were adjusted only for age, sex, assessment center, and the first 10 principal components. Estimates in Model 2 (BMI model) were further adjusted for BMI. Estimates in Model 3 (full model) were based on Model 2 and additionally adjusted for income, Townsend deprivation index values, smoking, drinking, physical activity, hypertension, and dyslipidemia. We excluded from sensitivity analysis those participants with less than a year of follow-up or a diagnosis of CAD within a year after developing T2DM. We repeated all these analyses to assess the association between CAD and the risk of subsequent T2DM. All statistical analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC, USA). A two-sided P value of less than 0.05 was considered statistically significant.

Global and local genetic correlation analyses

Genetic correlation describes the average sharing of genetic effect between two traits that is free from environmental confounders. We first calculated the global genetic correlation using linkage-disequilibrium (LD) score regression software (https://github.com/bulik/ldsc). This method builds on the fact that the effect-size estimate for a given SNP in a GWAS aggregates the effects of all SNPs in LD with that SNP. The final genetic correlation estimates ranged from –1 to 1, with –1 indicating a perfect negative correlation and 1 indicating a perfect positive correlation. A Bonferroni-corrected P-threshold (P <0.025 = 0.05/2) was used to define statistical significance.

The overall genetic correlation quantifies the aggregated contribution of genome-wide variants. Even if two traits show negligible genome-wide genetic correlation, specific genomic regions might contribute to both traits. We next estimated local genetic correlation using ρ-HESS software (https://huwenboshi.github.io/hess/local_rhog/), an algorithm that partitions the whole genome into 1703 LD-independent regions (average length: 1.5 Mb) and quantifies the genetic correlations restricted to these regions. A Bonferroni-corrected P-threshold (P <2.94 × 10-5 = 0.05/1703) was used to define statistical significance.

Cross-trait meta-analysis

Significant genetic correlation reflects either horizontal pleiotropy (pleiotropy) or vertical pleiotropy (causality). We next performed a cross-trait meta-analysis to identify pleiotropic loci affecting both traits using the Cross-Phenotype Association software (CPASSOC, http://hal.case.edu/~xxz10/zhu-web/). CPASSOC integrates the GWAS summary statistics from multiple correlated traits to detect variants associated with at least one trait, controlling for population structure or cryptic relatedness. We further applied the PLINK (https://www.cog-genomics.org/plink/1.9/) clumping function (parameters:clump-p1 5e-8, clump-p2 1e-5, clump-r2 0.2,clump-kb 500) to obtain the independent shared SNPs fulfilling PCPASSOC <5 × 10-8 and PT2DM or CAD <1 × 10-5.

An index SNP was considered a novel SNP only if all of the following criteria were satisfied: (1) the SNP did not reach single-trait genome-wide significance (5 × 10-8 <PT2DM or CAD <1 × 10-5); (2) the SNP reached cross-trait genome-wide significance (PCPASSOC <5 × 10-8); and (3) the SNP was not in LD (r2 <0.05) with any of those previously reported genome-wide significant SNPs of single traits.

We also searched the GWAS Catalog (https://www.ebi.ac.uk/gwas/) and PhenoScanner (http://www.phenoscanner.medschl.cam.ac.uk/) to determine whether an association between the SNPs identified by CPASSOC and phenotypes other than T2DM and CAD (P <5 × 10-8) was already reported.

Functional annotation and tissue enrichment analysis

To gain biological insight into the identified shared variants, we mapped SNPs to genes using Ensembl Variant Effect Predictor (https://grch37.ensembl.org/info/docs/tools/vep/index.html). We conducted tissue enrichment analysis of these genes based on 54 tissue types available in Genotype-Tissue Expression (GTEx) project version 8, using the functional mapping and annotation (FUMA) platform of GWAS software (https://fuma.ctglab.nl/). FUMA produced a set of differentially expressed genes for each of the 54 tissue types. Our identified shared genes were compared against those differentially expressed genes using hypergeometric tests to determine whether they were enriched in a specific tissue. A Bonferroni-corrected P-threshold (P <9.26 × 10-4 = 0.05/54) was used to define statistical significance.

Univariable MR analysis

We finally performed a bidirectional two-sample MR using TwoSampleMR (https://mrcieu.github.io/TwoSampleMR/) software to make causal inferences. We used the inverse-variance-weighted (IVW) method[21] as our primary approach, which assumed that all IVs were valid and provided the greatest statistical power. We next conducted several sensitivity analyses, including MR-Egger regression[22] and the weighted-median method, to test for the robustness of the results.[23] We also repeated the IVW method, excluding palindromic IVs (i.e., A/T or G/C alleles) or pleiotropic IVs (SNPs associated with potential confounders), guaranteeing the "exclusion restriction" assumption. We further performed latent heritable confounder MR (LHC-MR) to simultaneously estimate the bidirectional causal effects between two traits while accounting for the effect of the potential heritable confounder, i.e., shared genetic basis, guaranteeing the "exchangeability" assumption.[24] A Bonferroni-corrected P-threshold (P <0.025 = 0.05/2) was used to define statistical significance.

We estimated phenotypic variance explained by IVs (R2)[25] and calculated the F-statistic to test for the strength of the IVs, validating the "relevance" assumption.[26] We also calculated MR statistical power (https://shiny.cnsgenomics.com/mRnd/).

Multivariable MR and mediation analysis

The potential confounding or mediating effects of BMI,[27] systolic blood pressure (SBP),[28] intake of 3-hydroxy-3-methylglutaryl coenzyme A (HMG CoA) reductase inhibitors (C10AA),[29] three known important confounders or mediators of the T2DM–CAD association,[30,31] were evaluated. We first performed multivariable MR to estimate the independent causal effects between T2DM and CAD after adjusting for BMI. We next performed a multivariable MR to assess the mediation effect of SBP and intake of C10AA utilizing the "difference in coefficients" method.[32]

Results Phenotypic association

The baseline characteristics of the UKB participants are presented in Supplementary Tables 6 and 7, https://links.lww.com/CM9/B771.

In the analysis of the risk of incident CAD associated with T2DM, the participants were followed for 5,303,441 person-years (11.3 ± 3.1 years), during which 3323 T2DM patients and 26,380 T2DM-free individuals developed CAD [Table 1]. After adjusting for age, sex, assessment center, and principal components, T2DM patients showed a significantly increased risk of CAD (hazard ratio [HR] = 2.70, 95% confidence interval [CI]: 2.60–2.81). The effect was attenuated to some extent (17%) by further adjustment for BMI but remained statistically significant (HR = 2.28, 95% CI: 2.19–2.37). The effect stabilized to 2.12 (95% CI: 2.01–2.24) in the full model. Similar results (HR = 1.66, 95% CI: 1.56–1.76) were observed in the sensitivity analysis [Supplementary Table 8, https://links.lww.com/CM9/B771].

Table 1 - Observational associations between T2DM and CAD. Exposure status during follow-up Cases/person-years Primary analysis Sensitivity analysis Basic model Basic model + BMI Full model Full model T2DM No 26,380/5,135,683 1.00 (ref) 1.00 (ref) 1.00 (ref) 1.00 (ref) Yes 3323/167,759 2.70 (2.60–2.81) 2.28 (2.19–2.37) 2.12 (2.01–2.24) 1.66 (1.56–1.76) CAD No 17,608/5,050,271 1.00 (ref) 1.00 (ref) 1.00 (ref) 1.00 (ref) Yes 3810/318,052 2.44 (2.35–2.53) 1.99 (1.92–2.06) 1.72 (1.63–1.81) 1.48 (1.40–1.56)

HRs are provided with 95% CIs. Basic model: adjusted for age, sex, assessment center, and the first 10 principal components. Full model: adjusted for sex, age, assessment center, income, Townsend deprivation index, smoking, drinking, physical activity (IPAQ), BMI, hypertension, dyslipidemia, and the first 10 principal components. BMI: Body mass index; CAD: Coronary artery disease; CI: Confidence interval; IPAQ: International Physical Activity Questionnaire; HRs: Hazard ratios; ref: Reference; ref: Reference; T2DM: Type 2 diabetes mellitus.

In the analysis of the risk of incident T2DM associated with CAD, the participants were followed for 5,368,323 person-years (11.3 ± 3.1 years), during which 3810 CAD patients and 17,608 CAD-free individuals developed T2DM. After adjusting for age, sex, assessment center, and principal components, CAD patients showed a significantly increased risk of T2DM (HR = 2.44, 95% CI: 2.35–2.53), which was weakened after adjusting for BMI (HR = 1.99, 95% CI: 1.92–2.06) and further attenuated to 1.72 (95% CI: 1.63–1.81) in the full model. Similar results (HR = 1.48, 95% CI: 1.40–1.56) were observed in the sensitivity analysis [Supplementary Table 9, https://links.lww.com/CM9/B771].

Global and local genetic correlations

We observed a strong positive overall genetic correlation, which remained statistically significant after Bonferroni correction (T2DM–CAD: rg = 0.39, P = 1.43 × 10-75) [Figure 2A]. When the effect of BMI was removed, the estimate of genetic correlation was attenuated to some extent (20%) but remained statistically significant (T2DMadjBMI–CAD: rg = 0.31, P = 1.20 × 10-36), indicating a shared genetic basis partly driven by, but also largely independent of, BMI.

Figure 2:

Genome-wide genetic correlation between T2DM and CAD. The pink bar (A) represents the magnitude of global genetic correlation. The black error bars represent plus or minus one standard error of the global genetic correlation estimate. Colored points (B,C) represent genomic regions that contribute significant genetic correlation as estimated by ρ-HESS (P <0.05/1703). Colored bars (D,E) represent loci that show significant local genetic correlation and covariance after multiple correction (P <0.05/1703). Genetic correlation withstood Bonferroni corrections (P <0.05/2) are marked with a single asterisk. BMI: Body mass index; CAD: Coronary artery disease; SNP: Single-nucleotide polymorphism; T2DM: Type 2 diabetes mellitus; T2DMadjBMI: Type 2 diabetes mellitus adjusted for BMI.

Partitioning the whole genome into 1703 LD-independent regions identified five genomic regions (2q36.2–q36.3, 5q11.2, 8p21.3, 9p21.3, and 12q24.23–q24.31; Figure 2B–E and Supplementary Table 10, https://links.lww.com/CM9/B771), with a statistically significant local genetic correlation in multiple corrections (P <2.94 × 10-5). Of these genomic regions, four (2q36.2–q36.3, 5q11.2, 8p21.3, and 9p21.3) remained significant for T2DMadjBMI and CAD when the effect of BMI was removed. An additional genomic region (6p21.33) showed a significant local signal specific to T2DMadjBMI and CAD. Of note, 9p21.3, which harbors several well-established CAD and T2DM susceptibility loci (e.g., Cyclin dependent kinase inhibitor 2A [CDKN2A] and Cyclin dependent kinase inhibitor 2B [CDKN2B]),[4] showed the strongest local effect in both analyses. These results revealed a substantial genomic overlap between T2DM and CAD.

Cross-trait meta-analysis

Pairwise CPASSOC [Supplementary Table 11 and Supplementary Figure 2, https://links.lww.com/CM9/B771] identified a total of 107 independent significant pleiotropic SNPs (PCPASSOC <5 × 10-8 and PT2DM or CAD <1 × 10-5).

Five novel SNPs (rs4135268, rs6449, rs35497503, rs12452590, and rs2306527) [Table 2] shared between T2DM and CAD were found after excluding SNPs that were genome-wide significant single-trait-associated loci (PT2DM or CAD <5 × 10-8) or were in LD (r2 ≥0.05) with any of the previously reported genome-wide significant loci [Supplementary Tables 12–14, https://links.lww.com/CM9/B771]. Among these SNPs, rs4135268 was reported to be associated with BMI, which, as expected, was no longer significant in the T2DMadjBMI–CAD CPASSOC results. SNP rs6449 was located near cytochrome P450 family 21 subfamily A member 2 (CYP21A2) and found to be associated with cardiometabolic morbidities.[33] SNP rs35497503 was located near cyclin dependent kinase 12 (CDK12) and found to be associated with high triglyceride levels in Arab populations.[34] SNP rs12452590 was located near mannose receptor C type 2 (MRC2), known as a molecular marker and a therapeutic target for diabetic nephropathy.[35] SNP rs2306527 was located near ubiquitin specific peptidase 36 (USP36) and found to be associated with T2DM in a meta-analysis of two United States nested case–control studies.[36]

Table 2 - Novel SNPs identified in cross-trait meta-analysis between T2DM and CAD. SNP CHR BP A1 A2 Beta P-T2DM P-CAD P-CPASSOC Mapped genes † T2DM CAD T2DM and CAD rs4135268* 3 12437237 C G -0.053 -0.044 4.00 × 10-6 1.97 × 10-6 10.64 × 10-10 PPARG rs6449 6 32006655 T C 0.034 0.027 2.00 × 10-6 3.33 × 10-6 3.32 × 10-10 TNXB, C4B-AS1, CYP21A2, C4B, XXbac-BPG116M5.15, C4A, C4A-AS1 rs35497503 17 37620627 T C -0.035 -0.026 3.60 × 10-6 5.74 × 10-6 7.72 × 10-10 CDK12 rs12452590 17 60720058 T G 0.032 0.024 2.50 × 10-6 7.63 × 10-6 1.17 × 10-9 MRC2 rs2306527 17 76798155 T C -0.031 -0.024 1.00 × 10-6 1.71 × 10-6 9.24 × 10-11 USP36 T2DMadjBMI and CAD rs4762753 12 20579969 T G -0.042 -0.031 5.70 × 10-6 4.83 × 10-7 3.17 × 10-10 PDE3A

A1, effect allele; A2, non-effect allele; BP, physical position of SNP (base pairs). *Previously reported loci associated with BMI according to GWAS Catalog or PhenoScanner. †VEP.　BMI: Body mass index; CAD: Coronary artery disease; C4B-AS1: C4B antisense RNA 1; CYP21A2: Cytochrome P450 family 21 subfamily A member 2; C4B: Complement C4B; C4A: Complement C4A; C4A-AS1: C4A antisense RNA 1; CDK12: Cyclin dependent kinase 12; CHR: Chromosome; GWAS: Genome-wide association study; MRC2: Mannose receptor C type 2; PDE3A: Phosphodiesterase 3A; P-T2DM: P value for association with T2DM; P-CAD: P value for association with CAD; P-CPASSOC: P value for association with T2DM and CAD; SNP: Single-nucleotide polymorphism; T2DM: Type 2 diabetes mellitus; T2DMadjBMI: Type 2 diabetes mellitus adjusted for BMI; USP36: Ubiquitin specific peptidase 36; VEP: Variant Effect Predictor.

A novel locus (rs4762753) was further identified for T2DMadjBMI and CAD. This SNP was located near phosphodiesterase 3A (PDE3A) and found to be associated with the nitric oxide (NO)–cyclic guanosine monophosphate (cGMP) signaling pathway.[37]

Of the total six novel SNPs identified across both analyses, SNP rs4762753 was not previously reported to be associated with any other traits [Supplementary Table 15, https://links.lww.com/CM9/B771].

Functional annotation and tissue enrichment analysis

While failing to remain significant after multiple corrections (Bonferroni-corrected P <9.26 × 10-4) [Supplementary Figure 3, https://links.lww.com/CM9/B771], suggestive enrichment for the expression of genes shared by T2DM and CAD was observed in the liver, skeletal muscle, brain frontal cortex Brodmann area 9 (BA9), endocervix, heart left ventricle, adrenal gland, brain, cervical c-1 spinal cord, brain anterior cingulate cortex BA24, brain cortex, and fallopian tubes (P <0.05). Suggestive enrichments for the expression of genes shared by T2DMadjBMI and CAD were observed in the brain frontal cortex BA9 region, pancreas, brain cortex, endocervix, whole blood, liver, transverse colon, small intestine terminal ileum, artery aorta, skeletal muscle, artery coronary, and brain caudate basal ganglia.

Univariable MR analysis

In the bidirectional two-sample MR analysis [Figure 3 and Supplementary Figures 4–7, https://links.lww.com/CM9/B771], genetically determined T2DM significantly increased the risk of CAD (odds ratio [OR] = 1.13, 95% CI: 1.11–1.16). Conversely, genetically determined CAD was also associated with an increased risk of T2DM (OR = 1.12, 95% CI: 1.07–1.18). This bidirectional causal relationship was further supported by sensitivity analysis and was not affected by directional pleiotropy. Removing the effect of BMI from T2DM did not seem to influence the bidirectional causal relationship (T2DMadjBMI→CAD: OR = 1.10, 95% CI: 1.07–1.13; CAD→T2DMadjBMI: OR = 1.14, 95% CI: 1.09–1.20). Furthermore, LHC-MR, accounting for the heritable confounder, showed similar results (T2DM→CAD: OR = 1.12, 95% CI: 1.10–1.15; CAD→T2DM: OR = 1.42, 95% CI: 1.21–1.67), corroborating the robustness of the findings. The bidirectional effect remained consistent when genetic instruments from multiancestry GWAS of T2DM were used (T2DM→CAD: OR = 1.13, 95% CI: 1.10–1.16; CAD→T2DM: OR = 1.08, 95% CI: 1.04–1.13; Supplementary Figure 8, https://links.lww.com/CM9/B771).

Figure 3:

Bidirectional Mendelian randomization analysis between T2DM and CAD. The boxes denote point estimate of the causal effects, and the error bars denote 95% CI. Inverse-variance-weighted approach was used as primary analysis, with MR-Egger and weighted median approaches used as sensitivity analysis. 95% CI: 95% confidence interval; BMI: Body mass index; CAD: Coronary artery disease; MR: Mendelian randomization; OR: Odds ratio; SNP: Single-nucleotide polymorphism; T2DM: Type 2 diabetes mellitus; T2DMadjBMI: Type 2 diabetes mellitus adjusted for BMI.

The mean F-statistic of our IVs was larger than 50 [Supplementary Table 5, https://links.lww.com/CM9/B771], indicating strong instruments. With the current sample size for the outcomes, assuming 2.47% (T2DM), 2.06% (T2DMadjBMI), and 1.44% (CAD) phenotypic variance explained by the IVs based on the data that we used, our study had more than 80% statistical power to detect an OR of 1.05 for genetically determined T2DM on the risk of CAD, 1.09 for genetically determined CAD on the risk of T2DM, 1.05 for genetically determined T2DMadjBMI on the risk of CAD, and 1.11 for genetically determined CAD on the risk of T2DMadjBMI.

Multivariable MR and mediation analysis

After adjusting for BMI [Supplementary Tables 16 and 17, https://links.lww.com/CM9/B771], the effect of T2DM on CAD (OR