Diabetic retinopathy (DR) is a usual microvascular complication of type 2 diabetes mellitus (T2DM), one of the major reasons for blindness (1). According to The Global Burden of Disease Study, DR was the only cause of age-standardized vision loss to increase over the past three decades (2). Over 103.12 million adults worldwide were diagnosed with DR in 2020, and with the prevalence of diabetes increasing at an alarming rate, it is estimated that the world DR population will grow by 55.6%(57.4 million) between 2020 and 2045 (3). A prevalence-based cost-of-illness model estimates that Indonesia will spend $8.9 billion on the healthcare of DR in 2025 (4). As DR is often asymptomatic until the later, even more, severe stages, early diagnosis, and intervention are essential and more cost-effective for public health and healthcare costs (5–7).

DR prevalence has been discussed in some studies in different age groups of T2DM. The ADVANCE Collaborative group (8) has reported that the course of diabetes is independently related to the risk of microvascular complications, and diabetes duration has a more significant impact on younger people than on older people. Middleton et al. (9) have found that DR seems more susceptible in people diagnosed with T2DM in middle age (or with a younger present age), and the odds of DR decreased with increasing age at diagnosis. They considered this difference to be caused by reducing insulin-like growth factor 1 and growth hormone with increasing age. DR is more likely to occur in the middle-aged population after diagnosis of T2DM than in the elderly (10), so a more targeted prediction model and intervention strategy are needed.

Several prediction models have been applied to the identification and diagnosis of DR (11–13). However, these prediction models were constructed for almost all age groups. They have shortcomings in predicting the development of DR in different age groups. This will limit their ability to stratify individual patients according to risk level and select the optimal treatment. To our knowledge, there is a lack of predictive models developed separately for the middle-aged population. We suggested that developing a separate DR prediction model for the middle-aged age group and narrowing the prediction model orientation may be more important for applying the model for early identification and prevention of DR. We developed a model predicting the development of DR in middle-aged people with T2DM based on data from the National Health and Nutrition Examination Survey (NHANES), which may provide more personalized screening and treatment options for middle-aged T2DM patients.

Nomogram is a useful and reliable forecasting tool that can produce individual probabilities of endpoint events by combining different variables and quantifying the risk individually (19). In the risk predictor analysis of this study, gender, use of insulin, renal failure, duration of diabetes, UACR, blood urea nitrogen, and serum phosphorus were related to the risk of DR in midlife patients with T2DM. Based on this, we used statistical analysis to screen five of these variables to construct and validate a novel DR risk predictive tool for middle-aged patients with T2DM. The model showed that being male, taking insulin now, longer duration of diabetes, higher UACR, and lower serum phosphorus were critical factors in determining the risk of DR in patients with T2DM, which has the same part of risk factors as those reported in previous studies (10, 20). To make it more convenient for physicians to provide early individualized intervention for middle-aged T2DM patients, we have built an online, free prediction tool. According to Anne et al. (21), the AUC value of 0.7 or higher is considered acceptable or good for model discrimination. Our model presented good discrimination and calibration ability, offering a personalized prediction of DR incidence.

For patients with T2DM, the course of the disease is an unchangeable risk factor. Unlike T1DM, disease duration has a more significant impact on patients with T2DM combined with DR (22). Compared to older T2DM patients, midlife patients have a longer survival time and will be exposed to the increased risk of complications associated with a longer disease course. Elevated blood glucose and lipid metabolism disorders in T2DM patients caused pathological reactions, such as oxidative stress and inflammatory response (23, 24), which were considered an important pathogenesis of DR (25–27), Longer disease duration means a sustained state of inflammation for a longer period, which raises the risk of DR. As reported by Singh et al. (28), DR prevalence was five times higher in patients with a disease duration of >15 years than it was in patients with an illness duration of < 5 years. According to Sun et al. (29), the duration of diabetes is strongly correlated with the risk of DR. In middle-aged patients with T2DM. It is critical to diagnose and intervene in the early stages of DR progression to minimize the risks that come with a longer disease course.

Gender has been discovered in some studies to be relevant to the risk of developing diabetes-related complications (30, 31). Middle-aged men were significantly more likely to have T2DM, indicating that gender factors are involved to some extent in the pathogenesis of T2DM and its complications in the middle-aged population (32, 33). Several studies have revealed sexual dimorphism in fat distribution, inflammatory signaling pathway activation, and T2DM risk (34–37). Middle-aged T2DM patients have demonstrated gender differences in disease progression and pathogenesis, and our model suggested that being male is significantly associated with DR in the middle-aged population. Studies based on national databases from the UK and Finland found that the male sex is an independent risk factor for advanced DR in T2DM and a risk predictor for disease progression (38, 39). Maric-Bilkan et al. (40) concluded that age-related differences in hormone levels, glycemic control, duration of diabetes, and ethnic background could explain the reported gender differences in DR risk. Although the pathological mechanisms of gender influence on DR progression are unclear now, the significantly different prevalence between gender implied different individualized care measures. Prevention strategies targeting modifiable risk factors are critical for the middle-aged T2DM population.

Since its first clinical use in 1922, exogenous insulin has become a widely used hypoglycemic drug for many forms of diabetic patients worldwide (41). Our results found that mid-aged T2DM patients on insulin therapy were at greater risk of developing DR. A meta-analysis based on seven cohort studies has shown a significant association between the use of insulin and the risk of DR (42). A systematic review conducted by Song et al. (43) discovered that insulin therapy was remarkably correlated with an increased prevalence of any DR. The correlation between insulin therapy and DR demonstrated in various studies indicated that clinicians need to be more cautious when applying insulin therapy to patients at high risk of developing DR. Besides when dealing with patients on long-term insulin therapy, DR should be detected more carefully.

UACR is a clinically used indicator of renal function and a marker of endothelial dysfunction and may affect the microvasculature of the kidney and retina. Wang et al. (44) found that UACR, in addition to being an important marker of chronic kidney disease, was also closely related to the progression of DR. A 10-year prospective follow-up study confirmed that both UACR and estimated glomerular filtration rate (eGFR) were significant risk factors for DR, but UACR had a more significant association than eGFR (45), which is consistent with our result. The current studies found that high UACR is linked to changes in retinal vascular geometry, that patients with high UACR appear to be potentially predisposed to systemic vascular endothelial cell disease, and glycemic control may not affect the inherent biological risk of developing microvascular complications (46, 47). For this reason, UACR may be a favorable and easily accessible biochemical indicator for predicting DR.

Although the prediction model developed in this research is meaningful for the early prevention and treatment of the middle-aged T2DM population, there are still some limitations. First, the population included in this study was the general middle-aged US population. Due to the differences in lifestyle and eating habits, the DR prediction nomogram may be limited in its generalization to other national people. Second, all patient data in this study were obtained from the NHANES database. Although we used data from different periods for validation, multicenter clinical validation is needed to assess the efficacy of the nomogram. Third, we could not refine our DR study according to whether it was proliferative due to the lack of data limitations of DR staging in the NHANES database data.

The study developed a new web-based nomogram for predicting DR prevalence in middle-aged T2DM patients. After internal and external validation, the nomogram demonstrated good predictive performance. The line chart includes 5 common clinical characteristics of gender, serum phosphorus, UACR, duration of diabetes, and use of insulin. This nomogram enabled early to identify the high-risk groups of DR in middle-aged T2DM patients and helped to develop an aggressive individualized prevention and treatment strategy to reduce the prevalence and slow down the progression of DR. More clinical prospective and multicenter trials are needed to confirm our nomogram.

This study uses data from a free and open public database, which can be found here: www.cdc.gov/nchs/nhanes/.

Ethical approval was not provided for this study on human participants because each participator provided written informed agreement before inclusion in the NHANES database, which was examined and allowed by the National Center for Health Statistics Ethics Review Board. Anonymously processing the data makes it available to the public. The researchers then can transform the data into a form suitable for analysis following privacy-preserving. Based on the study’s data usage guidelines, all data will be analyzed statistically, and all studies will comply with all relevant laws and standards. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.

G-XW and X-YH contributed equally to this study. All the authors contributed to the article and approved the final version.