Type 2 Diabetes and Hypertension: A Study on Bidirectional Causality

Abstract

Rationale:

In observational studies, type 2 diabetes (T2D) has been associated with an increased risk of hypertension (HTN), and vice versa; however, the causality between these conditions remains to be determined.

Objectives:

This population-based prospective cohort study sought to investigate the bidirectional causal relations of T2D with HTN, systolic and diastolic blood pressure (SBP and DBP) using Mendelian randomization (MR) analysis.

Methods and Results:

After exclusion of participants free of a history of heart failure, cardiovascular disease, cardiac procedures, and non T2D diabetes, a total of 318,664 unrelated individuals with qualified genotyping data of European descent aged 37–73 from UK Biobank were included. The genetically instrumented T2D and HTN were constructed using 134 and 233 single nucleotide polymorphisms (SNPs), respectively. Seven complementary MR methods were applied, including inverse variance weighted method (IVW), two median-based methods (simple and weighted), MR-Egger, MR-RAPS, MR-PRESSO, and multivariate MR. The genetically instrumented T2D was associated with risk of HTN (OR 1.07 [95% CI, 1.04–1.10], P=3.4×10⁻⁷), whereas the genetically determined HTN showed no relationship with T2D (OR 0.96 [0.88–1.04], P=0.34). Our MR estimates from T2D to BP showed that the genetically instrumented T2D was associated with a 0.67 mm Hg higher SBP (95% CI 0.41–0.93, P =5.75×10⁻⁷), but not with a higher DBP. There was no clear evidence showing a causal effect of elevated SBP or DBP on T2D risk. Positive pleiotropic bias was indicated in the HTN→T2D relation (OR of MR-Egger intercept 1.010 [1.004–1.016], P=0.001), but not from T2D to HTN (1.001 [0.998–1.004], P=0.556).

Conclusions:

T2D may causally affect HTN, whereas the relationship from HTN to T2D is unlikely to be causal. These findings suggest the importance of keeping an optimal glycemic profile in general populations, and BP screening and monitoring, especially SBP, in patients with T2D.

INTRODUCTION

Type 2 diabetes (T2D) and hypertension (HTN), the two leading components of the global burden of disease, are commonly found to coexist ^1–3. The co-existence of T2D and HTN confers a dramatically increased risk (2~4 fold) of cardiovascular disease, end-stage kidney disease, and death, compared to the normotensive and nondiabetic adults ³. Hence, understanding the bidirectional relations between T2D and HTN is of significant public health importance regarding disease prevention and management of complications.

A large group of prospective studies has associated T2D with an increased HTN risk ^4–6, and a similar amount of evidence has been reported on the positive relationship between blood pressure and incident T2D ⁷. Nevertheless, these prior observational data were limited for causal inference due to the potential bias introduced by confounding factors and/or reverse causality.

In recent years, Mendelian randomization (MR) analysis, a form of instrumental variable (IV) analysis that leverages the random assortment of genetic variants during gamete formation and therefore minimizes the influence of confounding and reverse causation, has been increasingly used in estimating causal inference between exposures and outcomes ⁸. In the present study, we performed bidirectional MR analyses for causal inference between T2D and HTN among 318,664 participants from the UK Biobank, as well as bidirectional MR association analysis of T2D with systolic and diastolic blood pressure (SBP and DBP).

METHODS

Because of the sensitive nature of the data collected for this study, requests to access the dataset from qualified researchers trained in human subject confidentiality protocols may be sent to UK Biobank & the Access Team at access@ukbiobank.ac.uk

Data sources and study participants.

The UK Biobank is a large prospective study of over half a million participants aged 37–73 years living in the United Kingdom. Participant recruitment was conducted in 22 centers across the UK between 2006 and 2010, with a variety of individual-level health information obtained from self-administrated questionnaires, physical measurements, biological sample tests, and linked health records ⁹. In the present study, we firstly excluded participants who withdrew from the cohort till Oct 16, 2018 (n=73), had a history of heart disease and procedures (n=61,827) or diabetes other than T2D¹⁰ (n=5989), non-European (n=29815), and without validated genotyping data (n=110,670), leaving 318,664 unrelated European participants for the final analyses (a flow chart of selection of study participants was shown in details in Figure 1). Definitions of above-mentioned diseases and heart procedures are presented in the Online Table I and Table II, respectively. All participants provided electronic informed consent, and the study was approved by the NHS National Research Ethics Service (Ref: 11/NW/0382), and Institutional Review Board of Tulane University Health Sciences Center (Study number: 2018–1872).

Figure 1. Flow Diagram.

ASCVD, atherosclerotic cardiovascular disease; T2D, type 2 diabetes; QC, quality control; PCA, principle component analysis.

Ascertainment of T2D and HTN.

Information on prevalent and incident T2D and HTN was regularly collected through cumulative medical records of hospital diagnoses and was supplemented by survey data from questionnaires and physical measures at baseline and in two repeated surveys. The International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) was used in death records; whereas both ICD-10-CM and ICD-9-CM were used in medical records. T2D was defined using a validated algorithm according to self-report diabetes data¹⁰, as well as ICD-9/10-CM coded hospital records (Online Table I). We defined HTN cases using ICD-10-CM code “I10” and ICD 9-CM”250” and “401”, self-reported physician-diagnosed HTN, and the use of blood pressure (BP) lowering medication not due to heart disease (including angiotensin-converting-enzyme inhibitor, angiotensin II receptor blocker, beta blocker, calcium channel blocker, and diuretics). For the BP measures in each survey, systolic BP (SBP) ≥140 mm Hg or diastolic BP (DBP) ≥90 mm Hg was used to define HTN.

Measurements.

Sociodemographic characteristics at recruitment were obtained from local NHS Primary Care Trust registries before arrival at the Assessment Centre, including age (measured in years), sex (female/male), ethnicity (British/Irish/Any other white ethnicity), and Townsend deprivation index (a proxy for the socio-economic position). Information on lifestyles including physical activity (metabolic equivalents (METs) minutes per week calculated on the basis of the International Physical Activity Questionnaire short form¹¹), smoking, and alcohol drinking status (never/previous/current/missing) were obtained using a touch-screen questionnaire. Body mass index (BMI) was derived from weight in kilograms (BC-418MA body composition analyzer) divided by standing height in meters (Seca 202 stadiometer) squared.⁹ SBP/DBP (mm Hg) were averaged over two repeated automated measurements (OMRON Healthcare Europe, NA, Hoofddorp) ⁹. For participants who reported to be taking BP medication (19.08% of individuals) at baseline and two repeated surveys, we adjusted for medication use by adding 15 and 10 mm Hg to SBP and DBP, respectively. For a small portion (4~5%) of participants who participated in more than one follow-up visits, an average value of above mentioned quantitative measures (including PA METs minutes, BMI, and BP) was used in our analyses. For participants with missing data on BMI (0.29%), SBP (5.46%), DBP (5.46%), Townsend deprivation index (0.12%), and physical activity (0.69%), a “predictive mean matching” multiple imputation approach¹² was applied.

Genotype and imputed data.

UK Biobank genotyping was conducted by the Affymetrix Research Services Laboratory in Santa Clara, California, USA, using two similar custom-designed chips (UK BiLEVE array and UK Biobank Axiom array). General quality control procedures (P for Hardy–Weinberg equilibrium test≥1.0×10⁻⁶, call rate ≥90%, and imputation R²≥0.3) were employed in the UK Biobank genetic data analysis. Forty genetic principal components (PCs) were calculated, accounting for the effects of the population structure and batch-based genotyping. Genotype imputation was further performed using the 1000 Genomes Phase 3 reference panel, resulting in a dataset of 92,693,895 variants in 487,409 individuals.

Genetic instrument variables for T2D and HTN.

Given the results of four genome-wide association studies (GWAS) in Europeans from DIAGRAM consortium^13–16, we identified 134 out of 193 T2D-related single nucleotide polymorphisms (SNPs) that passed genotyping QC, and restricted to a set of only biallelic SNPs on 22 auto-chromosomes not in a linkage disequilibrium (LD) clumping (R²<0.01 within 1 Mb using European population genotype data originated from Phase 3 (Version 5) of the 1000 Genomes Project) as reference. Similarly, 233 out of 262 novel and previously reported SNPs from a recent GWAS of BP conducted by Warren et al¹⁷ were satisfied with our inclusion criteria. Thus, a total of 367 SNPs were selected for genetic IVs (Online Table III).

Mendelian randomization analysis.

A diagram for MR was presented by using the genetic variants as IVs (Online Figure I). As recommended, six complementary MR approaches were adopted in our analyses to assess the causal effect of the exposure on the outcome and its robustness, including inverse variance weighted method (IVW), two median-based methods (simple and weighted), MR-Egger regression, MR-RAPS (Mendelian randomization using the robust adjusted profile score)¹⁸, and MR-PRESSO (Mendelian Randomization Pleiotropy RESidual Sum and Outlier)¹⁹. We conducted heterogeneity tests in MR analyses using IVW and MR-Egger, sensitivity analyses were performed using the weighted median method and Leave-one-out analysis²⁰ (Online Figure II-V). If there was no evidence of directional pleiotropy (P for MR-Egger intercept>0.05^8,21), the estimate from the inverse-variance weighted (IVW) method was considered as the most reliable indicator. Two types of pleiotropy-corrected MR estimates were also reported, including the MR-RAPS estimate (a method for correcting for pleiotropy using robust adjusted profile scores)¹⁸ and the MR-PRESSO estimate (a method for correcting for outliers in IVW)¹⁹. A consistent MR effect across the six methods might indicate a true causal effect.²² In addition, multivariable MR (MMR) analysis was performed by considering BMI and dyslipidemia as potential confounders or intermediators (Online Table IV).

A multivariate logistic or linear regression model was fitted for binary traits or continuous traits, respectively, as well as odds ratios (OR) and regression coefficients from two models used as the MR estimates. In the IV-exposure association analyses, we controlled for age, sex, Townsend deprivation index, assessment center (22 centers), batch effects (106 batches), and the first ten genetic PCs as covariates (model 1). In the IV-outcome association analyses, we further adjusted for BMI, dyslipidemia, smoking status, alcohol drinking status, and PA METs minutes per week (model 2).

All data analyses were conducted using R version 3.4.4. Missing data imputation was performed using the “MICE” R package, and MR analyses were conducted using the “Two sample Mendelian randomization”, “MR-RAPS” and “MR-PRESSO” R packages. The threshold of statistical significance was P<0.05 (2-sided α=0.05).

RESULTS

The average age of the 318,664 UK Biobank participants included was 56.2 years, 44.8% were men, and 93.1% were British (Table 1). Two-thirds of the participants were overweight or obese (BMI≥25.0). Current smokers and drinkers accounted for 10.2% and 93.7%, respectively. There were 13,931 (4.4%) and 172,344 (54.1%) participants with T2D and HTN, respectively. Of note, 85.1% of T2D patients had HTN, while 6.9% of hypertensive participants were diabetic.

Table 1.

Characteristics of Participants from UK Biobank Used in the Analysis

Variable	Participants, No. (%) (n=318,664)
Age, mean (SD), y	56.21 (8.01)
Race/ethnicity, No. (%)
British	296653 (93.09)
Irish	8744 (2.74)
Other Whites	13267 (4.16)
Male, No. (%)	142643 (44.76)
Townsend deprivation index, median (IQR)	−2.3 (-3.73 to 0.16)
BMI, kg/m2	27.21 (4.70)
BMI group
<18.5	1701 (0.53)
18.5~24.9	108898 (34.17)
25.0~29.9	135311 (42.46)
30.0~39.9	67283 (21.11)
≥40.0	5471 (1.72)
Physical activity, median (IQR), MET mins per week	1568 (690 to 3249)
Smoking status, No. (%)
Never	176347 (55.34)
Previous	108819 (34.15)
Current	32509 (10.20)
missing	989 (0.31)
Drinking status, No. (%)
Never	9730 (3.05)
Previous	10246 (3.22)
Current	298486 (93.67)
missing	202 (0.06)
Dyslipidemia, No. (%)	41556 (13.04)
Type 2 diabetes, No. (%)	13931 (4.37)
Hypertension in adults with type 2 diabetes, No. (%)	11855 (85.10)
SBP, mean (SD), mm Hg	140.6 (20.45)
DBP, mean (SD), mm Hg	84.16 (11.20)
Hypertension, No. (%)	172344 (54.08)
Type 2 diabetes in adults with hypertension, No. (%)	11855 (6.88)

SD, standard deviation; IQR, interquartile range; BMI, body mass index; MET, metabolic equivalent; SBP, systolic blood pressure; DBP, diastolic blood pressure.

In our bidirectional MR analysis, a total of 134 and 233 SNPs were included as genetic IVs for T2D and HTN, respectively (Figure 1). In the T2D→HTN MR analysis by using IVW method, the genetically instrumented T2D increased HTN risk (OR: 1.07; 95% confidence interval [CI], 1.04–1.10, P=3.4×10⁻⁷) without detected pleiotropy bias (P=0.56). In contrast, pleiotropy bias was indicated in the HTN→T2D MR analysis (P for MR-Egger intercept=0.001), and the IVW estimate showed there was no association between the genetically instrumented HTN and T2D (OR: 0.98; 95% CI, 0.90–1.08, P=0.70). Even after correcting for pleiotropy, our MR results still demonstrated that there was no causal effect of HTN on T2D (OR 0.96 [95%CI: 0.88–1.04], P=0.34 in MR-RAPS; and OR 0.95 [95%CI: 0.88–1.02], P=0.14 in MR-PRESSO). When analyzed using different MR methods, the causal effect of T2D on HTN was quite robust and consistent (ORs ranged from 1.06 to 1.09, all P<0.01), whereas the genetically instrumented HTN was not associated with T2D (ORs ranged from 0.95 to 1.05, all P>0.05 except for MR-Egger). Of note, in our MMR analysis, unbalanced horizontal pleiotropy owing to BMI but not dyslipidemia in the T2D→HTN relation was detected (OR, 0.94 [94% CI, 0.90–0.98], P=0.004), the causal effect of T2D on HTN was further enhanced after controlling for such negative bias (OR: 1.08; 95% confidence interval [CI], 1.05–1.11, P=8.7×10⁻⁸) (Online Table IV). No horizontal pleiotropy due to BMI or dyslipidemia was detected in our HTN→T2D MR analyses (Online Table IV).

Furthermore, our MR analyses showed that the genetically instrumented T2D was consistently associated with a higher SBP (regression coefficients [β] in mm Hg ranged from 0.39 to 0.83, all P<0.01 except for P for MR-Egger=0.066), but not with a higher DBP (β ranged from −0.07 to 0.32) (Table 2). When SBP/DBP→T2D relations were analyzed, neither the genetically instrumented SBP nor DBP increased T2D risk (OR ranged from 0.976 to 1.003 for the SBP→T2D relation, and OR ranged from 0.961 to 1.013 for the DBP→T2D relation) (Table 3).

Table 2.

Mendelian Randomization Associations of Type 2 Diabetes with Systolic and Diastolic Blood Pressure using Genetic Instrument Variables

	NO. of genetic IVs	Effect Size (95% CI)	P
T2D → SBP
IVW	134	0.67 (0.41 to 0.93)	5.75×10−7
Simple median	134	0.83 (0.47 to 1.18)	4.64×10−6
Weighted median	134	0.57 (0.29 to 0.85)	5.49×10−5
MR-Egger	134	0.39 (−0.02 to 0.81)	0.066
MR-RAPSa	134	0.75 (0.48 to 1.01)	2.58×10−8
MR-PRESSO	127c	0.65 (0.42 to 0.88)	1.21×10−7
Test for Heterogeneity: P = 2.29×10−40 (IVW) and P = 4.21×10−39 (MR-Egger)
Test for Horizontal pleiotropy: MR-Egger intercept = 0.025 (−0.004 to 0.054), P = 0.10
T2D → DBP
IVW	134	0.20 (0.05 to 0.36)	0.008
Simple median	134	0.32 (0.11 to 0.53)	0.003
Weighted median	134	−0.07 (−0.22 to 0.07)	0.333
MR-Egger	134	−0.05 (−0.28 to 0.19)	0.680
MR−RAPSb	134	0.22 (0.07 to 0.38)	0.005
MR-PRESSO	131d	0.17 (0.03 to 0.31)	0.021
Test for Heterogeneity: P = 3.59×10−38 (IVW) and P = 1.20×10−34 (MR−Egger)
Test for Horizontal pleiotropy: MR-Egger intercept = 0.02 (0.006 to 0.04), P = 0.01

The effect size was presented as a regression coefficient and its 95% confidence interval. CI, confidence interval; T2D, type 2 diabetes; SBP, systolic blood pressure; DBP, diastolic blood pressure; IVW, the inverse-variance weighted (IVW) method; IV, instrument variables; MR, Mendelian randomization; MR-RAPS, an MR method for correcting for horizontal pleiotropy using robust adjusted profile scores; MR-PRESSO, an MR method for correcting for pleiotropy residual sum and outlier.

^aMR-RAPS estimates were given after pruning two SNPs (rs73455744 and rs7041847) with extraordinarily large direct effects¹⁸;

^bMR-RAPS estimates were given after pruning six SNPs (rs10922502, rs2760061, rs17477177, rs12628032, rs10948071, and rs449789) with extraordinarily large direct effects¹⁸;

^cIV outliers detected: rs11786613, rs1061810, rs10830963, rs11063018, rs12899811, rs78761021, and rs7578326;

^dIV outliers detected: rs1531583, rs1061810, and rs78761021.

Table 3.

Mendelian Randomization Associations of Systolic and Diastolic Blood Pressure with Type 2 Diabetes using Genetic Instrument Variables

	NO. of genetic IVs	Effect Size (95% CI)	P
SBP → T2D
IVW	233	0.999 (0.990 to 1.008)	0.748
Simple median	233	1.003 (0.992 to 1.014)	0.633
Weighted median	233	0.998 (0.988 to 1.008)	0.704
MR-Egger	233	0.976 (0.958 to 0.995)	0.012
MR-RAPS	233	0.996 (0.988 to 1.005)	0.361
MR-PRESSO	229a	0.996 (0.988 to 1.003)	0.268
Test for Heterogeneity: P = 2.72×10−26 (IVW) and P = 1.83×10−24 (MR-Egger)
Test for Horizontal pleiotropy: MR-Egger intercept = 1.009 (1.002 to 1.016), p = 0.008
DBP → T2D
IVW	233	0.995 (0.979 to 1.011)	0.556
Simple median	233	1.013 (0.992 to 1.034)	0.215
Weighted median	233	0.994 (0.978 to 1.011)	0.499
MR-Egger	233	0.961 (0.932 to 0.989)	0.008
MR-RAPS	233	0.992 (0.977 to 1.007)	0.284
MR-PRESSO	229a	0.992 (0.979 to 1.006)	0.248
Test for Heterogeneity: P = 3.20×10−26 (IVW) and P = 2.76×10−24 (MR-Egger)
Test for Horizontal pleiotropy: MR-Egger intercept = 1.008 (1.002 to 1.014), P = 0.006

The effect size was presented as odds ratio and its 95% confidence interval. CI, confidence interval; T2D, type 2 diabetes; SBP, systolic blood pressure; DBP, diastolic blood pressure; IVW, the inverse-variance weighted (IVW) method; IV, instrument variables; MR, Mendelian randomization; MR-RAPS, an MR method for correcting for horizontal pleiotropy using robust adjusted profile scores; MR-PRESSO, an MR method for correcting for pleiotropy residual sum and outlier.

^aIV outliers: rs2071518, rs2782980, rs8059962, and rs76326501.

DISCUSSION

By using individual-level data for 318,664 UK Biobank participants, our bidirectional MR analyses showed consistent evidence that the genetically instrumented T2D increased HTN risk, whereas the MR estimates for the HTN→T2D relation were unlikely to be causal. In addition, the genetically instrumented T2D was strongly associated with a higher SBP rather than a higher DBP.

Our study is the first to provide strong evidence for a causal relationship of T2D with HTN risk, dominantly driven by the causal effect of T2D on a higher SBP instead of DBP. Goharian et al ²³, however, didn’t find a causal relationship between fasting glucose and SBP/DBP in healthy children and adolescents, likely due to insufficient statistical power (n=1506), narrowed variance of BP in the young population, a lack of cumulative impact of raised glucose level on BP in childhood, or pleiotropy bias. However, our findings are in line with prospective cohort studies showing T2D and hyperglycemia were associated with incident HTN ^4–6, as well as two MR studies conducted in general adults showing that higher glucose level ²⁴ and greater genetic predisposition to T2D ²⁵ were associated with increased arterial stiffness, which coincided with the development of HTN. The precise mechanisms for our findings of a causal relationship of T2D with a higher SBP but not a higher DBP are largely unknown. We speculate that an accelerated arterial stiffness resulting from T2D was associated with a greater increase in SBP instead of a higher DBP during the aging process ^1,26,27. Furthermore, a recent meta-analysis of 49 trials demonstrated that the antihypertensive treatment for lowering SBP rather than DBP in patients with diabetes reduced the risk of all-cause and cardiovascular mortality substantially, as well as incident myocardial infarction, stroke, heart failure, and end-stage renal disease ²⁸.

The T2D→HTN causality is biologically plausible. T2D shared broad cardiometabolic disorders, including obesity, insulin resistance (IR), β-cell dysfunction, inflammation, oxidative stress, vascular dysfunction, sodium retention, sympathetic excitation, renin-angiotensin-aldosterone system activation, and kidney damage, which has been widely proposed in the initiation and maintenance of HTN ^3,29. However, the magnitude of our genetic association of T2D on HTN (ORs ranged from 1.06 to 1.09) were much lower compared to the observational associations in the current study (multivariate-adjusted OR, 2.42 [95% CI, 2.32–2.53] in Online Table V) and in Framingham Offspring Study (3.14 [95% CI, 2.17–4.54])³⁰. First, observational associations might be over-estimated due to various residual confounding and other bias (e.g., subtle arterial stiffness and kidney disease ahead of T2D onset^3,5). Second, the genetically instrumented T2D in our MR analyses might not be comprehensively characterized by a complex network of T2D pathophysiologic mechanisms^31,32 on HTN development. Third, the role of detection bias in inflating the observational estimates should also be considered, where patients with diagnosed and treated diabetes will be much more likely to have close surveillance of their BP and start on antihypertensive medications for reasons (e.g., heart failure, left ventricular hypertrophy, and chronic kidney disease) rather than elevated BP ⁵. Additionally, as our MMR analysis showing a negative association between T2D-IVs and BMI, we speculated that the lower MR estimate might be as a result of undetected negative bias due to unbalanced pleiotropic effects of T2D-IVs on above mentioned biological pathways⁸. Further studies are warranted to elucidate the precise mechanisms.

Previous MR studies for a causal relationship from elevated BP to an increased risk of T2D have yielded inconsistent results. Aikens et al ³³ reported that 1-mm Hg genetic increase in SBP was associated with a 2% increased risk of T2D, by adopting a 2-sample MR approach integrating summary-level GWAS data from 37,293 T2D cases and 125,686 controls. In contrast, a more recent MR study conducted by Zhu et al ³⁴ using data from two community-based studies (n=162,030) showed that there were no causal relations from BP to T2D (OR, 1.07 [95%CI, 0.89–1.29]; P=0.44 for SBP→T2D; OR, 1.12 [95%CI, 0.94–1.33]; P=0.20 for DBP→T2D), which was in line with our findings. In a case-control setting, selection bias might be inferred as T2D patients were more likely to take BP lowering medications compared with controls, in which the use of β blocker and diuretics was associated with the increased risk of T2D ³⁵. Hence, MR estimates based on a general population would be less biased for causal inference between HTN and T2D risk ³⁶. Moreover, in compared with 29 BP-related SNPs accounted for only 2.2% of BP variance ³⁷ were adopted as IVs in the above two MR studies ^33,34, a total number of over 200 SNPs here we used as IVs for BP might be more indicative as over 3.56% of the variance could be explained¹⁷.

Previous reported observational associations of HTN with T2D risk were unlikely to be causal, and instead, might be the results of two of bias ― collider bias or pleiotropy ⁸. As mentioned previously ³³, a false association between HTN and T2D might occur when the study sample comprised an excess number of undiagnosed coronary artery disease cases in T2D patients (collider stratification) ³⁸. Our MR estimates, however, were less likely to be affected by collider bias as we excluded patients with prevalent and incident cardiovascular diseases before running MR analyses. But still, the causal HTN→T2D relationship can be true in terms of recently proposed biologic mechanisms that HTN manifests as vasoconstriction ³⁹, IR ⁴⁰, and inflammation ³³, which increase the risk of T2D.

Our study is the first to use a bidirectional MR approach to investigate the causal relationship between T2D and HTN/SBP/DBP among over 300 thousand adults, with a variety of sensitivity analyses and MR diagnostics performed for evaluating the robustness of our MR estimates. However, there are several limitations. First, the lack of data on glycemic traits (e.g., fasting glucose, insulin, and HbA1c) might have led to a small number of undiagnosed T2D cases, who were misclassified into controls. However, the information on the linked health records provided us with an alternative way to identify such cases. Second, the lack of data on insulin resistance, β cell function, chronic inflammation, and renal function limited further investigation of the precise mechanisms underlying the observed bidirectional associations between T2D and HTN. Third, the present MR analyses conducted in participants of European descent might limit the generalization of our findings in other ancestry groups.

In summary, our comprehensively bidirectional MR results suggest a potentially causal T2D→HTN relationship, especially a causal relationship of T2D with a higher SBP but not with a higher DBP. In contrast, the HTN→T2D association was unlikely to be causal. Our findings have clinical significance for maintaining an optimal glycemic profile for the general populations, and the importance of BP screening and monitoring, especially SBP, in patients with T2D.

Figure 2. Mendelian Randomization Association between Type 2 Diabetes and Hypertension using Genetic Instrument Variables.

T2D, type 2 diabetes; HTN, hypertension; IVs, instrument variables; IVW, the inverse-variance weighted (IVW) method; MR, Mendelian randomization; OR, odds ratio; CI, confidence interval; MR RAPS, an MR method for correcting for horizontal pleiotropy using robust adjusted profile scores.

^a, MR-RAPS estimates were given after pruning two SNPs (rs73455744 and rs7041847) with extraordinarily large direct effects¹⁸;

^b, MR-RAPS estimates were given after pruning six SNPs (rs10922502, rs2760061, rs17477177, rs12628032, rs10948071, and rs449789) with extraordinarily large direct effects¹⁸;

^c, MR-PRESSO IV outlier detected was rs12899811;

^d, MR-PRESSO IV outliers detected: rs4660293, rs62270945, rs2071518, rs2782980, rs8059962, and rs76326501.

NOVELTY AND SIGNIFICANCE.

What Is Known?

• Type 2 diabetes (T2D) is associated with an increased risk of hypertension (HTN), and vice versa, in observational studies.

What New Information Does This Article Contribute?

• T2D may causally affect hypertension, whereas the HTN→T2D relation is unlikely to be causal.

In this bidirectional Mendelian randomization analysis, the genetic predisposition to T2D is associated with the development of HTN and elevated SBP, but not with elevated diastolic BP; whereas the “HTN to T2D” relation is unlikely to be causal. BP control, especially for SBP control, in patients with T2D is essential in clinical practice and self-management, and it is of great importance for general populations to maintain an optimal glycemic profile and a normal BP level.

Nonstandard Abbreviations and Acronyms:

BMI

body mass index

BP

blood pressure

CI

confidence interval

DBP

diastolic blood pressure

GWAS

genome-wide association study

HTN

hypertension

IV

instrumental variable

IVW

the inverse-variance weighted method

MR

Mendelian randomization

OR

odds ratio

RAPS

robust adjusted profile scores

SBP

systolic blood pressure

SNP

single nucleotide polymorphism

T2D

type 2 diabetes