Introduction

Diabetic retinopathy (DR) is a common microvascular complication of diabetes and is one of the main causes of visual impairment and blindness in adults over 501. DR affected an estimated 103 million adults worldwide in 2020, with the number expected to increase to approximately 160 million by 20452. DR has become a critical public health problem due to the increasing global burden of disease3. Although the application of laser photocoagulation and intravitreal anti-vascular endothelial growth factor therapy has achieved satisfactory clinical results in the treatment of severe DR, some limitations still exist, such as a high cost and invasive administration route4,5,6. Thus, the identification of modifiable risk factors is crucial to prevent the onset and development of DR and reduce the economic and social burden.

Diabetes mellitus (DM) is one of the most common metabolic disorders. Dietary approaches are a basic part of DR management, which may be beneficial to prevent complications of DM7. Caffeine is the most widely used active pharmacologic substance worldwide8, and coffee, tea, and soft drinks are predominant sources of caffeine9. Some studies have shown appropriate caffeine intake may be associated with a low risk of DM and cardiovascular diseases10,11, while some studies worried about the side effects of caffeine on human health12,13,14. Moreover, many studies have investigated the effect of caffeine on retinal vascular diseases15,16. Two recent studies reported the protective effect of self-reported caffeine intake for DR in patients with DM17,18. However, previous studies have examined the association between coffee consumption and DR, and the results were controversial19,20. These studies relied on self-reported coffee or caffeine use, and this may cause unavoidable recall bias. The concentrations of urinary caffeine and caffeine metabolites are potential biomarkers of caffeine intake21. Therefore, this study aims to evaluate the relationship between urinary caffeine and caffeine metabolites and DR in patients with DM using a US representative sample of the National Health and Nutrition Examination Survey (NHANES).

Methods

Study design and population

This cross-sectional study used data from the NHANES, a cross-section series of interviews and examinations of the civilian, noninstitutionalized US population conducted by the National Center for Health Statistics (NCHS) of the Centers for Disease Control and Prevention. NHANES used a stratified multistage sampling methodology with a weighting scheme to accurately assess the health and nutritional status among the US individuals22. NHANES aims to provide the US de-identified health statistics made publicly available. The NHANES was conducted in accordance with the ethical principles of the Declaration of Helsinki. Written informed consent was collected from all participants. The NCHS research ethics review board reviewed and approved the protocols.

Definition of DM and DR

According to the previous studies23,24, DM was defined as serum fasting glucose ≥ 126 mg/dl, 2-hour post-load glucose after oral glucose tolerance test (OGTT) ≥ 200 mg/dl, glycated hemoglobin (HbA1c) ≥ 6.5%, self-reported insulin or diabetes pill use, or self-reported DM.

DR was identified from the question “Has a doctor ever told you that diabetes has affected your eyes or that you had retinopathy”25,26.

Measurements of urinary caffeine and caffeine metabolites

Urine samples for participants in this study were collected after a 9-hour overnight fast during the morning session27. The high-performance liquid chromatography-electrospray ionization-tandem quadrupole mass spectrometry (HPLC–ESI–MS/MS) and internal standards labeled with stable isotopes were used to measure urinary caffeine and caffeine metabolites28. Caffeine and 14 caffeine metabolites were examined, including 1-methyluric acid (1-MU), 3-methyluric acid (3-MU), 7-methyluric acid (7-MU), 1,3-dimethyluric acid (1,3-DMU), 1,7-dimethyluric acid (1,7-DMU), 3,7-dimethyluric acid (3,7-DMU), 1,3,7-trimethyluric acid (1,3,7-TMU), 1-methylxanthine (1-MX), 3-methylxanthine (3-MX), 7-methylxanthine (7-MX), 1,3-dimethylxanthine (1,3-DMX), 1,7-dimethylxanthine (paraxanthine, 1,7-DMX), 3,7-dimethylxanthine (theobromine, 3,7-DMX), 1,3,7-trimethylxanthine (caffeine, 1,3,7-TMX), and 5-acetylamino-6-amino-3-methyluracil (AAMU). In cases where the result was below the limits of detection (LODs), the value for that variable is the LOD divided by the square root of two28. In order to control the concentration dilution of urine, urinary caffeine and its metabolites levels were calibrated with urine creatinine for further analysis.

Covariates

The following covariates were included to control the potential effects of confounding factors: Age, sex (male or female), race/ethnicity (non–Hispanic White, non–Hispanic Black, or others), body mass index (BMI, kg/m2), urine creatinine (mg/dl), smoking (smoking at least 100 cigarettes in life was considered as smokers), and drinking (having at least 12 alcohol drinks per year was considered as alcohol users), HbA1c, self-reported hypertension, and total energy intake (kcal/day). Duration of DM was calculated as the age at the interview minus the age of individuals when first told had DM (< 10 or ≥ 10 years).

Statistical analysis

All statistical analyses in this study were performed according to the guidelines of CDC for the complex oversampling of data with sample weight. Basic characteristics of the study population are reported as means ± standard deviation (SD) or percentages. The t-test and design-adjusted Rao-Scott Pearson χ2 were used to compare baseline characteristics between different groups. Spearman correlation coefficient analysis was conducted to show the pairwise correlations among these 14 metabolites as well as dietary caffeine intake, we did not adjust for multiple testing because the results were descriptive. To avoid the influence of the skew distribution of urinary caffeine and its metabolites, the natural logarithm-transformed levels were added into the model as continuous variable, and the original concentration was also analyzed in medians (low urinary level: ≤50%; high urinary level: >50%). Logistic regression models were used to calculate odds ratios (OR) and 95% confidence intervals (CI). Crude model was not adjusted for any covariate; model 1 was adjusted for age, sex, and race; based on model 1 model 2 was further adjusted for BMI, smoking, alcohol drinking, duration of diabetes, hypertension, HbA1c, and total energy intake. Subgroup analysis was performed stratifying by sex, according to the previous study. All analyses were conducted with R software, version 4.3.2. A two-sided P value < 0.05 was considered statistically significant.

Results

Data collection

All data was collected from NHANES 2009–2014. A total of 50,974 participants were included in this study. We excluded those participants who were without DM (n = 47789). Participants who did not answer the question for the diagnosis of DR were excluded (n = 1030). Then, participants with missing data on urinary caffeine and its metabolites (n = 1536) and covariates (n = 105) [BMI (n = 10), smoking (n = 3), drinking (n = 46), hypertension (n = 1), diabetes duration (n = 4), total energy intake (n = 21), HbA1c (n = 20)], were excluded. Finally, 514 participants were enrolled in this study (Fig. 1). Supplementary Table S1 showed the difference between included and excluded participants with diabetes.

Fig. 1
figure 1

Selection of study population.

Full size image

Characteristics of the study population

The basic characteristics of the participants were presented in Table 1. The prevalence of DR was 16.9% (87/514) among individuals with DM. No significant variations were found in age, sex, race, BMI, smoking, alcohol drinking, hypertension, and total energy intake between two groups. DR patients had higher HbA1c and longer duration of DM. Spearman correlation analysis showed significantly positive correlations between metabolites, with correlation coefficients ranging from 0.30 to 0.96. (Fig. 2). Caffeine intake showed moderate correlations with nine caffeine metabolites including 1-MU, 1,3-DMU, 1,7-DMU, 1,3,7-TMU, 1-MX, dimethylxanthine (1,3-DMX), paraxanthine (1,7-DMX), caffeine (1,3,7-TMX), and AAMU (Spearman p = 0.52–0.57, P < 0.0001).

Table 1 Weighted characteristics of participants without and with DR.
Full size table
Fig. 2
figure 2

Spearman correlation coefficients among 16 substances. Abbreviations: 1-MU, 1-methyluric acid, 3-MU, 3-methyluric acid; 7-MU, 7-methyluric acid; 1,3-DMU, 1,3-dimethyluric acid; 1,7-DMU, 1,7-dimethyluric acid; 3,7-DMU, 3,7-dimethyluric acid; 1,3,7-TMU, 1,3,7-trimethyluric acid; 1-MX, 1-methylxanthine, 3-MX, 3-methylxanthine; 7-MX, 7-methylxanthine; 1,3-DMX, 1,3-dimethylxanthine; 1,7-DMX, 1,7-dimethylxanthine (paraxanthine); 3,7-DMX, 3,7-dimethylxanthine (theobromine); 1,3,7-TMX, 1,3,7-trimethylxanthine (caffeine); AAMU, 5-acetylamino-6-amino-3-methyluracil.

Full size image

Compared to females, males were prone to be smokers and drinkers. Males had higher energy intake, 1-MU, 1,3-DMU, 1-MX,7-MX, paraxanthine, and AAMU than females (Supplementary Table S2).

The association between urinary caffeine and caffeine metabolites concentrations and DR

After adjusting for age, sex, race, BMI, smoking, alcohol drinking, duration of diabetes, HbA1c, hypertension, and total energy intake, ln-transformed 1-MU (OR = 1.27, 95% CI, 1.05 to 1.56, P = 0.016) and AAMU (OR = 1.16, 95% CI, 1.01 to 1.34, P = 0.043) were associated with an increased DR risk (Table 2).

Table 2 Association between urinary caffeine and caffeine metabolites as continuous variables and DR.
Full size table

In the median analysis, compared to lower levels (≤ 50%), higher levels (> 50%) of 1,7-DMU (OR = 1.92, 95% CI, 1.20 to 3.09, P = 0.008), caffeine (OR = 2.00, 95% CI, 1.27 to 3.16, P = 0.004), and AAMU (OR = 1.48, 95% CI, 1.04 to 2.10, P = 0.029) were associated with an increased DR risk (Table 3).

Table 3 Association between urinary caffeine and caffeine metabolites and DR in median analysis.
Full size table

Subgroup analysis

Subgroup analysis was conducted stratified by sex. Ln-transformed 1-MU (OR = 1.48, 95% CI, 1.10 to 1.98, P = 0.012), 1,3,7-TMU (OR = 1.24, 95% CI, 1.04 to 1.49, P = 0.018) and caffeine (OR = 1.29, 95% CI, 1.03 to 1.60, P = 0.028) were associated with an increased DR risk in males but not in females (Table 4).

Table 4 Subgroup analysis stratified by sex.
Full size table

In the median analysis, compared to lower levels (≤ 50%), higher levels (> 50%) of 1,7-DMU (OR = 2.75, 95% CI, 1.33 to 5.70, P = 0.008), 1,3,7-TMU (OR = 2.26, 95% CI, 1.02 to 5.01, P = 0.044), caffeine (OR = 3.23, 95% CI, 1.53 to 6.82, P = 0.003), and AAMU (OR = 2.93, 95% CI, 1.16 to 7.40, P = 0.024) were associated with an increased DR risk in males but not in females (Table 5).

Table 5 Association between urinary caffeine and caffeine metabolites and DR stratified by sex in median analysis.
Full size table

Discussion

This cross-sectional study was the first to explore the association between urinary caffeine and its metabolites and DR risk. After adjusting for age, race, BMI, smoking, alcohol drinking, duration of diabetes, HbA1c, hypertension, and total energy intake, we found that high urinary levels of 1-MU, 1,7-DMU, 1,3,7-TMU, caffeine (1,3,7-TMX) and AAMU were associated with an increased risk of DR in US males with DM. Furthermore, the results of correlation analysis between caffeine metabolites and caffeine intake were strong, and these were consistent with previous studies29,30,31. However, all metabolites are analyzed separately in the different models, but as there is high correlation between the metabolites, it could be that a significant finding of one metabolite could be caused by increases in the other metabolites as well. As these correlations are very strong, it might be that not all three metabolites separately are associated with DR, but only 1,3,7-TMU which as very high correlations with both 1-MU and caffeine (1,3,7-TMX).

Clinical studies on the association between coffee or caffeine and DR risk in the DM patients are still scarce. A previous study has reported coffee intake was not associated with DR in Asians20. Another study also did not find significant association between coffee consumption and the risk of DM microvascular complications in Europeans32. In addition, coffee consumption was inversely correlated with the DR risk in Koreans under 65 years with DM, while this association was not observed over 65 years19. Another study reported intake of caffeine decreased DR risk using NHANES 2003–201817. Moreover, a study with a mean age of 60.4, revealed that high caffeine intake was non-linear rather than linear correlated with reduced DR risk18. Our results were inconsistent with previous studies. Some factors of study design may contribute to this difference. First, food frequency questionnaires and dietary recall are two main methods used to quantify the intake of caffeine/coffee. Despite some advantages such as convenience, these methods rely on self-reported data which lead to unavoidable recall bias33. In addition, coffee usually contains multiple dietary components such as milk and chocolate, and coffee making process varies in different regions that may confuse the real effects of caffeine34,35. Considering these limitations, this study collected urinary caffeine and its metabolites to reflect caffeine intake, and this method was validated useful before29,36,37. Second, the population existed race difference between our studies and previous studies, which focused on Asians19,20 or Europeans18,32. The metabolism, clearance, pharmacokinetics, and pharmacodynamics of caffeine are significantly affected by race38. Third, DR diagnosis was based on fundus images in several previous studies of the association between coffee/caffeine intake and DR risk18,19. The diagnosis of DR was based on self-reported in this study, and this may underestimate DR prevalence compared with the diagnosis from fundus images39.

The results of experimental studies on the effects of caffeine and caffeine metabolites on DR were controversial. Caffeine is a well-known adenosine receptor antagonist which can cause vasoconstriction and is transformed into three active metabolites including theobromine (11%), theophylline (5%), and paraxanthine (82%) in human liver, catalyzed by cytochrome P450 1A2 (CYP1A2)40. Many studies have revealed that caffeine can reduce retinal blood flow and vessel density including macular and peripapillary region41,42,43. The decrease of macular and capillary vessels density were reliable indicators for DR screening44,45, and this may be attributed to increased levels of vascular endothelia growth factors caused by the retinal ischemia and tissue hypoxia46. In addition, recent studies have reported the positive relationship between urinary caffeine and its metabolites and insulin resistance (IR) in the US population31, and IR plays a crucial role in DR development47,48. Two caffeine metabolites including theophylline and theobromine could decrease the angiogenic activity of mononuclear cells in DM patients with proliferative DR49. However, previous studies found that aminophylline (an analogue of theophylline) and theophylline could lead to the formation of active E-rosette in DR patients, indicating abnormal lymphocytes reactivity50,51. Besides, a recent vitro study revealed the protection effects of caffeine on diabetic macular edema, which is the leading cause of vision loss in patients with DR52. In addition, caffeine could inhibit endothelial cell migration in vitro and choroidal neovascularization in vivo16. However, the latest research reported that caffeine eyedrop did not improve either the permeability of retinal vascular or reactive gliosis in db/db mice18. Moreover, an analysis of plasma metabolomic profiling indicated that caffeine metabolism differed between non-DR and proliferative DR in the patients with DM53. Based on these controversial findings, further studies are needed to explore the potential protection effects and mechanisms of caffeine and its metabolites in the onset and development of DR.

In addition, the results of sex-based analysis showed the positive association was significant in males but not in females. Several potential reasons may explain this difference. First, males had higher intake of caffeine and therefore higher urinary caffeine and its metabolites in this study. Second, compared with females, males were prone to be older, smokers, and drinkers. Age and unhealthy lifestyles behaviors are risk factors for DR in patients with DM54. Third, estrogens are reported as competitive inhibitors of caffeine metabolisms and the use of exogenous estrogen may inhibit caffeine metabolism in older females55.

Several limitations existed in this study that should be taken into rigorous consideration when interpreting these results. First, the causality could not be determined ought to the design of a cross-sectional study. Second, DR diagnosis was based on the interview questionnaire rather than retinal images, and this may lead to recall bias and misclassification where other eye conditions may be mistaken for DR. Self-reported DR may underestimate the true prevalence of DR, as undiagnosed or asymptomatic cases are unlikely to be captured, and the severity of DR was not recorded. Third, the single measurement of urine sample provides only a snapshot of metabolite levels, which may not reflect habitual caffeine consumption. Additionally, fasting and the morning-only sample collection may not represent real-world patterns, limiting generalizability. Forth, this study may not be representative of the broader population with diabetes in NHANES due to the high rate of exclusions. Besides, although our findings indicate that the potential DR risk of consuming caffeine for older adults, more research is needed to provide dietary guidelines regarding the caffeine intake. Fifth, this study did not adjust genetic factors, such as CYP1A2 polymorphisms. Sixth, multiple testing was not performed in correlation analysis. However, although multiple testing corrects for number of independent tests and these metabolites are most likely not independent due to their intercorrelation.

Conclusions

This study indicated that high urinary levels of 1-MU, 1,3,7-TMU, and caffeine were associated with an increased risk of DR in US males with DM. Prospective studies are needed to verify these findings in the future.