Prevalence of diabetic retinopathy across different diabetes phenotypes

Abstract

Diabetic retinopathy (DR) is a glucose-defining complication with varying prevalence reported depending upon the duration of diabetes, degree of glycemic control, and associated other microvascular complications, particularly nephropathy, defined by albuminuria and reduced glomerular filtration rate. The prevalence of DR also varies with the diabetes phenotype, with ketosis-prone diabetes having DR as prevalent as type 2 diabetes, but less than type 1 diabetes, matched for age and gender. An increasing prevalence in ketosis-prone diabetes plausibly could be due to ketonemia increasing permeability of blood-retinal layer contributing to inflammation. The guidelines for initial screening and evaluation of DR suggest similar strategies across diabetes phenotypes. Recent advances in DR diagnosis with advanced imaging technologies with artificial intelligence shall enable earlier detection with improved accuracy across varied diabetes phenotypes to reduce preventable blindness due to DR.

Key Words: Diabetic retinopathy; Ketosis prone diabetes; Type 2 diabetes; Type 1 diabetes; Maturity-onset diabetes of the young; Ketonemia; Vascular permeability

Core Tip: Diabetic retinopathy (DR) is glucose defining complication with varying prevalence reported depending upon the duration of diabetes, degree of glycemic control, associated other microvascular complications particularly nephropathy defined by albuminuria and reduced glomerular filtration rate. Interestingly, the prevalence of DR also varies with the diabetes phenotype with ketosis prone diabetes having DR as prevalent as type 2 diabetes but less than type 1 diabetes matched for age and gender.

INTRODUCTION

Diabetic retinopathy (DR) is a microvascular complication of diabetes mellitus characterized by progressive retinal damage due to chronic hyperglycemia, oxidative stress, and inflammation. It affects approximately one-third of individuals with diabetes worldwide and is a major cause of visual impairment[1].

In a large, population-based study in the United States, a prevalence rate of 26.43% (95% uncertainty interval: 21.95-31.60) was reported among people with diabetes, which is similar to that reported in a recent study by Xu et al[2,3]. The prevalence of DR depends upon the duration of diabetes, the degree of glucose control, and associated microvascular complications[4,5]. Out of these, the duration of hyperglycemia (diabetes) is the strongest risk factor for the development of DR, with the risk increasing by 8% with each year. Therefore, the recommendations are for detailed fundoscopic examination at diagnosis of type 2 diabetes (T2DM) and annually thereafter.

Interestingly, there is a non-linear correlation between the occurrence of DR and the duration of diabetes. In the first 8 years of diabetes, a positive correlation exists between diabetes mellitus duration and DR, but the effect plateaus after 8 years, and no significant correlation is observed thereafter[4]. The degree of glucose control also appears to suggest the risk of DR. Studies in both type 1 diabetes (T1DM) and T2DM have suggested that optimal and early glycemic control can aid in preventing or delaying the onset and progression of DR[5,6]. The importance of early management of blood glucose levels to achieve intensive treatment targets cannot be underestimated to postpone the development of DR.

WHAT ARE THE DIFFERENCES IN PREVALENCE OF DR ACROSS DIABETES PHENOTYPE?

Recently, The DR prevalence in the study by Xu et al[3] was 9.5% in ketosis-onset diabetes and 12.3% in T2DM, both significantly higher than in T1DM (5.7%). The definition of ketosis-prone diabetes in the study by Xu et al[3] was arbitrary, though autoimmune diabetes was excluded by the absence of antibodies at diagnosis. In other forms of diabetes, the prevalence of retinopathy in maturity-onset diabetes of the young (MODY) patients has been reported to be 21.5% in patients with hepatocyte nuclear factor 1 alpha-MODY (formerly MODY3) and 5.4% in patients with glucokinase-MODY (formerly MODY2), suggesting a cumulative glycemic burden, as glucokinase-MODY has less severe hyperglycemia[7]. The incidence rates of any degree of DR in patients with MODY were 880 (270-1490) cases per 10000 person-years, which is more than published incidence rates for T2DM (200 to 300 per 10000 person-years) and comparable with rates reported in T1DM (500-660 cases per 10000 person-years). The prevalence rates appear to be similar to those reported in the present study[8].

The incidence of any DR was found to be similar between type 3c DM (secondary to chronic pancreatitis) and T2DM when matched for the duration and degree of hyperglycaemia[9]. They found that DR prevalence increases substantially with age, requiring age-matched populations to assess differences in DR prevalence. They also found that, with age over 80 years, the prevalence of DR decreases, which was attributed to higher mortality among elderly individuals with DR. Overall, it seems that, irrespective of the type of diabetes, the prevalence of DR correlates mainly with glycemic burden and the duration of exposure to hyperglycemia.

PATHOPHYSIOLOGY AND MECHANISTIC DIFFERENCES FOR DR

Glycemic burden and temporal exposure

The duration and pattern of hyperglycemia are key determinants of DR severity. In T1DM, early onset results in prolonged lifetime exposure to hyperglycemia, with sustained glycemic burden driving cumulative microvascular damage[10]. In T2DM, hyperglycemia often exists for years before diagnosis, leading to early DR at presentation and variable severity depending on prior glycemic control. Clinically, DR severity in T1DM is primarily driven by disease duration, whereas in T2DM, it reflects both prolonged hyperglycemia and delayed detection. Mechanistically, chronic hyperglycemia activates pathways such as advanced glycation end-product formation, protein kinase C signalling, and the polyol pathway, inducing oxidative stress, endothelial dysfunction, and capillary damage that lead to DR[11-13].

Insulin deficiency and insulin resistance

A key distinction between diabetes phenotypes lies in insulin physiology, which directly impacts retinal microvascular injury. In T1DM, absolute insulin deficiency leads to increased lipolysis and ketogenesis, causing direct metabolic stress on retinal cells[14,15]. In contrast, T2DM is characterized primarily by insulin resistance, accompanied by hyperinsulinemia, endothelial dysfunction, and altered nitric oxide signalling. Insulin resistance further impairs vascular autoregulation, reduces retinal perfusion, and increases ischemia, contributing to the earlier onset of diabetic macular edema in T2DM[16].

Differential inflammatory milieu

Chronic low-grade inflammation is a central, yet phenotype-specific contributor to DR progression. In T2DM, a pro-inflammatory systemic environment marked by elevated tumor necrosis factor-α, interleukin (IL)-6, IL-1β, increased adipokines, and innate immune activation promotes vascular permeability, disrupts the blood-retinal barrier, and induces leukostasis and capillary occlusion[17,18]. In contrast, inflammation in T1DM is primarily metabolically driven rather than obesity-mediated. Clinically, this translates to a higher prevalence of DME and vascular leakage in T2DM, whereas T1DM more commonly progresses toward ischemia-driven proliferative DR.

Oxidative stress and mitochondrial dysfunction

Oxidative stress is a common mechanism in DR but varies in magnitude and underlying drivers across phenotypes. Hyperglycemia-induced reactive oxygen species damage endothelial cells, pericytes, and the neural retina, leading to capillary dropout, microaneurysm formation, and neurodegeneration[19]. In T1DM, reactive oxygen species generation is primarily driven by glucose toxicity, whereas in T2DM, it is amplified by dyslipidemia, obesity, and mitochondrial dysfunction. This results in greater metabolic complexity and variability in DR progression in T2DM[15].

Microvascular structural differences

Early retinal microvascular changes in DR vary by diabetes phenotype. Pericyte loss, a hallmark of DR, causes capillary instability and microaneurysm formation. In T1DM, capillary degeneration is more uniform, leading to progressive ischemia and a higher risk of neovascularization[20]. In contrast, T2DM exhibits patchy microvascular damage with greater vascular leakage, predisposing to macular edema. These differences explain the higher prevalence of proliferative DR (PDR) in T1DM and DME in T2DM[21].

Role of comorbidities and systemic factors

Comorbid conditions significantly modulate DR severity, especially in T2DM (Table 1). Patchy microvascular lesions and vascular leakage are more prevalent in T2DM, contributing to greater heterogeneity in retinal damage and faster progression of DR in certain patient subgroups[22].

Table 1 Role of comorbidities and systemic factors.

Factors	Impact on diabetic retinopathy
Hypertension	Accelerates microvascular damage
Dyslipidemia	Promotes exudate formation
Obesity	Enhances inflammation
Chronic kidney disease	Reflects systemic microangiopathy

Neurodegeneration and neurovascular unit dysfunction

Emerging evidence indicates that DR is not only a vascular but also a neurodegenerative disease. Retinal neuronal dysfunction can precede visible vascular lesions and involves glutamate toxicity, mitochondrial dysfunction, and glial activation. Phenotype-specific differences exist: In T1DM, neurodegeneration is primarily driven by metabolic stress, whereas in T2DM, it results from a combination of metabolic and inflammatory insults[23].

Genetic and epigenetic influences

Genetic susceptibility is likely to play a significant role in DR risk across diabetes phenotypes, with polymorphisms in genes such as vascular endothelial growth factor, aldose reductase, and inflammatory mediators contributing to individual variability. Epigenetic mechanisms, including “metabolic memory”, can cause persistent retinal damage despite good glycemic control, particularly in patients with long-standing T1DM or poorly controlled T2DM[15].

WHAT ARE THE IMPLICATIONS FOR THE DIAGNOSIS OF DR ACROSS PHENOTYPES?

T1DM

Patients with T1DM typically develop DR after a longer duration of disease, with its prevalence closely linked to cumulative glycemic exposure over time. They have a higher lifetime risk of progressing to PDR, particularly after 10-15 years of disease duration. Due to absolute insulin deficiency, microvascular damage begins early, leading to a greater likelihood of developing advanced retinal lesions such as neovascularization and vitreous hemorrhage. Longitudinal studies, including the Diabetes Control and Complications Trial/Epidemiology of Diabetes Interventions and Complications cohorts, have demonstrated that intensive glycemic control significantly reduces the progression of DR in individuals with T1DM[24].

T2DM

DR may already be present at the time of diagnosis in T2DM, largely due to prolonged periods of asymptomatic hyperglycemia. As a result, DR is often detected earlier relative to the formal diagnosis of diabetes. Its development and progression are strongly associated with other components of metabolic syndrome, including hypertension, dyslipidemia, and obesity. Additionally, individuals with T2DM have a higher prevalence of DME. Unlike T1DM, DR progression in T2DM is influenced by multifactorial metabolic dysregulation, not solely hyperglycemia[25].

Latent autoimmune diabetes in adults

Latent autoimmune diabetes in adults (LADA) represents an intermediate phenotype between T1DM and T2DM, sharing features of both conditions. It is characterized by a slower progression compared to T1DM, yet a faster decline in β-cell function than typically seen in T2DM. Patients with LADA carry a moderate risk of developing DR, which is often underestimated in clinical practice. Delayed initiation of insulin therapy can further contribute to poorer glycemic control, thereby increasing the risk of DR. Emerging evidence suggests that autoimmune mechanisms, combined with insulin resistance, may contribute to retinal microvascular damage in LADA[26].

MODY

It comprises monogenic forms of diabetes with distinct genetic aetiologies, leading to considerable variability in clinical presentation and complication risk. The prevalence and severity of DR differ across MODY subtypes; for instance, individuals with glucokinase-MODY generally exhibit a lower risk due to persistent but mild hyperglycemia, whereas those with hepatocyte nuclear factor 1 alpha-MODY are more prone to progressive hyperglycemia and consequently a higher risk of DR. Therefore, accurate phenotype-specific diagnosis is essential, as the likelihood and progression of DR are closely influenced by the underlying genetic mutation[27].

Ketosis-prone and atypical diabetes phenotypes

Emerging diabetes phenotypes, such as ketosis-prone diabetes, exhibit variable risk profiles for DR. These conditions are characterized by intermittent insulin deficiency with periods of remission, leading to fluctuating glycemic control. Consequently, the risk of DR in these patients is more closely related to cumulative glycemic burden over time rather than the phenotype itself[28].

Gestational diabetes mellitus

Although DR is uncommon in gestational diabetes, pregnancy can accelerate the progression of pre-existing DR in women with prior diabetes. This is particularly evident in pregestational diabetes, where rapid worsening of DR may occur due to hormonal and hemodynamic changes that destabilize retinal vasculature. Therefore, close ophthalmic monitoring during pregnancy is essential to prevent vision-threatening complications[29].

WHY ARE THERE DIFFERENCES IN DR AMONGST DIABETES PHENOTYPES?

Phenotype-specific differences in diabetes significantly influence underlying pathophysiological mechanisms, including microvascular damage pathways, retinal neurodegeneration, and retinal perfusion abnormalities, thereby contributing to variability in the development and progression of DR across different diabetes subtypes (Table 2).

Table 2 Phenotype-specific pathophysiological differences.

Mechanism	T1DM	T2DM	LADA	MODY
Insulin deficiency	Absolute	Relative	Progressive	Variable
Insulin resistance	Minimal	Prominent	Moderate	Minimal
Inflammation	Moderate	High	Moderate	Low-moderate
Onset of diabetic retinopathy	Delayed	Often early	Intermediate	Variable

T1DM: Type 1 diabetes; T2DM: Type 2 diabetes; LADA: Latent autoimmune diabetes in adults; MODY: Maturity-onset diabetes of the young.

Xu et al[3] suggest an increased prevalence of DR in ketosis-prone diabetes compared to T1DM, which is intriguing. It has been suggested that this may be due to fluctuations in glucose levels and abnormalities in lipid metabolism, but none of these parameters were studied. It would be interesting to know the differing prevalence of DR across sub phenotypes of ketosis-prone diabetes based on beta-cell function and glycemic variability, which were unfortunately not measured in the above study. However, the common pathogenesis of ketonemia seems plausible, as ketones are known to increase permeability of the blood-brain barrier and possibly the blood-retinal barrier, contributing to inflammation in retinal layers and causing DR. On the other hand, ketone diets that may increase beta-hydroxy butyrate have neuroprotective effects as ketone can be utilized as a direct energy source in the retina during hypoxic conditions. Ketones can have a protective effect on photoreceptors and retinal ganglion cells by shifting metabolism, potentially protecting retinal pericytes and improving hypoxia, as seen with sodium-glucose cotransporter 2 inhibitor[28].

Xu et al[3] have found an association between DR and renal function, including epidermal growth factor receptor (eGFR) and albumin excretion. Large population-based studies have also highlighted a strong association between any retinopathy and renal dysfunction that was independent of age, hypertension, diabetes, dyslipidemia, and other risk factors. This association is understandable, given shared microvascular damage due to hyperglycemia-induced oxidative stress, activation of the polyol and protein kinase C pathways, the pentose phosphate pathway, and inflammation, all of which contribute to vascular dysfunction. However, the connotations of reduced growth factor receptor and albuminuria differ, reflecting distinct pathophysiology: Mesangial expansion and glomerular basement membrane thickening define albuminuria akin to retinal pericyte leakage, followed by tubulointerstitial fibrosis linked with declining eGFR[30]. Studies have shown that the prevalence of DR closely mimics that of albuminuria and less strongly with eGFR. Another interesting finding from the study by Xu et al[3] is that the prevalence of PDR in ketosis-prone diabetes is almost one-tenth that of non-ketotic T2DM. It is known that DR is a continuum that progresses from non-PDR to PDR over the years of diabetes duration. However, patients of ketosis-prone diabetes included by Xu et al[3] were of recent onset, explaining the differences in the prevalence of PDR.

WHAT ARE THE RECENT ADVANCES IN RETINAL IMAGING TECHNOLOGIES?

Early-stage DR is often asymptomatic, making timely screening essential. Conventional diagnostic tools such as slit-lamp biomicroscopy and fundus photography are widely used but have limitations in detecting early microvascular changes. Recent technological advancements aim to overcome these limitations through enhanced imaging, automation, and data integration.

Optical coherence tomography and optical coherence tomography angiography

Optical coherence tomography (OCT) has revolutionized retinal imaging by providing high-resolution cross-sectional visualization of retinal layers. Its advanced extension, OCT angiography (OCTA), further enhances this capability by allowing non-invasive assessment of retinal microvasculature without the need for dye injection. Recent studies have shown that OCTA can quantitatively evaluate key microvascular parameters, including capillary non-perfusion, vessel density, and alterations in the foveal avascular zone. These metrics serve as sensitive early biomarkers of DR, often detectable before the appearance of clinically visible lesions[31-34].

Ultra-widefield imaging

Ultra-widefield imaging allows visualization of up to 200° of the retina, greatly enhancing the detection of peripheral lesions such as ischemia and neovascularization. Evidence suggests that peripheral lesions identified via ultra-widefield imaging are associated with increased risk of DR progression[35,36].

Adaptive optics imaging

Adaptive optics enables near-cellular resolution imaging of the retina, allowing early detection of photoreceptor damage and capillary remodeling. Although its use is currently confined to research settings, it holds significant promise for preclinical diagnosis of DR[37].

Enhanced fundus photography

Recent advancements in retinal imaging, including high-resolution digital fundus cameras and smartphone-based imaging systems, have greatly facilitated mass screening programs, making early detection of DR more accessible, especially in low-resource settings[38,39].

BIOMARKERS FOR EARLY DETECTION

Various biomarkers have been studied for the prediction or earlier detection of DR, including vascular, functional, and molecular biomarkers based on the known pathophysiology of DR secondary to hyperglycemia[40]. Key retinal biomarkers, including retinal thickness changes, microaneurysm turnover rate, and capillary dropout, can be quantitatively assessed using OCT and OCTA, providing objective measures for early detection and monitoring of DR. Vascular biomarkers such as vessel tortuosity, fractal dimension, and vessel caliber provide insight into early microvascular dysfunction, serving as important indicators for the onset and progression of DR[41]. Functional biomarkers, assessed using techniques such as electroretinography and contrast sensitivity testing, can detect retinal dysfunction in DR before structural damage becomes apparent[42]. Emerging molecular and systemic biomarkers, vascular endothelial growth factor, inflammatory cytokines such as IL-6 and tumor necrosis factor-α, and metabolomic or proteomic profiles, offer the potential for personalized risk stratification in DR[43].

ARTIFICIAL INTELLIGENCE IN DR DIAGNOSIS

Artificial intelligence (AI) has emerged as a transformative tool in the diagnosis and management of DR. Deep learning-based screening systems, particularly those using convolutional neural networks, have demonstrated high diagnostic performance, often exceeding 90% sensitivity and specificity for detecting referable DR. Several autonomous AI systems have been validated for clinical use, enabling rapid image analysis, automated DR grading, and reduced dependence on ophthalmologists, thereby improving both screening efficiency and accessibility[44-47].

Multimodal AI approaches further enhance diagnostic accuracy by integrating fundus images, OCT/OCTA data, and clinical variables such as glycated hemoglobin A1c (HbA1c) and diabetes duration. This comprehensive approach enables more precise disease staging and improves detection of DME, providing a more complete assessment than single-modality imaging[47,48]. To increase clinical trust and adoption, explainable AI frameworks have been developed. Techniques such as saliency maps and heatmaps highlight pathological regions within images, enabling clinicians to interpret AI outputs and verify that predictions align with known disease patterns[49].

Recent AI models have also emphasized precision diagnostics across diabetes phenotypes. By incorporating clinical metadata such as diabetes type, disease duration, and HbA1c levels, these models can provide phenotype-specific risk prediction, identify high-risk individuals early, and personalize screening intervals. This is particularly important because DR is not a uniform complication; its prevalence, onset, and severity vary significantly among diabetes phenotypes due to interacting metabolic, inflammatory, vascular, and genetic mechanisms. While chronic hyperglycemia is the central unifying factor, phenotype-specific modifiers influence the trajectory of retinal injury[50,51]. For example, T1DM typically manifests with duration-dependent microvascular degeneration, whereas T2DM involves multifactorial metabolic dysregulation, often leading to earlier onset but more heterogeneous disease expression[52]. Understanding these mechanistic differences supports the development of phenotype-specific screening strategies, biomarker-driven monitoring, and personalized therapeutic interventions in DR.

Finally, generalizability and real-world deployment of AI systems remain active areas of research. Techniques such as domain adaptation, federated learning, and bias mitigation aim to improve AI model robustness across diverse populations and imaging devices, protect patient data, and ensure equitable diagnostic performance. These advancements collectively enhance the reliability, scalability, and clinical utility of AI in diabetic eye care.

TELEMEDICINE AND DIGITAL HEALTH

Teleophthalmology has expanded DR screening by enabling remote image acquisition, cloud-based analysis, and AI assisted triage. This approach is especially valuable in rural and underserved populations, improving screening coverage and reducing diagnostic delays[53]. Additionally, mobile health platforms and smartphone-based fundus imaging are increasingly integrated with AI systems, facilitating point-of-care diagnostics. Recent AI models have explored diagnosis using anterior segment images (e.g., iris and sclera), demonstrating promising accuracy. These approaches may offer non-invasive, rapid screening alternatives. Integration of imaging, genomics, and clinical data supports the development of precision ophthalmology, enabling individualized risk prediction and management[54,55]. AI-enabled handheld devices allow real-time DR detection in primary care settings, reducing dependence on specialized infrastructure. Future DR diagnostics may incorporate continuous or periodic retinal monitoring, similar to glucose monitoring. Digital biomarkers derived from retinal imaging, wearable devices, and systemic physiological data, integrated with electronic health records, could enable dynamic risk assessment. This approach allows real-time disease tracking, early detection of progression, and personalized follow-up intervals[56].

CHALLENGES AND LIMITATIONS

Despite significant advances, several challenges hinder the widespread adoption of AI and advanced imaging in DR care. These include a lack of standardized imaging protocols, limited external validation of AI models, data privacy and ethical concerns, difficulties integrating technology into clinical workflows, and cost or infrastructure barriers. Addressing these issues is essential for broader implementation and equitable access[47].

FUTURE DIRECTIONS IN THE DIAGNOSIS OF DR

The diagnostic landscape of DR is undergoing rapid transformation, driven by advances in imaging, AI, systems biology, and digital health. Future directions are expected to move beyond detection toward prediction, personalization, and prevention, aligning with broader goals in precision diabetology.

Future research in DR should prioritize longitudinal predictive modelling of disease progression, the use of federated learning for privacy preserving AI, and seamless integration of AI tools into diabetes care pathways[57]. Efforts should also focus on developing hybrid human AI diagnostic systems and conducting large-scale validation studies across diverse populations to ensure accuracy, equity, and clinical utility that will increasingly leverage multimodal data fusion, integrating retinal imaging (fundus, OCT, OCTA), clinical parameters (HbA1c, blood pressure, lipid profile), and molecular data from genomics, proteomics, and metabolomics[58-60]. This integrative approach shall enable the identification of novel biomarkers, enhance disease phenotyping, and support precision risk stratification. Additionally, multi-omics profiling may reveal distinct molecular signatures of DR across different diabetes phenotypes, guiding personalized management strategies[61,62].

Overall, the approach to DR is shifting from traditional detection of established retinal lesions toward predictive and preventive diagnostics that aim to identify disease risk before structural damage occurs. Emerging longitudinal models integrate factors such as glycemic variability from continuous glucose monitoring, duration of diabetes, and retinal biomarkers including OCTA metrics to forecast progression to PDR and DME, enabling earlier and more personalized interventions. At the same time, advances in AI are transforming DR care through explainable models that improve clinician trust, federated learning frameworks that allow training on decentralized data while preserving patient privacy, and self-supervised or foundation models that reduce reliance on labelled datasets and enhance scalability. Parallel innovations in real-time and point-of-care diagnostics such as smartphone-based fundus imaging, handheld OCT devices, and AI enabled screening tools integrated into primary care are expanding access, particularly in rural and resource-limited settings[47,48].

Recognizing the heterogeneity of DR across diabetes phenotypes, future screening strategies are expected to become increasingly personalized, with risk-based intervals tailored to individual profiles incorporating genetic susceptibility, comorbidities, and disease subtype[63]. The diagnostic advancements are being closely integrated with therapeutic decision-making, where imaging biomarkers and AI assisted systems can predict treatment response, optimize timing, and monitor outcomes, creating a more dynamic and individualized care pathway. The emerging non-retinal and non-invasive approaches including anterior segment imaging, systemic biomarkers in blood or tears, and functional retinal assessments such as electrophysiology offer promising avenues for early, rapid, and less invasive detection of DR before visible retinal changes manifest[64].

CONCLUSION

The prevalence of DR may vary by diabetes phenotypes, with patho-physiologically ketosis-prone diabetes as a subset with almost similar risk of DR as T2DM, which warrants similar screening strategies as other forms of diabetes. Recent advances in DR diagnosis are driven by the convergence of advanced imaging technologies, AI, and biomarker discovery. These innovations enable earlier detection, improved accuracy, and scalable screening solutions. Integration of these tools into routine diabetes care has the potential to significantly reduce the global burden of vision loss due to DR.