1. Introduction
Type 2 diabetes mellitus is a chronic, systemic metabolic disorder that typically arises from the interplay of genetic and long-term environmental factors. It stands as one of the most prevalent and consequential diseases within the endocrine system. [1,2] According to the “IDF Diabetes Atlas (10th edition)” report released in 2021, approximately 537 million adults aged 20 to 79 worldwide are afflicted with diabetes, constituting 1 in every ten adults. Projections foresee a rise to 643 million by 2030, with an anticipated escalation to 783 million by 2045.
Simultaneously, the global population is expected to grow by 20%, and within this timeframe, diabetes cases are projected to surge by 46%.[3,4] This notable developmental trajectory underscores the increasingly burdensome impact associated with diabetes.[5] The substantial increase in complications, encompassing neurological, cardiovascular, renal, and ocular issues, attributable to type 2 diabetes mellitus, has resulted in a heightened mortality rate among patients, positioning diabetes as the ninth leading cause of human mortality.[6–10]
In contemporary society, an individual’s socioeconomic status is often gauged by their educational attainment (EA). Numerous studies have demonstrated a clear association between higher EA and a reduced risk of various health issues, including stroke, type 2 diabetes mellitus, and liver diseases.[11–13] Despite this, the causal link between EA and complications of type 2 diabetes mellitus remains unclear. Additionally, it is widely acknowledged that obesity and unfavorable lifestyle habits play pivotal roles as risk factors for type 2 diabetes mellitus and its complications, [14–18] clinicians routinely advocate for maintaining a healthy weight and adopting positive lifestyle habits as fundamental measures to improve the prognosis of type 2 diabetes mellitus.
Expanding on existing observational clinical studies, our hypothesis posits that specific body weight indicators and lifestyle habits may act as mediating factors in the intricate relationship between EA and type 2 diabetes complications. In this regard, Mendelian randomization (MR) stands out as a method supporting causal inference in current observational data.[19] This approach employs single nucleotide polymorphisms (SNPs) closely associated with the exposure, identified through genome-wide association studies (GWAS), as instrumental variables to mitigate the impact of various unconsidered confounding factors. Utilizing the 2-step MR technique allows for an extensive exploration of potential mediating effects.[20] Hence, the primary objective of this study is to delve into the causal relationship between EA and the 4 complications of type 2 diabetes, unraveling potential mediating pathways.
2. Method
2.1. Study design
The objective of this study is to examine the influence of EA on type 2 diabetes with neurological complications (DNC), type 2 diabetes with ophthalmic complications (DOC), type 2 diabetes with peripheral circulatory complications (DPCC), and type 2 diabetes with renal complications (DRC) and to explore potential mediating factors. Initially, we applied the Univariable Mendelian Randomization (UVMR) method to analyze the causal connection between EA and DNC, DOC, DPCC, and DRC. Following this, we compiled a list of 25 potential risk factors from existing studies that could potentially serve as crucial mediators.
Subsequently, utilizing two-step MR to assess the distinct impacts of these mediating factors, we acquired genetic data related to type 2 diabetes complications from the Finn Gen public database.
2.2. Data sources for exposure, mediating factors, and outcomes
All the data utilized in this MR study were publicly available GWAS summary statistics (Table 1). Ethical approval and informed consent were obtained in the original studies. This secondary analysis used de-identified aggregate data, involved no new participant recruitment, and required no additional ethical approval or informed consent.
2.3. Exposure
The genetic instruments corresponding to EA were obtained from an extensive meta-analysis conducted by the Social Science Genetic Association Consortium (SSGAC), which involved 766,345 individuals of European descent. Following the 2011 International Standard for Education, EA was categorized and converted into years of schooling in the United States. The data were then standardized, with 1 standard deviation equivalent to 4.2 years of schooling.
Out of the initial 10,101,242 SNPs, independent SNPs significantly associated with education (P < 5 × 10−8), pairwise linkage disequilibrium r2 < 0.001, and a distance threshold >10,000 kb were selected as instrumental variables.
2.4. Mediating factors
Among the 27 candidate mediators, 6 pertain to adiposity traits (body mass index [BMI], waist-to-hip ratio (WHR), visceral adipose tissue, body fat percentage, waist circumference, childhood obesity), 5 are lipids (LDL-c, HDL-c, triglycerides, apolipoprotein B, total cholesterol), 2 are related to glucose metabolism (fasting glucose, fasting insulin), systolic blood pressure (SBP), (diastolic blood pressure [DBP]), 3 are urinary biomarkers (sodium in urine, potassium in urine, urinary albumin excretion), 3 are physical activity or sedentary behaviors (moderate to vigorous physical activity levels (MVPA), watching television (TV), computer using), 1 is a stress-related trait (major depression), 4 are smoking or dietary behaviors (smoking initiation, alcohol intake frequency, coffee intake, fizzy drink intake), and 1 is a socioeconomic factor (total household income).
Candidate mediators were screened using associations of EA with each mediator and of each mediator with at least 1 complication, together with the direction of the corresponding associations. Each mediator was evaluated separately using two-step MR. The resulting path estimates were interpreted as exploratory candidate mediating pathways.
2.5. Outcomes
This study acquired summary-level data for 4 complications of type 2 diabetes, namely DNC (911 cases/183,185 controls), DOC (2119 cases/183,185 controls), DPCC (1049 cases/183,185 controls), and DRC (1296 cases/183,185 controls), from the Finn Gen study, a nationwide GWAS study from Finland.[21] Distinct study projects do not ensure the absence of participant overlap with the SSGAC exposure GWAS or mediator GWASs. The available summary results do not allow participant overlap to be quantified. The renal endpoint was E4_DM2REN (Open GWAS ID finn-b-E4_DM2REN), with 1296 cases and 183,185 controls.
The Finn Gen DF5 endpoint specification, corresponding to the R5 endpoint with this case count, defines cases using International Classification of Diseases –10 E11.2 (stored as E112) or Finnish International Classification of Diseases –9 2503A in hospital discharge or cause-of-death records.[22] The DF5 control specification excludes E4_DIABETES. This is a registry diagnosis of type 2 diabetes mellitus (T2D) with renal complications, rather than a phenotype defined specifically by albuminuria, estimated glomerular filtration rate stages 3 to 5, end-stage renal disease, or dialysis. These clinical stages cannot be distinguished using this endpoint.
The comparison is not restricted to patients with T2D without renal complications and should not be interpreted as an estimate of progression within an established T2D cohort.
2.6. Statistical analysis
The MR analysis devised in this study must adhere to the following assumptions: in the UVMR analysis, instrumental variables (IVs) exhibit significant associations with EA; IVs solely impact complications of type 2 diabetes through exposure factors; IVs remain unaffected by any confounding factors.[23] Effect sizes will be denoted using odds ratios (ORs), β coefficients, or mediation proportions, accompanied by corresponding 95% confidence intervals (CIs).
2.7. Total effect of education on the four complications of type 2 diabetes
This study employed the UVMR method to assess the overall impact of a 1-standard-deviation increase in genetically predicted education on the risk of 4 complications of type 2 diabetes. The primary analytical approach utilized was the random-effects inverse variance weighted (IVW) method.
2.8. Mediation MR analysis
The indirect effect for each candidate mediator was calculated as the product of the EA-to-mediator estimate (β1) and the mediator-to-outcome estimate (β2), and divided by the total EA-to-outcome effect to obtain the mediation proportion.[24,25] These separately estimated proportions may capture overlapping pathways and should not be added or interpreted as mutually adjusted independent contributions. The reported mediation intervals could not be independently reconstructed during revision.
2.9. Sensitivity analysis
This study conducted various sensitivity analyses, including MR Egger, weighted median, weighted mode, simple mode, and Mendelian randomization pleiotropy residual sum and outlier to validate the accuracy of IVW results in UVMR analysis. Additionally, the UVMR Egger method was employed to verify the accuracy of IVW results in UVMR analysis. The MR Egger method allows intercept-free estimation, serving as a measure of directional pleiotropy bias to some extent.[26] To detect and correct for any outlier SNPs reflecting potential levels of heterogeneity bias in MR causal estimates, the Mendelian randomization pleiotropy residual sum and outlier method was applied.
This study employed MR Egger’s intercept test to assess pleiotropy. In the absence of horizontal pleiotropy, the genetic instrument’s effect on the disease can be attributed exclusively to the exposure factor. Furthermore, Cochran’s Q test assesses heterogeneity between instruments.[27] Causal relationships between IVW estimates and at least 1 sensitivity analysis were confirmed only when they exhibited the same direction and statistical significance, with no evidence of pleiotropy (P > .05).
All MR analyses were performed using the “Two Sample MR” and “MR” R packages in R 4.3.1.
3. Result
3.1. Total effect of education on the four complications of type 2 diabetes
Based on the IVW analysis results from data obtained from Finn Gen, the prediction from a genetic perspective reveals that every 4.2-year increase in EA reduces the risk of 4 complications in type 2 diabetes mellitus patients. Specifically, the risk reductions are as follows: DNC is 0.33 (0.20, 0.54), DOC is 0.57 (0.41, 0.80), DPCC is 0.60 (0.39, 0.93), and DRC is 0.47 (0.32, 0.69) (refer to Figure 1), with the directions of effects consistent with the IVW method observed in other testing methods. Moreover, the MR Egger intercept estimation method and Cochran’s Q test indicate that except for DOC, the other 3 diseases show no evidence of pleiotropy and heterogeneity in all analyses (refer to Table S1, Supplemental Digital Content 1).
Table S2, Supplemental Digital Content 2 indicates that the reverse MR analyses did not identify any reverse causal relationships between the 4 type 2 diabetes-related complications and EA, thereby confirming that subsequent mediation MR analyses can be conducted.
3.2. Effects of education on the intermediate factors
Among the 27 candidate intermediate factors, a total of 25 risk factors met the inclusion criteria and underwent mediation MR analysis (refer to Figure 2). Predicting from a genetic perspective, for every 4.2-year increase in EA, there is an association with lower BMI (odds ratio: 0.76; 95%CI: 0.72,0.81), WHR (0.88; 0.85,0.92), Visceral adipose tissue (0.73; 0.68,0.78), Body fat percentage (0.78; 0.75,0.81), Waist circumference (0.78; 0.74,0.81), Childhood obesity (0.60; 0.44,0.82), Body fat percentage (−0.26; −0.28, −0.24), Waist circumference (−0.25; −0.27, −0.23), Triglycerides (0.81; 0.77,0.84), apolipoprotein B (0.96; 0.94,0.99), Fasting glucose (0.97; 0.95,1.00), Fasting insulin (0.95; 0.93,0.97), Sodium in urine (0.77; 0.74,0.80), SBP (0.14; 0.08,0.27), DBP (0.41; 0.28,0.62), MVPA (0.91; 0.88,0.94), Watching TV (0.67; 0.65,0.69), Major depression (0.79; 0.73,0.85), Smoking initiation (0.69; 0.65,0.74), Alcohol intake frequency (0.61; 0.58,0.64), and Fizzy drink intake (0.93; 0.90,0.96), while being associated with higher high-density lipoprotein cholesterol (1.24;1.20,1.29), Total cholesterol (1.04;1.01,1.07), Urinary albumin excretion (1.03; 1.01,1.05), Computer using (1.40; 1.36,1.44), Coffee intake (1.10; 1.08,1.13), and Total household income (1.92; 1.85,1.99). The directions of effects in other testing methods are also consistent with the IVW method (refer to Table S3, Supplemental Digital Content 3).
Except for Visceral adipose tissue, Childhood obesity, and Fizzy drink intake, there is heterogeneity in the association between EA and other intermediate factors (refer to Table S4, Supplemental Digital Content 4), but there is no evidence of pleiotropy for educational level and all intermediate factors (refer to Table S4, Supplemental Digital Content 4). Among the excluded candidate intermediate factors, 2 were excluded because they had no causal relationship with education, and 2 were excluded because the direction of the effects of education on the intermediates and the intermediates on the 4 complications of type 2 diabetes was the same.
3.3. Effects of intermediate factors on the four complications of type 2 diabetes
According to the UVMR results, an increased risk of the 4 complications of type 2 diabetes is associated with higher BMI, WHR, body fat percentage, total cholesterol, and fasting insulin. Additionally, an elevated waist circumference, fasting glucose, fasting insulin, and alcohol intake frequency are correlated with an increased risk of DNC. Childhood obesity, fasting glucose, SBP, watching TV, and alcohol intake frequency are associated with an increased risk of DOC.
Childhood obesity, SBP, DBP, and watching TV are linked to a higher risk of DPCC. Additionally, fasting glucose, SBP, DBP, watching TV, alcohol intake frequency, and total household income are correlated with an elevated risk of DRC (Tables S5, S6, S7 and S8, Supplemental Digital Content 5). Sensitivity results for the separate mediator analyses are reported in Tables S9, S10, S11 and S12, Supplemental Digital Content 6.
Although Cochran’s Q test detected heterogeneity in the causal relationships between most intermediate factors and the 4 complications of type 2 diabetes, various sensitivity analysis methods consistently demonstrated directional agreement. However, directional agreement does not remove heterogeneity or establish independent mediation. MR Egger intercept regression indicated horizontal pleiotropy in the causal relationships between HDL-c, triglycerides, and the 4 complications of type 2 diabetes.
3.4. Mediating effects of intermediate factors on complications of type 2 diabetes
The reported mediation proportion point estimates varied by outcome (Figure 3 and Table S13, Supplemental Digital Content 7). For DPCC, watching TV and waist circumference had point estimates of 85.1% and 81.8%, respectively, but both reported intervals ranged from 0% to 99.8%. These separate estimates may capture overlapping pathways, cannot be added, and do not establish that either factor contributes independently or more strongly than another mediator.
BMI, waist circumference, and watching TV are highlighted as candidate pathways rather than a definitive ranking.
4. Discussion
In this 2-step MR study, we uncovered new evidence indicating that, based on genetic predictions, each additional 4.2 years of school education is causally associated with a 33%, 57%, 60%, and 47% reduction in the risk of DNC, DOC, DPCC, and DRC, respectively. Furthermore, we identified candidate risk factors that may lie on pathways from education to the 4 complications of type 2 diabetes. The reported mediation proportion point estimates for BMI were 31.8%, 52.1%, 73.8%, and 39.1% for the respective complications.
Our study illuminates the causal protective effects of education on DNC, DOC, DPCC, and DRC, suggesting candidate mediating pathways involving physical obesity characteristics, lipids, glucose metabolism, and lifestyle habits. These exploratory findings provide evidence for the prevention and treatment of complications associated with type 2 diabetes.
Education constitutes a pivotal element in the landscape of social inequality. Numerous observational studies conducted across diverse countries consistently highlight a strong correlation between higher EAs and reduced risks of mortality, cardiovascular diseases, and metabolic disorders.[28,29] The relationship between education and all-cause mortality in adults exhibits a dose-response pattern, with an average risk reduction of 1.9% for each additional year of education (95% uncertainty interval 1.8–2.0). Specifically, for adults aged 18 to 49, each extra year of education correlates with an average mortality risk reduction of 2.9% (2.8–3.0), whereas for adults aged 70 and above, the average reduction in mortality risk is 0.8% (0.6–1.0).[30]
In the context of type 2 diabetes, multiple studies underscore the association between higher EAs and a diminished risk of developing the condition.[31] This observation is further substantiated by a MR study, which suggests a 47% risk reduction for each 1-standard deviation increase in education (equivalent to 4.2 years).[32] Notably, a nationwide cohort study conducted in Denmark by Emma Neble Larsen et al[33] discovered that lower educational levels during childhood are linked to an elevated risk of complications in type 2 diabetes (HR_low, 1.14; 95% CI, 1.05 to 1.24; HR_medium, 1.11; 95% CI, 1.05–1.17). The research by Yi-Shu Liao et al[34] also demonstrates a gradual decrease in the incidence of type 2 diabetes complications with increasing EAs. Moreover, 2 cross-sectional studies from Japan and Denmark highlight lower EAs as significant risk factors for diabetic retinopathy.[35,36] Consistent with our research findings, a cohort study by Doris Winitzki et al[37] identified a higher risk of diabetic nephropathy in individuals with low educational levels (odds ratio 1.65, 95% CI: 1.36–2.0).
On the other hand, education not only exerts a comprehensive impact on the complications of type 2 diabetes but also facilitates individuals’ access to higher-quality healthcare services. [38] Through a two-step MR approach, this study has explored candidate intermediary factors in the relationship between education and the 4 types of complications in type 2 diabetes. These factors predominantly encompass physical obesity characteristics, partially involve chemical obesity characteristics, and include features related to sugar metabolism and lifestyle habits, whose independent mediating roles remain uncertain. Notably, these findings align with recognized risk factors for complications in type 2 diabetes, encompassing obesity, lipid abnormalities, sugar metabolism irregularities, and unhealthy lifestyles; [17] these factors may share common metabolic pathways.
Various studies suggest that obesity can instigate a chronic inflammatory state by elevating pro-inflammatory factors such as TNF-α, thereby impeding insulin action and prompting inflammatory diseases across multiple organs, ultimately progressing into complications associated with diabetes.[39–41] Intriguingly, despite ample observational evidence supporting the notion that coffee consumption can reduce the incidence risk of complications in type 2 diabetes,[42] no causal relationship was identified in this study. This discrepancy may stem from the susceptibility of observational studies to diverse confounding factors.
Watching TV may reflect a broader lifestyle and socioeconomic cluster, rather than physical inactivity alone. Genetic evidence links leisure sedentary behaviors with education and BMI,[43] providing context for potentially overlapping behavioral and adiposity pathways. Associations with dietary habits and socioeconomic circumstances are plausible explanations, but were not established as mechanisms in this study.
Sedentary behavior and MVPA describe different dimensions of behavior and can coexist. Differences in instruments, measurement, the total-effect denominator, and uncertainty may contribute to the larger watching-TV point estimates.
UKPDS 33 and ADVANCE showed benefits of intensive glucose control for microvascular outcomes in T2D.[44,45] Our analysis included fasting glucose and fasting insulin, but not glycated hemoglobin and therefore cannot compare glycated hemoglobin with BMI or watching TV. Adiposity, insulin resistance, glycemia, and sedentary behavior may share pathways. Differences between their mediation proportion point estimates do not establish a hierarchy of treatment benefits, particularly when intervals are wide.
Genetically predicted lifetime differences also differ from the effects of glucose-lowering or behavioral interventions. These results do not diminish the clinical value of glycemic control.
One notable strength of this study lies in the utilization of SNPs as genetic instruments, minimizing biases stemming from confounding and reverse causation to the greatest extent possible. We leveraged the latest large-scale GWAS data to identify significantly associated SNPs, conducted precise MR analyses, and validated the reliability of the results through various sensitivity analyses. We initially selected widely recognized and significant risk factors as candidate mediators based on existing literature, establishing stringent screening criteria to identify trustworthy intermediate factors.
However, it is important to acknowledge potential limitations in this study. Firstly, it cannot fully elucidate the mediating effects of education on the 4 complications of type 2 diabetes, as certain risk factors, such as age and gender, were not included in the analysis. Secondly, a majority of the GWAS data utilized in this study were derived from European populations, necessitating caution when extrapolating the findings to other ethnic groups.
Lastly, sample overlap between GWAS studies may introduce bias, potentially causing MR estimates to lean toward observational association estimates.[46] The available GWAS summary information was insufficient to support a verified joint multivariable Mendelian randomization analysis, and multiple mediators were not analyzed jointly. Correlation between EA and a mediator, and among adiposity and behavioral mediators, may produce overlapping pathways and collinearity. In a joint model, collinearity can reduce conditional instrument strength, increase standard errors, and make estimates unstable, even when instruments are strongly associated with each exposure separately.
Without joint estimates and conditional instrument-strength diagnostics, we cannot distinguish independent mediator contributions or establish their relative importance. The separately estimated mediation proportions are therefore exploratory and non-additive; a sum above 100% does not represent a jointly mediated fraction. Participant overlap cannot be excluded or quantified for exposure-outcome, exposure-mediator, or mediator-outcome comparisons.
SSGAC pooled multiple cohorts, so differing project names or countries are insufficient to establish independence. Overlap may increase weak-instrument bias and inflate type I error; no verified filtering or overlap correction can be claimed from the retained records.[46] The renal endpoint does not distinguish early albuminuria from advanced chronic kidney disease or dialysis, and its controls are not a T2D-only comparison group. Unverifiable supplementary P values and mediation interval boundaries further limit the certainty of specific pathways.
In conclusion, this MR study systematically investigated the causal relationship between EAs and the occurrence of DNC, DOC, DPCC, and DRC. The study explored candidate mediating pathways, including obesity, lipid profiles, glucose metabolism-related traits, and lifestyle habits. Our findings contribute valuable insights into the identification of risk factors associated with DNC, DOC, DPCC, and DRC, offering new evidence to guide government initiatives aimed at strengthening school education.
Further validation is needed before clinical practitioners can target the candidate mediators in the pathway from education to the 4 complications of type 2 diabetes, potentially alleviating the societal burden of type 2 diabetes complications.
Acknowledgments
We are grateful to the authors and participants of all GWASs for the summary statistics used.
Author contributions
Formal analysis: Wei Tao Han.
Methodology: Wei Tao Han.
Software: Wei Tao Han.
Visualization: Wei Tao Han.
Funding acquisition: Qin Wan.
Validation: Qin Wan.
Writing – review & editing: Qin Wan.
Writing – original draft: Wei Tao Han.