Impact of Glucose Level on Micro- and Macrovascular Disease in the General Population: A Mendelian Randomization Study
Abstract
OBJECTIVE To evaluate whether high glucose levels in the normoglycemic range and higher have a causal genetic effect on risk of retinopathy, neuropathy, nephropathy, chronic kidney disease (CKD), peripheral arterial disease (PAD), and myocardial infarction (MI; positive control) in the general population.
RESEARCH DESIGN AND METHODS This study applied observational and one-sample Mendelian randomization (MR) analyses to individual-level data from 117,193 Danish individuals, and validation by two-sample MR analyses on summary-level data from 133,010 individuals from the Meta-Analyses of Glucose and Insulin-Related Traits Consortium (MAGIC), 117,165 from the CKDGen Consortium, and 452,264 from the UK Biobank.
RESULTS Observationally, glucose levels in the normoglycemic range and higher were associated with high risks of retinopathy, neuropathy, diabetic nephropathy, PAD, and MI (all P for trend <0.001). In genetic causal analyses, the risk ratio for a 1 mmol/L higher glucose level was 2.01 (95% CI 1.18–3.41) for retinopathy, 2.15 (1.38–3.35) for neuropathy, 1.58 (1.04–2.40) for diabetic nephropathy, 0.97 (0.84–1.12) for estimated glomerular filtration rate (eGFR) <60 mL/min/1.73 m2, 1.19 (0.90–1.58) for PAD, and 1.49 (1.02–2.17) for MI. Summary-level data from the MAGIC, the CKDGen Consortium, and the UK Biobank gave a genetic risk ratio of 4.55 (95% CI 2.26–9.15) for retinopathy, 1.48 (0.83–2.66) for peripheral neuropathy, 0.98 (0.94–1.01) for eGFR <60 mL/min/1.73 m2, and 1.23 (0.57–2.67) for PAD per 1 mmol/L higher glucose level.
CONCLUSIONS Glucose levels in the normoglycemic range and higher were prospectively associated with a high risk of retinopathy, neuropathy, diabetic nephropathy, eGFR <60 mL/min/1.73 m2, PAD, and MI. These associations were confirmed in genetic causal analyses for retinopathy, neuropathy, diabetic nephropathy, and MI, but they could not be confirmed for PAD and seemed to be refuted for eGFR <60 mL/min/1.73 m2.
Introduction
High glucose levels are observationally and causally associated with a high risk of ischemic heart disease in individuals with and those without diabetes (1,2). Similarly, hyperglycemia is a well-known risk factor for microvascular disease in individuals with diabetes, and treatment with glucose-lowering drugs reduces the risk of microvascular complications in such individuals (3,4). Whether high glucose levels increase the risk of a spectrum of peripheral micro- and macrovascular diseases in the general population, and whether the magnitude of such risk varies between vascular compartments, is less clear. Also uncertain is whether the effect of glucose-lowering drugs on vascular disease is caused by glucose lowering per se or whether other mechanisms also are involved, as most glucose-lowering drugs have pleiotropic effects, including improvements in concomitant obesity, hyperlipidemia, and hypertension (5,6).
In order to determine whether a risk factor is causally related to a disease, several lines of evidence should be considered, including mechanistic studies, longitudinal epidemiological studies, genetic Mendelian randomization (MR) studies, and randomized controlled trials (7,8). MR is an epidemiological approach to assessing causality that uses the random assortment of alleles at conception to largely circumvent confounding and reverse causation (8). Applying the principles of MR in a general population, we tested the hypothesis that high glucose levels within the normal range and higher causally associate with high risks of retinopathy, peripheral neuropathy, and diabetic nephropathy, representing microvascular disease, or with high risks of chronic kidney disease (CKD) and peripheral arterial disease (PAD), representing a mix of micro- and macrovascular PAD. Myocardial infarction (MI) was included as a positive control for the genetic instrument, that is, as a confirmation of the association between the genetic instrument and the risk of MI.
In two cohorts from the Danish general population—participants in the Copenhagen City Heart Study (CCHS) and those in the Copenhagen General Population Study (CGPS)—we first tested whether random plasma glucose levels at baseline were prospectively associated with risk of disease. Seven variants of the genes GCP62/ABCB11, GCK, DGKB, ADCY5, CDKN2A/B, and TCF7L2 have been shown to affect glucose levels (1,9,10); thus, second, we tested whether they also did so in our populations. Third, we tested whether the genetic variants associated with high glucose levels also associated, as an indication of causality, with risk of disease. Fourth, we analyzed instrumental variables to obtain causal risk estimates per 1 mmol/L (18 mg/dL) higher glucose levels. Fifth and last, we validated our results using a two-sample MR design with summary-level data based on 26 genetic variants associated with fasting glucose levels in individuals without diabetes from the Meta-Analyses of Glucose and Insulin-Related Traits Consortium (MAGIC), and end point data from the UK Biobank and the CKDGen Consortium (11–14).
Research Design and Methods
Study Population
We included 117,193 individuals from two similar studies of the Danish general population: the CCHS and the CGPS (1). Participants were white and of Danish descent, and none were included in more than one study. For detailed descriptions of the cohorts, see Supplementary Appendix 1. Both studies were approved by institutional boards at Herlev and Gentofte Hospitals and the Danish National Committee on Health Research Ethics, Copenhagen, Denmark (KF-100.2039/91, KF-01–144/01, H-KF-01–144/01), and they were conducted according to the Declaration of Helsinki. Written informed consent was obtained from all individuals.
Plasma Glucose
Nonfasting plasma glucose was measured by using a colorimetric assay (Konelab; Boehringer, Mannheim, Germany). Blood samples were taken at random, irrespective of time since and content of the last meal. Time (hours) since the last meal was self-reported. In order to examine the observational association between glucose levels and disease, individuals without diabetes were categorized, on the basis of glucose level at baseline, into five groups that reflect stepwise increases in glucose level, ranging from ≥4.0 mmol/L to above the diabetes threshold of 11.1 mmol/L (15). Individuals with levels between 4.0 and 5.4 mmol/L were defined as the reference group.
Genotypes
An ABI PRISM 7900HT Sequence Detection System (Applied Biosystems Inc., Foster City, CA) and TaqMan-based assays were used in order to genotype seven variants of the genes GCP62/ABCB1 (rs560887), GCK (rs4607517), DGKB (rs2191349), ADCY5 (rs11708067), CDKN2A/B (rs10811661 and rs2383206), and TCF7L2 (rs7903146), which have been shown to be associated with high glucose levels (1,9,10). The variants were combined into a weighted allele score, which accounted for the effect of each allele on random plasma glucose and the frequency of each allele in the CCHS and CGPS populations (16). In order to compare the effect of the same genetic variants on fasting glucose levels, the allele score was also weighted on the basis of its effect on fasting glucose levels in MAGIC participants. The weighted allele scores were divided into five categories, cut at the 25th, 50th, 75th, and 95th percentiles, in order to reflect stepwise increases in glucose levels from below the median to extremely high values.
End Points
End points were based on diagnoses of retinopathy, peripheral neuropathy, diabetic nephropathy, PAD, and MI according to the codes from the World Health Organization’s ICD-8 and ICD-10. (For specific ICD codes, see Supplementary Table 1.) Relevant data were collected from 1 January 1977 through 10 April 2018 by reviewing diagnoses from all hospital admissions and outpatient clinic visits in the national Danish Patient Registry and the national Danish Registry of Causes of Death. CKD was defined as an estimated glomerular filtration rate (eGFR) <60 mL/min/1.73 m2, corresponding to the KDIGO definition of mildly to moderately decreased (stage G3) kidney function or worse (17). eGFR was calculated from creatinine by using the CKD-EPI creatinine equation (18). For detailed descriptions of end points and covariates, see Supplementary Appendix 2.
Statistical Analyses
We used Stata SE 14.2 to analyze the data. Deviation from Hardy-Weinberg equilibrium was tested by using the Pearson χ2 test. To test for trend across ordered categories of glucose levels, genotypes, and weighted allele scores, we used the nonparametric Cuzick extension of a Wilcoxon rank sum test.
First, the observational associations between glucose level and disease in individuals without diabetes were examined for predefined categories of glucose level (which increases with each category) by using Cox proportional hazards regression with age as the time scale and with delayed entry (left-truncation), excluding individuals with a glucose level <4.0 mmol/L (<72 mg/dL) because of nonlinearity (Supplementary Fig. 2). In the observational analyses, follow-up began at the first inclusion in a study and ended with censoring at the date of death, of an event, or of emigration (n = 610), or on 10 April 2018 (corresponding to the end of follow-up for the least updated registry)—whichever came first. We did not lose track of any individual, as all individuals living in Denmark have a Danish Civil Registration System number, which provides information, updated daily, on emigration and death. The models were adjusted for sex, birth year, current smoking, pack-years smoked, BMI, hypertension, LDL cholesterol, time since last meal, and menopausal status (for women). Missing values for covariates (0–2.4% missing) were imputed from age, sex, and cohort (i.e., whether the individual participated in CCHS or CGPS) by using multivariate imputation. Cohort was adjusted in order to accommodate for the time difference between the initiation of the two cohorts (1976 for the CCHS and 2003 for the CGPS), and for potential associated changes in patterns of risk behavior, diagnoses, and treatment options. We also examined the observational associations between glucose level (on a continuous scale) and disease by using restricted cubic splines with seven knots incorporated into Cox proportional hazards models including all individuals (those with and those without diabetes); the population-wide median glucose level was the reference value.
Second, to test whether genotypes and the weighted allele score were associated with increased risk of disease, we used unadjusted Cox proportional hazards regression (as genotypes generally have a constant effect throughout life and are largely unaffected by confounding factors), with age as the time scale. Because an individual’s genotype is constant throughout life, follow-up time began when the national Danish Patient Registry was established (1 January 1977) or on the individual’s 20th birthday, whichever came last; follow-up ended as described for the observational analyses. A critical assumption of the MR design is that the genetic instrument should influence the risk of disease only through the exposure of interest (i.e., plasma glucose). To test this, we used logistic regression to assess whether the potential confounders—age, sex, current smoking, BMI, LDL cholesterol, hypertension, alcohol intake, physical activity, education, and menopausal status (for women)—were associated with glucose levels and with the weighted allele score.
Third, instrumental variable analysis by two-stage least-squares regression (through the Stata ivreg2 and ivpois commands) was used in order to estimate the potential causal association between 1 mmol/L higher glucose levels and risk of disease (19,20). The strength of the genetic instrument (i.e., the association between genotypes and glucose levels) was confirmed with the F statistic for a weighted allele score of 98, explaining 0.3% of the variation in glucose levels (21).
Fourth, to validate and test the generalizability of the results—that is, how well they represent glucose levels per se and are able to be reproduced in another general population—we conducted two-sample MR analyses with a broader genetic instrument based on summary-level data of 26 genetic variants associated with glucose levels in the MAGIC participants (n = 133,010) (13). (For further descriptions of methods of the two-sample MR, see Supplementary Appendix 3.) Genetic variants with P values <1 × 10−8 were selected; variants in linkage disequilibrium, defined as an R2 value >0.6, were excluded. End point data for the variants were extracted from the UK Biobank (n = 452,264) (12) and the CKDGen Consortium (n = 133,413) (14). (For more details about the genetic variants used, see Supplementary Table 2.)
Finally, summary-level estimates of causal effects for the data from the CGPS, CCHS, and consortia populations were estimated for each end point through meta-analyses by using the metan command in Stata. Between-study heterogeneity was assessed by using the I2 statistic. If heterogeneity was <50%, the analyses were performed with a fixed effects model. If heterogeneity was >50%, a random effects model was chosen (22).
Results
A total of 117,193 individuals were included from the CCHS and the CGPS. Of these, 4,738 had a diagnosis of diabetes at baseline or a glucose level below 4.0 mmol/L, leaving 112,455 individuals for the observational analyses. Baseline characteristics for all individuals, and for individuals without diabetes by glucose category, are shown in Table 1. Mean glucose level as a function of time since the last meal is shown in Supplementary Fig. 1 and Supplementary Table 4. Genotype distributions did not deviate from Hardy-Weinberg expectations (all P > 0.05).
Baseline characteristics for all individuals and by random plasma glucose in categories for individuals without diabetes
Observationally High Glucose Levels and Risk of Disease
Figure 1 shows the prospective risk of vascular end points by categories of glucose level in individuals without diagnosed diabetes at baseline. Baseline glucose levels were 80% higher (mean ± SEM 8.5 ± 0.02 mmol/L) in the highest glucose category below the diabetes cutoff and 186% higher in the highest category above the cutoff (13.4 ± 0.01 mmol/L); the level in the reference category was 4.8 ± 0.01 mmol/L) (Fig. 1). Higher glucose levels were associated with stepwise higher risks of all end points, and higher risks also were observed for glucose levels within the normal range.
Observational prospective risk of retinopathy, peripheral neuropathy, diabetic nephropathy, PAD, and MI as a function of increasing random plasma glucose (p-glucose) in individuals without diabetes at baseline. Individuals with an event before baseline or with a glucose level below 4.0 mmol/L (n = 2,103) were excluded from the analyses. ∆%, percentage difference in mean glucose level.
Multifactorial adjusted hazard ratios were 13 (95% CI 8.0–20) for retinopathy, 2.5 (1.9–3.2) for peripheral neuropathy, 12 (7.6–18) for diabetic nephropathy, and 1.3 (1.1–1.6) for PAD in individuals with nonfasting glucose levels between 8.0 and 11.0 mmol/L. The risks of all end points were even higher in individuals with a nonfasting glucose level above the diabetes cutoff (≥11.1 mmol/L) (P for trend <0.001 for all end points) (Fig. 1). The known stepwise association between higher glucose levels and MI is shown as a positive control of the study’s power (P < 0.001). Restricted cubic spline analyses of the associations between glucose levels (on a continuous scale) and risk of retinopathy, peripheral neuropathy, diabetic nephropathy, and PAD for all individuals (i.e., those with and those without diabetes) showed an increasing risk of disease with glucose levels increasing from 5.2 mmol/L (Supplementary Fig. 2).
Genetically High Glucose Levels and Risk of Disease
The selected genetic variants of GCP62/ABCB11, GCK, DGKB, ADCY5, CDKN2A/B, and TCF7L2 were separately associated with stepwise higher glucose levels, as was the weighted allele score obtained when those variants were combined: the mean glucose level was 5% higher in individuals with an allele score above the 95th percentile (mean ± SEM 5.56 ± 0.02 mmol/L) than in individuals with an allele score below the 26th percentile (mean ± SEM 5.28 ± 0.01 mmol/L) (Fig. 2 and Supplementary Fig. 3). The effect of allele frequency and the direction and magnitude of the effect of each genetic variant on random plasma glucose in the individuals from CCHS and from CGPS were largely comparable to the frequency, direction, and magnitude of their effects on fasting plasma glucose in the MAGIC participants (Supplementary Table 5). Categories of genetically higher glucose levels were associated with stepwise higher risks of retinopathy (P for trend 0.002), peripheral neuropathy (P = 0.001), and diabetic nephropathy (P = 0.007) (Fig. 2). The risk of PAD was higher for individuals in the highest allele score category than for those in the lowest (HR 1.19 [95% CI 1.05–1.35]). Estimates for the known association between genetically higher glucose levels and MI is shown as a positive control of the genetic instrument.
Mean random plasma glucose (p-glucose) level as a function of weighted allele score categories, and the prospective risk of retinopathy, peripheral neuropathy, diabetic nephropathy, PAD, and MI as a function of allele score (by categories of increasing score). Risk of MI is included as a positive control of the genetic instrument. ∆%, percentage difference in mean glucose level.
Confounding Factors
We tested whether potentially confounding factors were associated with glucose level and the combined genetic variants. Age, sex, menopausal status (for women), BMI, hypertension, current smoking status, alcohol intake, physical activity, and education level were associated with glucose level but not with the combined genetic variants, indicating that pleiotropic effects are unlikely through any of the aforementioned factors (Supplementary Figs. 4 and 5).
Causal Association Between High Glucose Level and Disease
Multifactorial adjusted observational analyses and causal analyses of instrumental variables for risk estimates of disease per a 1 mmol/L higher glucose level showed observational and causal associations between high glucose levels and risk of retinopathy (observational risk ratio [RR] 1.32 [95% CI 1.29–1.35]; causal RR 2.01 [95% CI 1.18–3.41]), peripheral neuropathy (observational RR 1.16 [1.14–1.19]; causal RR 2.15 [1.38–3.35]), and diabetic nephropathy (observational RR 1.29 [1.25–1.32]; causal RR 1.58 [1.04–2.40]) (Fig. 3). For PAD, a 1 mmol/L higher glucose level had an observational RR of 1.06 (1.04–1.09) and a causal RR of 1.19 (0.90–1.58). To investigate whether the causal association between high glucose level and diabetic nephropathy also was valid for measured reduced kidney function, we performed the analyses for eGFR <60 mL/min/1.73 m2 and found an observational RR of 1.04 (1.02–1.06) and a causal RR of 0.97 (0.84–1.12) per 1 mmol/L higher glucose level. Sensitivity analyses of the instrumental variables gave similar results when using the seven genetic variants in an allele score weighted for their effects on fasting glucose level in participants in the MAGIC instead of their effect on random plasma glucose in participants in the CCHS or the CGPS (Supplementary Fig. 6).
Risk of retinopathy, peripheral neuropathy, diabetic nephropathy, CKD (defined as eGFR <60 mL/min/1.73 m2), PAD, and MI per 1 mmol/L (18 mg/dL) higher observational and causal glucose levels. Hazard ratios for a 1 mmol/L higher observational glucose level were calculated by using Cox regression in individuals without diabetes at baseline, and RRs for genetically higher glucose levels were derived from instrumental variable analyses (IVAs) of all individuals with available genotypes. Risk was also estimated by using summary-level data from the UK Biobank and the CKDGen Consortium on genetic variants tested in the MAGIC by using MR IVW estimates. The causal estimates from the CCHS and CGPS cohorts and the validation cohorts were combined through meta-analysis. The I2 values denote the percentages of between-study heterogeneity in the meta-analyses. N, number; p-glucose, plasma glucose.
On the basis of summary-level data for the 26 genetic variants associated with high fasting glucose levels in individuals without diabetes in the MAGIC, combined with end point data from the UK Biobank participants into a causal estimate by two-sample MR regression, the inverse-variance weighted (IVW) estimates were similar to the results from the CCHS and the CGPS combined, that is, an RR of 4.55 (95% CI 2.26–9.15) for retinopathy, 1.48 (0.83–2.66) for peripheral neuropathy, and 1.23 (0.57–2.67) for PAD (Fig. 3). We did not have access to diagnosis codes for diabetic nephropathy or eGFR measurements from the UK Biobank. Using summary-level data for the same 26 genetic variants from the MAGIC combined with end point data for reduced kidney function from the CKDGen Consortium, we found results similar to those in the CCHS and CGPS combined, that is, no evidence of causal association between high glucose level and eGFR <60 mL/min/1.73 m2 (RR 0.98 [95% CI 0.95–1.01]). Combining the causal estimates from the CCHS and CGPS with the IWV estimates from the UK Biobank and CKDGen Consortium by using meta-analyses, a 1 mmol/L higher glucose level was associated with a causal RR of 2.93 (95% CI 1.32–6.50) for retinopathy, 1.87 (1.32–2.67) for peripheral neuropathy, 0.98 (0.95–1.01) for eGFR <60 mL/min/1.73 m2, and 1.19 (0.90–1.56) for PAD. The corresponding two-sample MR estimates using MR Egger regression and weighted median regression showed similar results for all end points, and we found no indication of pleiotropy (all P values ≥0.10 for the MR Egger intercept) or bias due to invalid instruments (Supplementary Fig. 7). Estimates for the known observational and causal association between high glucose level and MI is shown for the CCHS and CGPS combined, the UK Biobank cohort, and the CARDIoGRAMplusC4D Consortium as a positive control of study power and of the genetic instruments (Fig. 3 and Supplementary Fig. 7).
Conclusions
In this cohort from the Danish general population, we found observational and causal associations between elevated glucose level and high risk of retinopathy, peripheral neuropathy, and diabetic nephropathy. Observationally, we found positive associations between high random plasma glucose level and risk of PAD and CKD (defined as eGFR <60 mL/min/1.73 m2); however, a causal association could not be confirmed for PAD and seemed to be refuted for an eGFR <60 mL/min/1.73 m2. These findings were validated by similar results from a two-sample MR analysis using summary-level data for fasting glucose levels from participants in the MAGIC (13) and end point data from the UK Biobank (12) and the CKDGen Consortium (14). Our findings indicate that a high glucose level within the normal range is an important causal risk factor for microvascular disease. Furthermore, our study replicates previous findings showing that a high glucose level within the normal range is a causal risk factor for MI (1), suggesting that in a general population, elevated glucose levels over time may be important in the development of both micro- and macrovascular disease.
Previous observational studies have shown that retinopathy, neuropathy, and signs of microvascular dysfunction are prevalent in individuals with prediabetes (23–25). Observational findings may be confounded by concomitant cardiometabolic risk factors such as obesity, hyperlipidemia, and hypertension. Our finding of a stepwise increase in the risk of vascular disease with increasing glucose levels within the normoglycemic range or higher support the idea that an elevated glucose level per se has a causal role in the pathogenesis of microvascular disease, as do levels below the diabetes cutoff. This is in line with the general understanding of the natural history of type 2 diabetes as a continuous process of declining β-cell function and increasing relative insulin deficiency, leading to a continuous increase in glucose that is initiated years before the diabetes threshold is reached (26). Randomized controlled trials have shown that lifestyle changes and treatment with glucose-lowering drugs can reduce the progression from prediabetes to diabetes (27–29). Recent 30-year follow-up data from a study of 577 Chinese individuals showed that lifestyle interventions in individuals with prediabetes reduce long-term risks of diabetes, a composite of microvascular complications, cardiovascular disease, cardiovascular mortality, and all-cause mortality (30). The effects of lifestyle intervention are not likely to be due to glucose lowering alone but to several beneficial metabolic effects. Our findings highlight the importance of early detection of glycemia and screening for prediabetes in asymptomatic individuals through the use of risk assessment tools—such as the one currently provided by the American Diabetes Association (www.diabetes.org/are-you-at-risk/diabetes-risk-test/) (15)—in order to prevent glycemic deterioration as early as possible. In addition to being a risk factor for diabetes and cardiovascular disease, our data suggest that an elevated glucose level below the diabetes cutoff is an important risk factor for microvascular disease. Screening for retinopathy, neuropathy, diabetic nephropathy, and additional risk factors such as obesity, hyperlipidemia, and hypertension might be indicated in individuals with prediabetes.
We found observational and causal associations between glucose level and diabetic nephropathy, but no causal association between glucose level and CKD (defined as an eGFR <60 mL/min/1.73 m2). These findings may be explained by the fact that these two end points represent different etiologies of kidney disease: Diabetic nephropathy is a microvascular disease with characteristic histopathological changes induced by hyperglycemia and is initially characterized by albuminuria, glomerular hyperfiltration, and—though only in late stages—reductions in eGFR (17,31). In contrast, CKD covers a variety of kidney abnormalities of different etiologies, and eGFR is the gold standard but an unspecific measure of a decline in kidney function. In many individuals with prediabetes or type 2 diabetes, CKD is probably attributable to a mix of several pathological changes leading to reduced kidney function, including diabetic nephropathy, glomerulopathies, previous episodes of acute kidney disease, and aging-related nephropathy with micro- and macrovascular components (31); hyperglycemia may not be the predominant causal factor. In line with this, the renoprotective effects of sodium–glucose cotransporter 2 inhibitors seen in clinical intervention trials do not seem to be mediated by lowering glucose, but by other mechanisms that target common pathways of CKD progression, regardless of etiology (31,32).
The mechanisms by which glucose contributes to the pathogenesis of macrovascular disease such as MI and PAD are not completely known. Mechanistic studies suggest that metabolic changes induced by hyperglycemia, together with increased insulin resistance and free fatty acids, accelerate the atherosclerotic process through increased oxidative stress in arterial endothelial cells, the formation of advanced glycation end products, and nonenzymatic glycation of LDL, apolipoproteins, and clotting factors, collectively resulting in vasoconstriction, inflammation, and thrombosis (33,34). These mechanisms are likely to be similar in coronary arteries and large peripheral arteries, and even though in this study we could not confirm a causal association between glucose level and PAD (defined on the basis of ICD codes), a previous study of individuals without diabetes reported a causal association between fasting glucose level and carotid intima-media thickness (35).
A potential limitation of MR is that the selected genetic variants have pleiotropic effects on other risk factors of the diseases studied (21). However, we did not find any associations between the one-sample MR genetic instrument and age, sex, BMI, smoking, hypertension, LDL cholesterol, alcohol intake, physical activity, years of education, or, for women, menopause, and we found no indications of directional pleiotropy using MR Egger regression in the two-sample MR. Also, even though our aim was to study individuals from the general population, the confirmation cohorts used in the two-sample MR consisted of selected samples that may not be representative of the general population and may thus not be completely comparable (participation rates: CCHS, 61%; CGPS, 43%; UK Biobank, 5%) (36).
A limitation to our study is that in CCHS and CGPS, glucose was measured from samples obtained from participants in a nonfasted state, and oral glucose tolerance tests were not performed. Consequently we could not classify individuals according to impaired fasting glucose or impaired glucose tolerance status. In the observational analyses, estimates were adjusted for time since the last meal, and in the genetic analyses the results were validated by using summary data for fasting glucose levels. Another limitation is that eGFR was calculated from a single measurement of plasma creatinine, which may lead to some misclassification of kidney disease. However, the prevalence of eGFR <60 mL/min/1.73 m2 among the CCHS/CGPS population was ∼10%, which corresponds well with prevalence estimates from other general populations (37). Also, we did not have access to albuminuria measurements, which could have strengthened the validity of the nephropathy end point.
Strengths of the study include the examination of a large number of individuals from a genetically homogenous general population, access to highly valid data from individual participants, no losses to follow-up, and the use of MR, which allowed us to assess potential causal effects of high glucose levels on the risk of disease, minimizing residual confounding and reverse causation. In addition, the similar results found by using the two-sample MR approach increase the generalizability and validity of the results, making them reproducible with a different genetic instrument and in another population. The two-sample approach consisted of three separate analyses: the conventional IVW analysis unadjusted for pleiotropy; an analysis that used MR Egger regression adjusting for directional pleiotropic effects; and a weighted median regression accounting for up to 50% of information coming from invalid or weak instruments (38,39). Risk estimates from these three two-sample MR analyses were largely in the same direction and showed no indication of pleiotropy, confirming the validity of the instrument (38,39). It is still possible, however, that the genetic instruments influence the risk of disease through unknown glucose-independent pathways that we could not control for. An assumption of two-sample MR analyses is that the two samples represent similar but nonoverlapping populations. Summary data used for the two-sample MR analyses in this study were from the MAGIC and the CKDGen Consortium, both general population cohorts of European ancestry (13,14), and the overlap between two samples was low (<4% of individuals participating in both consortia). The UK Biobank is a general population cohort comprising participants of mainly white and British descent, and is not a part of the MAGIC (12,13). The CCHS and the CGPS were not part of the MAGIC or the CKDGen Consortium.
In conclusion, in this cohort from the Danish general population, random plasma glucose levels within the normal range and higher were causally associated with high risks of retinopathy, neuropathy, diabetic nephropathy, and MI. A causal association could not be confirmed for PAD and seemed to be refuted for eGFR <60 mL/min/1.73 m2. The findings were validated with similar results by using summary-level data for fasting glucose levels from the MAGIC and end point data from the UK Biobank and the CKDGen Consortium. These findings suggest that elevated glucose levels should be identified as an important risk factor for micro- and macrovascular disease in the general population and that screening for microvascular disease may be recommended, along with screening for additional cardiovascular risk factors, in individuals with prediabetes.
Article Information
Acknowledgments. The authors thank the staff and participants of the Copenhagen General Population Study and the Copenhagen City Heart Study, and the participants of the Meta-Analyses of Glucose and Insulin-Related Traits Consortium (MAGIC), the CKDGen Consortium, the UK Biobank, and the CARDIoGRAMplusC4D Consortium for their generous participation. The authors also thank the consortia for making their data publicly available.
Funding. This study was supported by Det Frie Forskningsråd (the Danish Council for Independent Research) (#4183-00171B to M.B.).
The funder was not involved in the design, conduct, analyses of the study, or the interpretation of the results.
Duality of Interest. B.G.N. reports acting as a consultant to AstraZeneca, Sanofi, Regeneron, Akcea Therapeutics, Amgen, Kowa Company, Ltd., Denka Seiken Co., Ltd., Amarin Corporation, Novartis, Novo Nordisk, and Silence Therapeutics and has received lecture honoraria from Amgen, Kowa Company, Ltd., and Amarin Corporation. No other potential conflicts of interest relevant to this article were reported.
Author Contributions. F.E. researched data, performed the analyses, and wrote the manuscript. S.M. reviewed and edited the manuscript. A.T.-H. and B.G.N. collected data and reviewed and edited the manuscript. M.B. performed the analyses and reviewed and edited the manuscript. All authors discussed the manuscript. M.B. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Prior Presentation. Parts of this study were presented in abstract form at the 79th Scientific Sessions of the American Diabetes Association, San Francisco, CA, 7–11 June 2019.
Footnotes
This article contains Supplementary Data online at https://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc19-1850/-/DC1.
- Received September 17, 2019.
- Accepted January 23, 2020.
- © 2020 by the American Diabetes Association.
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 https://www.diabetesjournals.org/content/license.
References
In this Issue
