GRAPHICAL ABSTRACT

Highlights

• HbA1c and HDL-C were identified as key biomarkers for DR.

• Mendelian randomization reveals causal links between biomarkers and DR progression.

• Cheese intake shows protective effects, while alcohol intake raises DR risk.

• Biomarker-based nomogram predicts DR risk with high accuracy and clinical utility.

• Dietary habits influence DR risk via biomarkers like HbA1c and HDL-C.

INTRODUCTION

Diabetic retinopathy (DR), a leading cause of vision loss worldwide, significantly impacts individual health and public healthcare systems, with approximately one-third of people with diabetes developing severe visual impairments that can lead to irreversible blindness [1,2]. The progression of DR is mainly driven by prolonged hyperglycemia, oxidative stress, and inflammation, causing retinal microvascular damage [3,4]. Although anti-vascular endothelial growth factor (anti-VEGF) therapies have significantly improved DR management, about 40% of patients do not respond adequately, highlighting the need for more diverse treatments such as steroid therapies and combination protocols [5-8]. The damage from DR progresses from non-proliferative diabetic retinopathy (NPDR), severe non-proliferative diabetic retinopathy (SNPDR) to the more severe proliferative diabetic retinopathy (PDR), which can cause significant vision loss [9]. Given the risks associated with diagnosing PDR through fluorescein angiography, it is crucial to explore noninvasive, reliable biomarkers for predicting PDR risk [10].

Blood glucose (GLU) and glycosylated hemoglobin (HbA1c) levels are considered the primary biomarkers of diabetes and are closely related to DR’s progression and severity [11-13]. Besides, in recent years, a series of biomarkers, such as blood proteins, blood metabolites, lipids like cholesterol and apolipoproteins, and inflammatory and angiogenic biomarkers, have been found to have a causal relationship with diabetes mellitus (DM) and DR [14-17]. Additionally, some studies have found significant associations between different dietary habits and food intake and the incidence and progression of DR [18]. For instance, a study found that a high intake of fat and saturated fatty acids may affect the development of DR, while no significant associations were found between DR prevalence and the intake of monounsaturated or polyunsaturated fatty acids [18,19]. A study found that optimal combinations of specific vitamins and certain medical foods have been identified to protect the retina and choroid, and recommendations have been developed for clinicians to support conventional therapy for DR [18]. However, no studies have yet investigated the causal relationships between different dietary habits, food intake, varying degrees of DR, and DR-related biomarkers.

Mendelian randomization (MR) is an analytical method used to establish causality between a gene product and an intermediate phenotype, effectively serving as a randomized clinical trial [20]. This approach leverages the fact that genetic variants are randomly assigned during gamete formation and conception, thus eliminating reverse causation and confounding bias [21]. MR has shown promise in elucidating connections between DR and various biomarkers. The UK Biobank (UKB) conducted laboratory tests on more than 30 serum and urine biomarkers and 226 dietary habits and intake for a cohort of over 480,000 unrelated individuals, collecting comprehensive phenotype and genotype data [22]. This study uniquely integrates MR analysis and cross-sectional data to address the causal relationships between biomarkers, dietary factors, and DR, a combination not previously explored in depth. This study pioneers the integration of MR and cross-sectional analyses to uncover causal links between biomarkers, dietary factors, and DR progression, addressing gaps in understanding dietary impacts on DR. In this study, we employed a systematic MR to explore the causal relationships between 35 biomarkers, 226 dietary intake, and various stages of DR. Subsequently, by utilizing data from the National Health and Nutrition Examination Survey (NHANES), we investigated the association between these biomarkers and DR, further validating their clinical significance. In addition, we conducted a retrospective study on the DR cohort at Tianjin Medical University Eye Hospital and developed an effective yet simple prediction tool to assess the roles of various biomarkers in the progression of DR.

METHODS

Study design

This study was carried out in two phases, as illustrated in Fig. 1. In the first phase, we employed MR to investigate the causal relationships between 35 biomarkers and DR, NPDR, SNPDR, and PDR (Fig. 1A). To ensure the validity of the MR analysis, three criteria had to be met: (1) The genetic variants used must show a significant association with the exposure; (2) the selected genetic variants serving as instrumental variables (IVs) must be independent of confounding factors that influence both the exposure and the outcome; and (3) there should be no horizontal pleiotropy, meaning the IVs should influence DR solely through the exposure pathway [23]. In the second phase, we conducted an observational study using data from the NHANES, which serves as a foundational resource for health monitoring among the United States civilian population (Fig. 1B). We verified the biomarkers identified as having a causal relationship with DR in the MR study within the NHANES dataset’s patient cohort, confirming a significant association between these biomarkers and DR. In the third phase, we collected data on DR patients from the Tianjin Medical University Eye Hospital between January 2018 and December 2023, all from China (Fig. 1C). In this cohort, we validated the biomarkers identified through MR analysis as having a causal relationship with DR. Additionally, we developed and internally validated a medication non-adherence risk nomogram to investigate the risk relationship between these biomarkers and DR to predict DR and its progression using a series of biomarkers [24].

Data source and study population in MR

The UKB is a large-scale prospective cohort study that recruited more than half a million individuals, aged 40 to 69, from 22 different assessment centers across the United Kingdom during the period from 2006 to 2010 [25]. To minimize the impact of ancestry-related confounding factors, we focused our analyses exclusively on unrelated individuals of European heritage [22]. Our study involved a thorough examination and detailed fine mapping of loci linked to 35 biomarkers within a cohort of 363,228 participants [26]. In this study, four DR phenotypes were utilized as outcome measures. The Genome-Wide Association Study (GWAS) of DR was derived from the FinnGen Release 9 (R9), which is a genotype dataset from Finnish biobanks including more than 500,000 individuals.

Instrument selection of MR

We implemented a meticulous method for choosing IVs by initially applying a P value threshold of >5E−08 to identify suitable single-nucleotide variants. To guarantee the inclusion of only independent variants in the MR analysis, we excluded any variants that were in linkage disequilibrium with the most significant single nucleotide polymorphisms (SNPs) (using a clumping r2 cut-off of 0.001 and a clumping window of 10,000 kb). In MR analysis, the inverse variance weighted (IVW) approach was primarily used to estimate causal effects by pooling data from genetic instruments. Additional methods such as the MR-Egger intercept test assessed horizontal pleiotropy. The F-statistic was calculated for each instrument, with values >10 indicating sufficient instrument strength.

MR analysis

In our two-sample MR study, we evaluated the impact of biomarkers on DR utilizing the IVW approach with a random-effects model, implemented in TwoSampleMR version 0.5.6 (https://mrcieu.github.io/TwoSampleMR/). The data harmonization and subsequent analyses were performed using the same version of the TwoSampleMR package. Heterogeneity was evaluated using the ‘mr_heterogeneity()’ function, with a heterogeneity P value (Q_pval) below 0.05 indicating significant heterogeneity. To detect directional pleiotropy, we applied the MR-Egger intercept test, considering pleiotropy present if the MR-Egger intercept significantly differed from zero (P<0.05). Each SNP’s strong association with exposure was confirmed using a calculated F-statistic, where F>10 was the threshold. In addition to the IVW method, we utilized MR-Egger, weighted median, and weighted mode methods for supplementary analyses to validate the findings. Furthermore, reverse MR analysis was performed on these biomarkers with causal associations to rule out bidirectional interactions between exposure and outcome.

Mediation analysis

The study proceeded with two-step MR to evaluate and quantify the mediating effects of selected mediators on the causal pathway between exposures and DR. The initial step involved using MR analysis to determine the causal impact (β1) of exposures on each mediator. Following this, the second step utilized MR analysis to assess the causal influence (β2) of each mediator on the risk of DR, factoring in adjustments for exposures. The contribution of each mediator to the overall effect was calculated by dividing the product of the mediation effect (β1×β2) by the total effect. The standard errors for these mediation effects were computed employing the delta method [27].

Data sources and study population of NHANES analysis

The NHANES is an ongoing, cross-sectional survey designed to assess the health and nutritional status of non-institutionalized civilians in the United States. It uses a multistage probability sampling approach to select a nationally representative sample. The survey includes household interviews, physical exams, and laboratory testing. Conducted by the National Center for Health Statistics (NCHS) under the Centers for Disease Control and Prevention, the sampling methodology and data collection processes have been detailed in previous publications. All analyses accounted for the complex survey design of NHANES, incorporating sampling weights, stratification, and clustering, ensuring that estimates are representative of the United States population. This rigorous approach enhances the external validity of the findings and supports their generalizability to the broader population. The study was approved by the NCHS Ethics Review Board (no. Protocol 2005-06), and all participants provided written informed consent prior to participation [28].

To investigate the association between biomarkers and DR subtypes, we used publicly available NHANES survey data from the 2005/2006 and 2007/2008 cycles. Initially, we identified a total of 7,081 participants aged 40 years and older. Subsequently, based on the results of the previous MR analysis, we identified high-density lipoprotein cholesterol (HDL-C) as the focus of our study to investigate their association with DR. We excluded 1,656 participants who lacked information on these biomarkers. After these exclusions, a final cohort of 5,425 participants was identified (Fig. 1). The assessment of retinal photographs, available on the NHANES website, was meticulously conducted by at least two experienced experts using a rigorous procedure for the diagnosis and classification of DR. Among them, there were a total of 670 DR patients and 4,755 controls, with 545 participants classified as NPDR, 97 as SNPDR and 28 confirmed as PDR. In the NHANES surveys, participants were required to provide written informed consent before enrollment.

Data sources and study population of the Tianjin cohort

This retrospective study collected patients from the Tianjin Medical University Eye Hospital, focusing on individuals who were diagnosed and treated for DR between March 2018 and March 2024. The inclusion criteria for DR patients were as follows: (1) the age range for participants is 18 to 80 years; (2) patients had a confirmed diagnosis of DR, verified through clinical examination and imaging results, such as fluorescein angiography, fundus photography and optical coherence tomography; and (3) inclusion may be limited to treatment-naive patients or those who have not received specific treatments (e.g., anti-VEGF injections, corticosteroids) within a defined period before enrollment.

Exclusion criteria were in one of the following points: (1) patients with a history of ocular trauma; (2) patients coexisting retinal or optic nerve conditions affecting visual function assessment; (3) patients who are unable to complete follow-up will be excluded from the study; (4) patients with severe visual impairment from other causes not related to DR are usually excluded, as this could affect the assessment of the study’s primary endpoints.

We identified a total of 1,889 DR participants and 745 Cataract participants (control group). Subsequently, we excluded 473 participants who lacked information on DR and some biomarkers like HDL-C and suffered from severe DM complications. After these exclusions, a final cohort of 2,161 participants were identified (Fig. 1). Among them, there were a total of 1,511 DR patients and 650 controls, with 914 participants classified as NPDR, 495 as SNPDR and 102 confirmed as PDR. In the Tianjin cohort, participants were required to provide written informed consent before enrollment.

Covariates in analysis

The regression models were adjusted for covariates that have been previously linked to biomarkers and DR, including age, sex, race/ethnicity (classified into five categories), and education level (grouped into three categories: less than high school education, high school diploma or equivalent, and more than high school education). Additional factors considered in the models included body mass index, history of hypertension, as well as other relevant indicators. Covariates were selected based on their known relevance to both biomarkers and DR risk, including demographic, metabolic, and lifestyle factors. Some covariates, such as diabetes duration, were excluded due to data availability issues.

Statistical analyses

In this study, we performed data analysis using EmpowerStats software (https://www.empowerstats.net) and logistic regression models to evaluate clinical outcomes. Baseline characteristics of the study population were categorized according to DR subgroups and described using statistical methods. Continuous variables were expressed as mean±standard deviation and analyzed through weighted linear regression models. To explore the association between biomarkers and DR, we conducted multivariate linear regression analysis, calculating beta coefficients and 95% confidence intervals (CIs). Three models were used for multivariate testing: model 1 without adjustments, model 2 adjusted for gender, age, and race, and model 3 adjusted for all relevant covariates. Smoothed curve fitting was performed with the same set of adjustments. A threshold effect analysis was employed to identify the relationship and potential inflection points between biomarkers and DR. The same statistical approach was applied to gender-specific subgroups. A significance level of P<0.05 was considered statistically significant. To reduce variability within the dataset, we employed a weighting method during analysis.

The predictive nomogram study

Research approval was obtained from Tianjin Medical University Eye Hospital Ethics Committee. Patients were recruited and collected from the Tianjin Medical University Eye Hospital, from March 2018 to March 2024, and they came from all over China, as described in the previous part. Statistical analysis was performed using the R software version 4.3 (R Foundation for Statistical Computing, Vienna, Austria).

The least absolute shrinkage and selection operator (LASSO) method, effective for high-dimensional data, was used to select key predictive features among risk factors in patients with DR [29]. Features with nonzero coefficients were chosen, and a multivariable logistic regression model was built incorporating these features [30]. Odds ratios (ORs) with 95% CIs and P values were calculated. Sociodemographic variables with P≤0.05 and all disease- and treatment-related variables were included in the model, which aimed to predict the risk of DR within the cohort [31-33]. Calibration curves were used to evaluate the accuracy of the risk nomogram, with a significant test indicating imperfect calibration [34]. Harrell’s C-index measured the model’s discrimination performance, and bootstrapping (1,000 resamples) was applied to obtain a corrected C-index [35]. Decision curve analysis assessed the clinical utility of the nomogram by calculating net benefits across different threshold probabilities, accounting for the balance between true positives and the potential harm of unnecessary interventions [36,37]. External validation was conducted using data from the NHANES and Tianjin cohorts. Calibration curves, decision curve analyses, and bootstrapping techniques confirmed the model’s predictive accuracy.

Ethics approval and consent to participate

Since the data adopted in this study were all publicly available data from the FinnGen database, UKB, NHANES (Protocol 2005-06), Tianjin Medical University Eye Hospital (2024KY13), all data-related studies were approved by their respective ethical review committees and received written informed consent from patients. Therefore, this study does not need additional ethics approval.

RESULTS

Causal effect of biomarkers with DR in MR

We first analyzed the causal relationship between 35 biomarkers and DR (which includes NPDR, SNPDR, and PDR) using MR analysis (Fig. 2A). Of the 35 biomarkers, under the criterion of P<0.05, the IVW method found that eight biomarkers were significantly associated with DR, including HbA1c, GLU, alanine aminotransferase (ALT), ALT/aspartate aminotransferase (AST) ratio, insulin-like growth factor 1 (IGF1), apolipoprotein A (ApoA), HDL-C, sex hormone-binding globulin (SHBG). Among them, AST/ALT ratio (OR, –0.346; P=6.52E-6), ApoA (OR, –0.185; P=1.65E-3), HDL-C (OR, –0.136; P=6.60E-3), and SHBG (OR, –0.144; P=2.82E-2) showed a significant negative causal relationship with DR, suggesting a protective role in the progression of DR, while HbA1c (OR, 0.645; P=1.81E-20), GLU (OR, 0.831; P=3.50E-9), ALT (OR, 0.443; P=9.39E-7), and IGF1 (OR, 0.151; P=9.03E-4) showed a positive causal relationship, indicating a potential increase in the risk of DR. Consistent results of HbA1c were found by MR analysis using other MR methods, including weighted mode, weighted median and MR-Egger. Similarly, the results for GLU were doubly validated in the weighted median method, and the results for ALT were doubly validated in MR-Egger and weighted median methods, the results for AST/ALT were doubly validated in the weighted median method, the results for IGF1 were doubly validated in the MR-Egger method, the results for ALT were doubly validated in MR-Egger and weighted median methods, the results for ApoA were doubly validated in MR-Egger and weighted mode methods, the results for HDL-C were doubly validated in weighted median and weighted mode methods. We did not find heterogeneity and horizontal pleiotropy in these results (Supplementary Data 1).

Causal effect of biomarkers with DR subtypes in MR

To further explore the causal relationships between these biomarkers and different types and stages of DR, we conducted MR studies on NPDR, SNPDR, and PDR separately (Fig. 2B-D), hoping to gain more support for the use of biomarkers in early diagnosis, typing, and an indication of disease progression in DR. Frankly, HbA1c and GLU exhibited a significant causal relationship in different forms of DR, and all were positive, suggesting that HbA1c and GLU have a significant effect on the risk and progression of DR. Furthermore, ALT also exhibited a significant positive causal relationship in NPDR, SNPDR, and PDR, similarly suggesting an important effect on DR. On the other hand, HDL-C, and ApoA showed a negative causal relationship in different types of DR, suggesting its potential protective role in the progression of DR. We did not find heterogeneity and horizontal pleiotropy in these results (Supplementary Data 1).

Baseline characteristics of population-based study from NHANES

In this investigation, a total of 5,425 adults aged over 40 were selected according to specific inclusion and exclusion criteria (Table 1, Supplementary Data 2). The average age of the participants of the control group was 59.33 years and of the NPDR, SNPDR, and PDR were 62.57, 61.60, and 62.96 years relatively. Of the control cohort, 49.32% were male and 50.68% were female and the gender of NPDR (57.25% male and 42.75% female), SNPDR (50.52% male and 49.48% female), and PDR (39.29% male and 60.71% female) cohort. Building on prior MR studies, we assessed the relationship between HbA1c, GLU, ALT, AST/ALT ratio, HDL-C, and DR within this demographic. Regarding biomarkers, significant differences were observed in clinical laboratory examinations between the DR group and the control group for AST/ALT, HDL-C, GLU, and HbA1c. We also found that these biomarkers showed significant differences among the different groups of DR. Specifically, the levels of HDL-C were significantly lower in the DR group compared to the control group, with the lowest levels observed in the late DR group (PDR). On the other hand, ALT, HbA1c, GLU levels were significantly higher in the DR group and reached their highest in the late group. The results from the NHANES study were consistent with the findings of the previous MR analysis.

Table 2 displays the outcomes from the multivariate regression analysis. Alkaline phosphatase and red blood cells (RBCs) were strongly associated with DR. Yet, this significant positive correlation turned insignificant in model 3 after adjustments for all the variables were made. However, AST/ALT, HDL-C, creatinine (CREA), and hemoglobin (HGB) exhibited a significant correlation with DR in all three models, displaying an important relationship in DR. These findings are consistent with the results from previous MR studies.

Baseline characteristics of population-based study from the Tianjin cohort

In this investigation, a total of 2,161 participants were collected according to specific inclusion and exclusion criteria (Table 3, Supplementary Data 3). The average age of the participants of the control group was 69.62 years and of the NPDR, SNPDR, and PDR were 57.91, 57.50, and 61.10 years relatively. Of the control cohort, 58.15% were male and 41.85% were female and the gender of NPDR (47.05% male and 52.95% female), SNPDR (47.88% male and 52.12% female), and PDR (51.96% male and 48.04% female) cohort. We assessed the relationship between biomarkers and DR within this demographic. Regarding biomarkers, significant differences were observed in clinical laboratory examinations between the DR group and the control group for white blood cells (WBC), RBC, ALT, AST, albumin (ALB), AST/ALT, urea, uric acid (UA), triglyceride (TG), low-density lipoprotein cholesterol (LDL-C), ApoA, apolipoprotein B, HDL-C, GLU, and HbA1c. However, some biomarkers showed no significant differences among the DR subgroups. Through comparisons between DR subgroups, we found that RBC, ALT, AST/ALT, HDL-C, and ApoA exhibited significant differences. In the meantime, we also found that the levels of HDL-C and ApoA were significantly lower in the DR group compared to the control group, with the lowest levels observed in the late DR group (PDR). On the other hand, the ALT level was significantly higher in the DR group and reached its highest in the late group. The results from the Tianjin cohort study were also consistent with the findings of the previous MR analysis.

Table 4 presents the results of the multivariate regression analysis. In models 1 and 2, AST/ALT and γ-glutamyltransferase (GGT) showed a strong association with DR. However, after adjusting for all variables in model 3, this positive correlation became insignificant. In contrast, HDL-C, ApoA, and RBC maintained a significant correlation with DR across all three models, indicating a consistent and important relationship with DR. These findings align with those reported in previous MR studies. After synthesizing the results from the NHANES cohort and the Tianjin cohort, we found that HDL-C had a significant causal relationship and association with DR in both populations as well as in the MR analysis.

Development of an individualized prediction model

In this study, the Tianjin cohorts were divided into DR and control groups. Among the demographic, disease, and characteristics, 17 features were identified as potential predictors based on the Tianjin cohort (Fig. 3A and B) and had nonzero coefficients in the LASSO regression model. The results of the logistic regression analysis among the use of age, gender, WBC, HGB, platelet, ALT, AST, AST/ALT, GGT, ALB, CREA, urea, UA, total cholesterol, TG, HDL-C, and LDL-C. The model that incorporated the above independent predictors was developed and presented as the nomogram (Fig. 3D).

Apparent performance of the risk nomogram in the cohort

The calibration curve for the risk nomogram, used to predict risk in DR patients, showed strong agreement within the cohort (Fig. 3C). The C-index for the prediction model was 0.833 (95% CI, 0.796 to 0.870), and bootstrapping validation confirmed this with a value of 0.787, indicating good discriminatory ability. The risk nomogram demonstrated strong predictive performance, reflecting its reliable capability in forecasting disease risk. Additionally, we used the receiver operating characteristic (ROC) curve to calculate the model’s true positive rate and false positive rate, with the final area under the curve (AUC) being 0.814, demonstrating the model’s high clinical value (Fig. 3E). The decision curve analysis for the nomogram is presented in Fig. 3F. The decision curve showed that if the threshold probability is >18% and <92%, respectively, using this nomogram to predict DR risk adds more benefit than the scheme. Within this range, net benefit was comparable with several overlaps, on the basis of the risk nomogram.

Causal effect of diets with DR subtypes in MR

To investigate the relationship between dietary habits and different subtypes of DR, we analyzed the causal relationship between 226 diets and DR and subtypes (Fig. 4). Of the 226 diets, under the criterion of P<0.05, the IVW method found that 14 diets were significantly associated with DR. Among them, cheese intake (OR, –0.494; P=6.21E-5), dried fruit intake (OR, –0.509; P=1.20E-2), chocolate sweet intake (OR, –0.427; P=2.20E-2), cereal intake (OR, –0.407; P=2.30E-2), squash intake (OR, –0.400; P=2.40E-2), white rice intake (OR, –0.601; P=2.40E-2), ham intake (OR, –0.372; P=2.60E-2), prune intake (OR, –0.729; P=3.40E-2), and latte (OR, –0.665; P=4.20E-2) showed a significant negative causal relationship with DR, suggesting a protective role in the progression of DR, while alcohol intake (OR, 0.257; P=2.00E-3), coffee intake (OR, 0.673; P=3.00E-3), boiled/baked potatoes intake (OR, 0.362; P=8.00E-3), whole egg intake (OR, 0.821; P=2.00E-2), and goat cheese intake (OR, 0.966; P=3.60E-2) showed a positive causal relationship, indicating a potential increase in the risk of DR.

In NPDR, cheese intake, vegetarian alternatives intake, Scotch eggs intake, sausage intake, unsalted nuts intake, white rice intake, flavored mile intake showed a significant negative causal relationship, and pure fruit/vegetable juice intake, soya dessert intake, boiled/baked potatoes intake, whole eggs intake, spirits intake showed a positive causal relationship, indicating a potential increase in the risk of NPDR. Similarly, certain foods showed significant positive or negative causal relationships in both SNPDR and PDR, suggesting that dietary habits may play an important role in the management and progression of DR (Supplementary Data 4-7).

The mediation effect and potential therapeutic role of diets in DR

We used mediation MR analyses to explore the role of biomarkers in the relationship between diets and DR (Table 5). In the previous MR analysis and cross-sectional studies across different populations, we identified that ALT, ApoA, HbA1c, and HDL-C play important roles in the onset and progression of DR. Additionally, we found a group of diets with significant causal effects on DR. To explore how these diets influence DR through the regulation of biomarkers, we conducted a study using the mediation MR. We found that cheese intake may exert a protective effect on DR and its various subtypes by reducing risk factors such as ALT and HbA1c and enhancing protective factors like ApoA and HDL-C. In contrast, alcohol intake increases the risk of DR by promoting risk factors such as HbA1c and ALT, while suppressing protective factors like ApoA and HDL-C. This suggests that alcohol intake may be an important risk factor for DR. Additionally, dried fruit intake can also act on these four biomarkers to exert a protective effect against DR, while chocolate sweet intake may protect through the promotion of ApoA and HDL-C (Supplementary Data 8-11).

DISCUSSION

In this study, we employed MR analysis alongside cross-sectional data to investigate the relationships between biomarkers, dietary habits, and DR. Our results reveal the pathogenic roles of biomarkers such as HbA1c, ALT, and HDL-C in the development of DR, as well as the protective and risk-modifying effects of certain diets. Furthermore, we explored the mediation effects and interactions between diet, biomarkers, and DR. This research is the first to integrate MR analysis with a cross-sectional study based on different populations including the NHANES database and clinical data, offering a comprehensive evaluation of the causal links and correlations between multiple exposures and DR. By validating findings from both approaches, we have increased the robustness of our results, suggesting that ALT, HbA1c, and GLU could serve as critical indicators of DR risk, while HDL-C and ApoA may act as protective factors. Moreover, certain dietary interventions could potentially serve as therapeutic strategies for managing DR.

DR is a major ocular complication of DM poses a significant global health challenge and occurs in about 30 to 40% of diabetic individuals [1,38]. In previous studies, researchers have regarded HbA1c as one of the most important biomarkers for diabetes [39]. However, the role of HbA1c as a biomarker in the development of DR, particularly at different stages of DR, remains a subject of debate. In one study, researchers found that high levels of HbA1c play an important role in the occurrence of DR [40]. However, in another study, it was found that the rapid reduction of HbA1c is not associated with the progression of mild or moderate NPDR [41]. In our MR study, we found a strong causal relationship between HbA1c and DR, as well as NPDR, SNPDR, and PDR. HbA1c is one of the most important risk factors for DR and its different stages. Subsequently, we conducted an analysis using clinical data from the NHANES database and the Tianjin cohort and similarly found a significant association between HbA1c and DR. HbA1c is a key risk factor for DR.

Diabetic patients often present with dyslipidemia, and HDL-C, a common marker in lipid profiles, has recently emerged as an important biomarker for various diabetic complications [42]. In a cross-sectional study, Costacou et al. [43] found that significantly elevated HDL-C levels were linked to a higher risk of coronary heart disease in individuals with long-term type 1 DM. Another recent study also suggested that HDL-C may be involved in the relationship between elevated HGB levels and kidney function in diabetes [44]. In an MR study, researchers found that HDL-C was inversely related to SNPDR and PDR [45]. However, few studies are focusing on the association between DR and HDL-C, and this relationship remains controversial. In this study, we first used MR analysis to explore the causal relationship between HDL-C and DR. Subsequently, we conducted a cross-sectional study to investigate the association between DR and HDL-C using clinical data from DR samples in the NHANES database and the Tianjin cohort. We found that HDL-C has a protective effect on the progression of DR. In addition to HDL-C, the role of ApoA in DR also remains controversial. A study on the DM population found that the levels of HDL and ApoA were significantly higher in the nonDR group compared to the DR group, with the lowest levels observed in the PDR group [46]. In another study, it was found that the ratio of HDL-C/ApoA was positively correlated with the risk of developing DR [47]. HDL-C and ApoA exert protective effects in DR by mitigating oxidative stress and inflammation. Their roles in cholesterol efflux and anti-inflammatory pathways are well-documented in metabolic and ocular diseases. In our study, we first used MR analysis and found a significant inverse causal relationship between ApoA and DR. This was further validated in a cross-sectional study of the Tianjin cohort, where we observed that ApoA levels in the control group were significantly higher than in the DR group, and decreased progressively with the severity of DR. Additionally, our MR analysis revealed a causal relationship between ALT and DR, suggesting that ALT may be a risk factor for DR. This was further validated through sample analyses from the NHANES database and the Tianjin cohort, where we found that ALT levels were higher in the DR group compared to the control group and increased progressively with the severity of DR within the DR group.

Nomograms are now widely used as prognostic tools in oncology and medicine due to their user-friendly interfaces, improved accuracy, and clear prognostic outcomes, aiding clinical decision-making [48]. Our study is the first to apply a nomogram to DR. We developed and validated a new prediction tool for assessing risk in DR patients using biomarkers as variables. By integrating demographic, and disease risk factors into a user-friendly nomogram, this tool enables individualized predictions of risk for DR patients. Internal cohort validation showed strong discrimination and calibration abilities, with a high C-index from interval validation, supporting the tool’s broad and accurate applicability due to its large sample size. In our model, factors such as age, gender, and various relevant biomarkers are assigned a score. By calculating each patient’s score, we can predict their risk of developing DR. Patients with a score above 215 have over a 90% risk of developing DR. Therefore, this model can be used in routine check-ups and assessments for DM patients, providing timely guidance on treatment and lifestyle adjustments to prevent the onset of DR and protect vision.

Many eye diseases are treatable and preventable, especially in their early stages, and lifestyle, including nutrition and physical activity, plays a crucial role, particularly for DR [49]. A Study suggests that consuming fish, seafood, plant-based foods, and fresh fruits and vegetables may inhibit increased vascular leakage and pericyte loss, while regulating VEGF protein levels in the retina, thereby suppressing the progression of DR. Eating fruits, vegetables, fish, and olive oil may be associated with a reduced risk of developing DR [50]. A study suggests that dietary polyphenols may slow the progression of retinal diseases and seem to indirectly control DR by activating metabolic pathways responsible for enhancing cellular antioxidant capacity, rather than through direct mechanisms [51]. Some scholars have suggested that the type of diet and vitamin intake will play an important role in the treatment and management of DR [52]. Our study used diet GWAS data from the UKB to perform MR analysis on DR. We identified a series of diets with positive or negative causal relationships with DR. Studies have found that cheese can effectively reduce the risk of type 2 DM and cardiovascular disease [53,54]. Our study found that cheese intake may reduce the risk of DR, possibly due to the anti-inflammatory effects of casein, whey protein, and lactoferrin found in cheese [55]. The relationship between alcohol intake and DR remains controversial. Some studies have found that alcohol consumption appears to have a protective effect on diabetes complications [56], while other research has identified a significant positive correlation between alcohol intake and DR [19]. In our MR analysis, a significant positive causal relationship was found between alcohol intake and DR, suggesting that alcohol intake may be a risk factor for DR. It is recommended that DR patients strictly control their alcohol consumption. Practical dietary recommendations include consuming cheese and dried fruits weekly to support retinal health and reduce DR risk. Additionally, certain diets can influence the progression of DR by acting on biomarkers such as HDL-C, ApoA, and HbA1c, suggesting that dietary management should be emphasized in the care of DR patients. Additionally, our study found that the intake of dried fruit, chocolate, white rice, and cereal has a protective effect on different types of DR, while the consumption of coffee and whole eggs increases the risk of DR. These findings should provide guidance for the daily diet of DR patients.

The current study has several limitations. (1) As this study exclusively used GWAS data from the European population, it may not accurately represent other ethnicities or races worldwide. (2) In the cross-sectional study, despite our efforts to include a broad range of participants from NHANES, the number of participants was still limited, especially since the PDR group has only 28 patients, potentially leading to bias in the results. The PDR group in the Tianjin cohort has 102 patients, which is also a small proportion of the whole DR population. (3) Some covariates may not have been considered in the multiple regression analysis. (4) Despite efforts to identify and eliminate outlier variants, the potential for horizontal pleiotropy to affect the findings cannot be entirely ruled out. (5) A significant limitation of our study is the reliance on GWAS data predominantly derived from European populations and the use of NHANES data and the Tianjin cohort, which includes limited sample sizes for certain DR categories, particularly PDR. This reliance on a specific ethnic group limits the generalizability of our findings to other populations. Additionally, the small sample size for PDR (n=28 in NHANES and n=102 in the Tianjin cohort) may reduce statistical power and increase variability. Future studies should prioritize larger, diverse datasets to validate these findings. Therefore, our findings should be interpreted with caution, and future research should aim to validate these associations in larger, more diverse populations. This approach will help ensure that the results are applicable across different ethnicities and enhance the overall reliability of the study. Therefore, further validation of this study’s findings is necessary through multi-center epidemiological research and genetic engineering experiments, employing larger sample sizes and diverse populations.

In conclusion, using MR analysis, this study was the first to explore the causal impact of biomarkers on DR and DR subtypes from a genetic perspective, with findings validated by a cross-sectional study based on NHANES data and the Tianjin cohort. The results confirmed a causal association between HbA1c, GLU, ALT, HDL-C, ApoA, and DR. They also explored the protective effects and therapeutic potential of diets in DR. These findings support integrating dietary guidelines emphasizing HDL-C-enhancing and anti-inflammatory foods into public health strategies for DR prevention. However, the accuracy and validity of these findings require further verification through additional basic and clinical studies on DR.

SUPPLEMENTARY MATERIALS

NOTES

CONFLICTS OF INTEREST

No potential conflict of interest relevant to this article was reported.

AUTHOR CONTRIBUTIONS

Conception or design: X.C., D.W., X.L.

Acquisition, analysis, or interpretation of data: X.C., D.W., J.X.

Drafting the work or revising: X.C., D.W., X.L.

Final approval of the manuscript: all authors.

FUNDING

This study was funded by the Cambridge Trust, the Addenbrooke’s Charitable Trust, Cambridge University Hospitals, and the NIHR Cambridge Biomedical Research Centre (NIHR 203312). This research was supported by the National Natural Science Foundation of China (82171085) and the Tianjin Key Medical Discipline (Specialty) Construction Project (TJYXZDXK-037A).

ACKNOWLEDGMENTS

We sincerely thank the FinnGen consortium, the UK Biobank, and Open GWAS Project for publicly providing all the data for this MR analysis. We also sincerely thank the NHANES database and Tianjin Medical University Eye Hospital.

Fig. 1.

Overview of this study design. (A) The design of Mendelian randomization (MR) analysis. (B) The data of National Health and Nutrition Examination Survey (NHANES) cohort. (C) The data of the Tianjin cohort. SNP, single nucleotide polymorphism; HDL-C, high-density lipoprotein cholesterol; DM, diabetes mellitus.

Fig. 2.

The results of the causal relationship between biomarkers and diabetic retinopathy (DR). (A) The result of the causal relationship between biomarkers and DR. (B) The result of the causal relationship between biomarkers and non-proliferative diabetic retinopathy (NPDR). (C) The result of the causal relationship between biomarkers and severe non-proliferative diabetic retinopathy (SNPDR). (D) The result of the causal relationship between biomarkers and proliferative diabetic retinopathy (PDR). nsnp, numbers of single nucleotide polymorphism; se, standard error; pval, P value; OR, odds ratio; CI, confidence interval; HbA1c, glycosylated hemoglobin; MR, Mendelian randomization; IVW, inverse variance weighted; ALT, alanine aminotransferase; AST, aspartate aminotransferase; IGF1, insulin-like growth factor 1; HDL, high-density lipoprotein; SHBG, sex hormonebinding globulin; ALB, albumin; GGT, γ-glutamyltransferase.

Fig. 3.

Demographic and clinical feature selection using the least absolute shrinkage and selection operator (LASSO) binary logistic regression model to predict the risk of diabetic retinopathy. (A) LASSO coefficient profiles of the clinical features. (B) Optimal parameter (lambda) selection in the LASSO model. (C) Nomogram-predicted probability of non-adherence. (D) The non-adherence nomogram was developed in the cohort, with the use of clinical feathers and biomarkers. (E) Area under the curve (AUC) curve analysis for the non-adherence nomogram. (F) Decision curve analysis for the non-adherence nomogram. WBC, white blood cell; HGB, hemoglobin; PLT, platelet; ALT, alanine aminotransferase; AST, aspartate aminotransferase; GGT, γ-glutamyltransferase; ALB, albumin; CREA, creatinine; UA, uric acid; TC, total cholesterol; TG, triglyceride; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol.

Fig. 4.

The results of the causal relationship between diets and diabetic retinopathy (DR). (A) The result of the causal relationship between diets and DR. (B) The result of the causal relationship between diets and non-proliferative diabetic retinopathy (NPDR). (C) The result of the causal relationship between diets and severe non-proliferative diabetic retinopathy (SNPDR). (D) The result of the causal relationship between diets and proliferative diabetic retinopathy (PDR). nsnp, numbers of single nucleotide polymorphism; pval, P value; OR, odds ratio; CI, confidence interval; IVW, inverse variance weighted.

REFERENCES

• 1. Yau JW, Rogers SL, Kawasaki R, Lamoureux EL, Kowalski JW, Bek T, et al. Global prevalence and major risk factors of diabetic retinopathy. Diabetes Care 2012;35:556-64.PubMedPMC
• 2. Duh EJ, Sun JK, Stitt AW. Diabetic retinopathy: current understanding, mechanisms, and treatment strategies. JCI Insight 2017;2:e93751.ArticlePubMedPMC
• 3. Furino C, Boscia F, Reibaldi M, Alessio G. Intravitreal therapy for diabetic macular edema: an update. J Ophthalmol 2021;2021:6654168.ArticlePubMedPMCPDF
• 4. Kang Q, Yang C. Oxidative stress and diabetic retinopathy: molecular mechanisms, pathogenetic role and therapeutic implications. Redox Biol 2020;37:101799.ArticlePubMedPMC
• 5. Stitt AW, Curtis TM, Chen M, Medina RJ, McKay GJ, Jenkins A, et al. The progress in understanding and treatment of diabetic retinopathy. Prog Retin Eye Res 2016;51:156-86.ArticlePubMed
• 6. Sorour OA, Levine ES, Baumal CR, Elnahry AG, Braun P, Girgis J, et al. Persistent diabetic macular edema: definition, incidence, biomarkers, and treatment methods. Surv Ophthalmol 2023;68:147-74.ArticlePubMed
• 7. Xie X, Lian C, Zhang Z, Feng M, Wang W, Yuan X, et al. Aflibercept for long-term treatment of diabetic macular edema and proliferative diabetic retinopathy: a meta-analysis. Front Endocrinol (Lausanne) 2023;14:1144422.ArticlePubMedPMC
• 8. Nawaz IM, Rezzola S, Cancarini A, Russo A, Costagliola C, Semeraro F, et al. Human vitreous in proliferative diabetic retinopathy: characterization and translational implications. Prog Retin Eye Res 2019;72:100756.ArticlePubMed
• 9. Chaudhary S, Zaveri J, Becker N. Proliferative diabetic retinopathy (PDR). Dis Mon 2021;67:101140.ArticlePubMed
• 10. Wang M, Garg I, Miller JB. Wide field swept source optical coherence tomography angiography for the evaluation of proliferative diabetic retinopathy and associated lesions: a review. Semin Ophthalmol 2021;36:162-7.ArticlePubMed
• 11. Long M, Wang C, Liu D. Glycated hemoglobin A1C and vitamin D and their association with diabetic retinopathy severity. Nutr Diabetes 2017;7:e281.ArticlePubMedPMCPDF
• 12. Sabanayagam C, Khoo EY, Lye WK, Ikram MK, Lamoureux EL, Cheng CY, et al. Diagnosis of diabetes mellitus using HbA1c in Asians: relationship between HbA1c and retinopathy in a multiethnic Asian population. J Clin Endocrinol Metab 2015;100:689-96.ArticlePubMed
• 13. Brinchmann-Hansen O, Dahl-Jorgensen K, Sandvik L, Hanssen KF. Blood glucose concentrations and progression of diabetic retinopathy: the seven year results of the Oslo study. BMJ 1992;304:19-22.ArticlePubMedPMC
• 14. Zou X, Ye S, Tan Y. Potential disease biomarkers for diabetic retinopathy identified through Mendelian randomization analysis. Front Endocrinol (Lausanne) 2024;14:1339374.ArticlePubMedPMC
• 15. Busik JV. Lipid metabolism dysregulation in diabetic retinopathy. J Lipid Res 2021;62:100017.ArticlePubMedPMC
• 16. Kastelan S, Oreskovic I, Biscan F, Kastelan H, Gverovic Antunica A. Inflammatory and angiogenic biomarkers in diabetic retinopathy. Biochem Med (Zagreb) 2020;30:030502.PubMedPMC
• 17. Du X, Yang L, Kong L, Sun Y, Shen K, Cai Y, et al. Metabolomics of various samples advancing biomarker discovery and pathogenesis elucidation for diabetic retinopathy. Front Endocrinol (Lausanne) 2022;13:1037164.ArticlePubMedPMC
• 18. Shi C, Wang P, Airen S, Brown C, Liu Z, Townsend JH, et al. Nutritional and medical food therapies for diabetic retinopathy. Eye Vis (Lond) 2020;7:33.ArticlePubMedPMCPDF
• 19. Sasaki M, Yuki K, Hanyuda A, Yamagishi K, Motomura K, Kurihara T, et al. Associations between fatty acid intake and diabetic retinopathy in a Japanese population. Sci Rep 2023;13:12903.ArticlePubMedPMCPDF
• 20. Smith GD, Ebrahim S. ‘Mendelian randomization’: can genetic epidemiology contribute to understanding environmental determinants of disease? Int J Epidemiol 2003;32:1-22.ArticlePubMed
• 21. Swerdlow DI, Kuchenbaecker KB, Shah S, Sofat R, Holmes MV, White J, et al. Selecting instruments for Mendelian randomization in the wake of genome-wide association studies. Int J Epidemiol 2016;45:1600-16.ArticlePubMedPMC
• 22. Bycroft C, Freeman C, Petkova D, Band G, Elliott LT, Sharp K, et al. The UK Biobank resource with deep phenotyping and genomic data. Nature 2018;562:203-9.ArticlePubMedPMCPDF
• 23. Wu F, Huang Y, Hu J, Shao Z. Mendelian randomization study of inflammatory bowel disease and bone mineral density. BMC Med 2020;18:312.ArticlePubMedPMCPDF
• 24. Wang H, Zhang L, Liu Z, Wang X, Geng S, Li J, et al. Predicting medication nonadherence risk in a Chinese inflammatory rheumatic disease population: development and assessment of a new predictive nomogram. Patient Prefer Adherence 2018;12:1757-65.ArticlePubMedPMCPDF
• 25. Sudlow C, Gallacher J, Allen N, Beral V, Burton P, Danesh J, et al. UK Biobank: an open access resource for identifying the causes of a wide range of complex diseases of middle and old age. PLoS Med 2015;12:e1001779.ArticlePubMedPMC
• 26. Sinnott-Armstrong N, Tanigawa Y, Amar D, Mars N, Benner C, Aguirre M, et al. Genetics of 35 blood and urine biomarkers in the UK Biobank. Nat Genet 2021;53:185-94.ArticlePubMedPMCPDF
• 27. Kong L, Ye C, Wang Y, Zheng J, Zhao Z, Li M, et al. Causal effect of lower birthweight on non-alcoholic fatty liver disease and mediating roles of insulin resistance and metabolites. Liver Int 2023;43:829-39.ArticlePubMedPDF
• 28. Centers for Disease Control and Prevention. The National Health and Nutrition Examination Survey (NHANES) analytic and reporting guidelines. Atlanta: CDC; 2006.
• 29. Sauerbrei W, Royston P, Binder H. Selection of important variables and determination of functional form for continuous predictors in multivariable model building. Stat Med 2007;26:5512-28.ArticlePubMedPDF
• 30. Kidd AC, McGettrick M, Tsim S, Halligan DL, Bylesjo M, Blyth KG. Survival prediction in mesothelioma using a scalable LASSO regression model: instructions for use and initial performance using clinical predictors. BMJ Open Respir Res 2018;5:e000240.ArticlePubMedPMC
• 31. Xing J, Min L, Zhu S, Zhang H, Zhao Y, Li H, et al. Factors associated with gastric adenocarcinoma and dysplasia in patients with chronic gastritis: a population-based study. Chin J Cancer Res 2017;29:341-50.ArticlePubMedPMC
• 32. Balachandran VP, Gonen M, Smith JJ, DeMatteo RP. Nomograms in oncology: more than meets the eye. Lancet Oncol 2015;16:e173-80.ArticlePubMedPMC
• 33. Iasonos A, Schrag D, Raj GV, Panageas KS. How to build and interpret a nomogram for cancer prognosis. J Clin Oncol 2008;26:1364-70.ArticlePubMed
• 34. Kramer AA, Zimmerman JE. Assessing the calibration of mortality benchmarks in critical care: the Hosmer-Lemeshow test revisited. Crit Care Med 2007;35:2052-6.ArticlePubMed
• 35. Pencina MJ, D’Agostino RB. Overall C as a measure of discrimination in survival analysis: model specific population value and confidence interval estimation. Stat Med 2004;23:2109-23.ArticlePubMed
• 36. Vickers AJ, Cronin AM, Elkin EB, Gonen M. Extensions to decision curve analysis, a novel method for evaluating diagnostic tests, prediction models and molecular markers. BMC Med Inform Decis Mak 2008;8:53.ArticlePubMedPMCPDF
• 37. Huang YQ, Liang CH, He L, Tian J, Liang CS, Chen X, et al. Development and validation of a radiomics nomogram for preoperative prediction of lymph node metastasis in colorectal cancer. J Clin Oncol 2016;34:2157-64.ArticlePubMed
• 38. Ruta LM, Magliano DJ, Lemesurier R, Taylor HR, Zimmet PZ, Shaw JE. Prevalence of diabetic retinopathy in type 2 diabetes in developing and developed countries. Diabet Med 2013;30:387-98.ArticlePubMed
• 39. Gillery P. HbA1c and biomarkers of diabetes mellitus in Clinical Chemistry and Laboratory Medicine: ten years after. Clin Chem Lab Med 2022;61:861-72.ArticlePubMed
• 40. Poshtchaman F, Dehnabi A, Poshtchaman Z, Birjandi B. HbA1C, proliferative and non-proliferative retinopathy in diabetic patients. Med Clín Pract 2023;6:100371.Article
• 41. Simo R, Franch-Nadal J, Vlacho B, Real J, Amado E, Flores J, et al. Rapid reduction of HbA1c and early worsening of diabetic retinopathy: a real-world population-based study in subjects with type 2 diabetes. Diabetes Care 2023;46:1633-9.ArticlePubMedPDF
• 42. Romero-Aroca P, Verges R, Pascual-Fontanilles J, Valls A, Franch J, Barrot J, et al. Effect of lipids on diabetic retinopathy in a large cohort of diabetic patients after 10 years of follow-up. J Clin Med 2023;12:6674.ArticlePubMedPMC
• 43. Costacou T, Evans RW, Orchard TJ. High-density lipoprotein cholesterol in diabetes: is higher always better? J Clin Lipidol 2011;5:387-94.ArticlePubMedPMC
• 44. Williams AN, Conway BN. Effect of high density lipoprotein cholesterol on the relationship of serum iron and hemoglobin with kidney function in diabetes. J Diabetes Complications 2017;31:958-63.ArticlePubMedPMC
• 45. Li N, Zhang X, Zhang M, Wu L, Li C, Pan Y, et al. Associations of genetically determined lipid traits and lipid-modifying agents with the risk of diabetic retinopathy: a Mendelian randomization study. Atherosclerosis 2023;369:9-16.ArticlePubMed
• 46. Hosny SS, Nasr MS, Elfattah MM, Dayem SH, Ibrahim RH. Evaluating the relation between serum apolipoprotein B (apo B), apolipoprotein A (apo A) and apo B/apo A ratio with diabetic retinopathy in a sample of type 2 Egyptian diabetic patient. Egypt J Intern Med 2023;35:60.ArticlePDF
• 47. Zhang C, Lin W, Xu Q, Li H, Xu C, Ma X, et al. Association between high-density lipoprotein cholesterol to apolipoprotein A ratio and diabetic retinopathy: a cross-sectional study. J Diabetes Complications 2023;37:108471.ArticlePubMed
• 48. Wei L, Champman S, Li X, Li X, Li S, Chen R, et al. Beliefs about medicines and non-adherence in patients with stroke, diabetes mellitus and rheumatoid arthritis: a cross-sectional study in China. BMJ Open 2017;7:e017293.ArticlePubMedPMC
• 49. Rondanelli M, Gasparri C, Riva A, Petrangolini G, Barrile GC, Cavioni A, et al. Diet and ideal food pyramid to prevent or support the treatment of diabetic retinopathy, age-related macular degeneration, and cataracts. Front Med (Lausanne) 2023;10:1168560.ArticlePubMedPMC
• 50. Bryl A, Mrugacz M, Falkowski M, Zorena K. A Mediterranean diet may be protective in the development of diabetic retinopathy. Int J Mol Sci 2023;24:11145.ArticlePubMedPMC
• 51. Fanaro GB, Marques MR, Calaza KD, Brito R, Pessoni AM, Mendonca HR, et al. New insights on dietary polyphenols for the management of oxidative stress and neuroinflammation in diabetic retinopathy. Antioxidants (Basel) 2023;12:1237.ArticlePubMedPMC
• 52. Simo R, Hernandez C. What else can we do to prevent diabetic retinopathy? Diabetologia 2023;66:1614-21.ArticlePubMedPMCPDF
• 53. Gao D, Ning N, Wang C, Wang Y, Li Q, Meng Z, et al. Dairy products consumption and risk of type 2 diabetes: systematic review and dose-response meta-analysis. PLoS One 2013;8:e73965.ArticlePubMedPMC
• 54. Guo J, Astrup A, Lovegrove JA, Gijsbers L, Givens DI, Soedamah-Muthu SS. Milk and dairy consumption and risk of cardiovascular diseases and all-cause mortality: dose-response meta-analysis of prospective cohort studies. Eur J Epidemiol 2017;32:269-87.ArticlePubMedPMCPDF
• 55. Nieman KM, Anderson BD, Cifelli CJ. The effects of dairy product and dairy protein intake on inflammation: a systematic review of the literature. J Am Coll Nutr 2021;40:571-82.ArticlePubMed
• 56. Jia MJ, Li HL. Association between tea intake and alcohol consumption and diabetes complications: a two sample Mendelian randomization study. Curr Sci 2024;4:496-509.Article

Figure & Data

References

Citations

Citations to this article as recorded by  

About this article

Cui X, Wen D, Xiao J, Li X. The Causal Relationship and Association between Biomarkers, Dietary Intake, and Diabetic Retinopathy: Insights from Mendelian Randomization and Cross-Sectional Study. Diabetes Metab J. 2025 Mar 31. doi: 10.4093/dmj.2024.0731. Epub ahead of print.

Received: Nov 17, 2024; Accepted: Dec 12, 2024

DOI: https://doi.org/10.4093/dmj.2024.0731.

Download citation ↓