Identifying Patients with Rapid Progression to Diabetic Retinopathy in Type 2 Diabetes: A Simple Prediction Model for Community Settings

Abstract

Background

Diabetic retinopathy (DR) progression in community-based type 2 diabetes (T2D) remains challenging to predict. This multi-site cohort study aimed to identify novel metabolic risk factors beyond glucose control and develop a resource-adaptable model for prioritizing DR risk screening.

Methods

A prospective, multi-site study was conducted among patients with T2D from 2021 to 2023 across 14 community health centers in Northwestern China (Shaanxi). Baseline and follow-up data on demographic, medical history, and laboratory data were collected. DR was assessed by trained retinal specialists through fundus photography and slit-lamp bio-microscopy. LASSO-penalized logistic regression with 10-fold cross-validation was first performed to identify candidate predictors, followed by multivariable logistic regression to construct the prediction model. Model discrimination was evaluated using receiver operating characteristic (ROC) curve analysis and bootstrap internal validation. Clinical characteristics before and after onset of DR were analyzed using the paired-sample Wilcoxon test.

Results

Among 363 participants who completed follow-up, 50 (13.77%) patients developed DR. Patients who developed DR had longer duration of diabetes, higher HbA1c levels, poorer glycemic control, and elevated urine pH. Multivariate logistic regression identified HbA1c (OR = 1.30), urine pH (OR = 2.55), and diabetes duration (OR = 1.06) as independent risk factors. The prediction model showed moderate discrimination, with an apparent AUC of 0.72 and an optimism-corrected AUC of 0.693 after bootstrap internal validation. In paired analyses, patients exhibited significantly higher BUN and eGFR levels and lower serum creatinine after DR onset (all P < 0.05).

Conclusions

In this community-based cohort, a prediction model incorporating urine pH, HbA1c, and diabetes duration showed potential for identifying individuals at high risk of incident DR. Urine pH may serve as a simple and accessible metabolic marker for risk stratification, while renal functional alterations may reflect early systemic changes beyond glycemic control.

Introduction

Diabetic Retinopathy (DR) is the most common microvascular complication of type 2 diabetes (T2D) and the leading cause of vision loss in the working-age population worldwide. Epidemiological studies show that 22.3% of diabetic patients develop DR. Of these, 6.2% progress to severe vision- threaten stages, such as proliferative DR or diabetic macular edema.¹ Although, over 90% of DR-induced blindness could be prevented through early intervention, the insidious onset of the disease often leads to delayed diagnosis, creating a significant burden on individuals, families and society.² Given this substantial clinical and socioeconomic impact, early identification of high-risk individuals—particularly during the preclinical stage, is essential to prevent retinal damage, preserve vision, and reduce long-term healthcare costs.

Currently established clinical risk factors for identifying high-risk patients include long duration diabetes, poor glycaemic control (elevated blood glucose and HbA1c levels), hypertension, increased body mass index (BMI), and impaired renal function.^3,4 However, there is considerable variability among individuals regarding the onset and severity of DR. Notably, some patients with good glycaemic control and without hypertension still develop DR, suggesting that additional unidentified factors may also play a significant role in the development of DR.^5,6 Furthermore, DR is a progressive chronic disease, yet most existing studies have focused on patients in the middle or later stages, resulting in a scarcity of truly “rapid progression” cases and their associated biomarkers.

Community-based populations offer a unique opportunity to investigate the natural progression and early characteristics of DR. Unlike clinic-based samples—which usually focus on later stage patients—community populations capture a broader spectrum of the disease, including undiagnosed and rapid progression cases.⁷ In recent years, most studies have relied on multi-center cohorts across different regions in China. Although these studies provide large sample sizes, they often suffer from considerable heterogeneity due to variations in patient demographics, diagnostic criteria, and healthcare practices, and frequently focus on patients with relatively severe DR phenotypes.^8,9 In addition, many existing predictive models are highly complex, incorporating numerous parameters that may limit their practical utility.^10,11 These factors collectively hinder the development of robust, clinically applicable models for early risk prediction.

Therefore, in the present study, we conducted a community-based investigation within a single geographic region, aiming to reduce population heterogeneity and capture the clinical features associated with rapid progression to DR. Leveraging these findings, we subsequently developed a simple and practical risk prediction model to facilitate the early identification of individuals at high risk of rapid DR progression.

Materials and Methods

Study Design and Participants

A multi-site study was conducted in Xi’an, Shannxi Province, China, from 2021 to 2023, involving a total of 14 community health centers. The inclusion criteria were as follows: (1) diagnosis of type 2 diabetes (T2D) according to the World Health Organization (WHO) criteria; (2) age ≥ 18 years; (3) permanent residency in the selected communities, with all participants registered in the local community chronic disease management system. The exclusion criteria included: 1) patients who were younger than 18 years; 2) type 1 or gestational diabetes; 3) patients with malignancies or severe systemic illnesses that could impact ocular status or data completeness (eg, end-stage renal disease on dialysis; 4) patients with ocular co-morbidities such as severe cataracts, retinal vein occlusion, or advanced age-related macular degeneration. Only participants who completed follow-up visits were included in the subsequent analysis.

During follow-up, participants who developed DR were categorized into the DR group, while those who remained free of DR were classified as the non-DR group. All participants provided written informed consent before enrollment. The study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of Xi’an People’s Hospital (Xi’an Fourth Hospital), (20220066).

Clinical Data Collection and Laboratory Test

Demographic and clinical parameters, including age, sex, systolic/diastolic blood pressure (SBP/DBP), diabetes history, and disease duration, were systematically recorded. Following ≥ 10-hour overnight fasting, venous blood and midstream urine samples were collected under standardized phlebotomy protocols. The complete blood count (CBC) was performed to evaluate hematological parameters, including white blood cell count (WBC), neutrophil count (NEU), lymphocyte count (LYM), red blood cell count (RBC), hemoglobin (Hb), hematocrit (Hct), and the neutrophil-to-lymphocyte ratio (NLR). Serum glucolipid metabolic indexes were measured using an automatic biochemical analyzer, including HbA1c, cholesterol (TC), triacylglycerol (TG), high-density lipoprotein cholesterol (HDL-C) levels, blood urea nitrogen (BUN), uric acid (UA), serum creatinine (Scr). Estimated glomerular filtration rate (eGFR) was calculated based on serum creatinine levels using the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation. Urinalysis consisted of routine urine analysis, urinary sediment examination, and urinary micro-albumin assessment. Urine pH was measured using pH test strips. The strip was immersed in a fresh midstream urine sample, and the pH value was determined by comparing the resulting color change to a standard color chart. All samples were tested within 2 hours of collection.

Definitions of Diabetic Retinopathy

Ophthalmic examinations included visual acuity assessment, intraocular pressure measurement (Icare ic100, Icare Finland Oy, Finland), slit-lamp examination (by KJ5S1, Suzhou Kangjie Medical Inc, China), and fundus photography. Fundus images were captured using a portable fundus camera (the Horus DEC 200, Miis, Hsinchu, Taiwan), acquiring two 45-degree color photographs centered on the optic disc and macula, respectively. Image acquisition was performed by specially trained ophthalmic technicians following standardized protocols, under both mydriatic and non-mydriatic conditions to ensure clear, accurately positioned, and sufficiently wide-field images. All images were uploaded to a dedicated electronic data capture (EDC) platform. Two experienced retinal specialists independently graded DR status based on the fundus photographs. In cases of disagreement, a senior ophthalmologist reviewed the images and conducted a consensus discussion to determine the final diagnosis.

Statistical Analysis

Statistical analysis was performed using the R 4.2.3 software. Demographic and clinical variables were reported as mean (SD) for continuous variables and frequency (percentage) for categorical variables. The χ2-tests were used for comparing categorical data, while Student’s t-tests were carried out to compare the continuous variables between groups. To reduce the risk of overfitting and identify the most informative predictors, least absolute shrinkage and selection operator (LASSO) penalized logistic regression with 10-fold cross-validation was performed using the glmnet package. The optimal penalty parameter (λ) was determined according to the minimum cross-validation criterion (λmin), and variables with non-zero coefficients were retained for subsequent analyses. Variables selected by LASSO were then entered into a multivariable logistic regression model to estimate odds ratios (ORs) and 95% confidence intervals (CIs) for factors associated with rapid DR progression.

Receiver operating characteristic (ROC) curves were generated to evaluate the predictive value of each risk factor, and the area under the curve (AUC), sensitivity, specificity, and optimal cutoff values were calculated accordingly. To assess model optimism and internal validity, bootstrap internal validation with 1000 resamples was performed using the rms package. Model discrimination was quantified by the apparent area under the ROC curve (AUC) and the optimism-corrected AUC estimated from bootstrap resampling. Clinical characteristics before and after DR onset were assessed using the paired-sample Wilcoxon signed-rank test. All analyses were two-tailed with p < 0.05 as statistical significance.

Results

Follow-Up Outcomes and Baseline Characteristics According to DR Development

A total of 363 participants completed at least one follow-up, including 318 with one follow-up and 45 with two follow-up visits. The mean age of the participants was 68.81 years, 148 were male (40.77%), and the mean duration of diabetes was 9.3 years.

During the 1–2-year follow-up, 50 participants (13.77%) progressed from diabetes without retinopathy at baseline to newly diagnosed DR, representing the primary outcome of this study. Most incident DR cases were classified as mild or moderate NPDR, with only one patient developing severe NPDR and one progressing directly to PDR. An additional 66 participants experienced other changes in DR status, including DR regression or progression among those with pre-existing DR. The remaining 247 participants maintained a stable non-DR status throughout follow-up.

To investigate the potential factors associated with the development of DR, the study population was divided into two groups: those who remained free of DR (NDR group) and those who developed DR during follow-up (DDR group). Compared with NDR group, patients in DDR group exhibited a significantly longer duration of diabetes (10.96 ± 8.22 vs 7.95 ± 6.57 years, P = 0.006), higher HbA1c (7.88 ± 1.64 vs 7.27 ± 1.51%, P = 0.02), and poorer glycemic control both at present and over the past six months (P = 0.002 and P = 0.007). Although no significant difference was found, patients in the DDR group tended to have higher SBP. Notably, the urine pH was significantly higher in the DDR group compared to the NDR group in this study (5.74 ± 0.84 vs 5.37 ± 0.57, P = 0.001). No significant differences were observed between the two groups in terms of other parameters (Table 1).

Table 1.

Baseline Characteristics of Patients Who Developed DR (DDR) and Those Without DR Progression (NDR)

Characteristics	DDR Group	NDR Group	P value
N	50	247
Age (year)	70.00 (6.83)	68.88 (8.48)	0.382
Female, n (%)	27 (54.0)	143 (57.9)	0.726
Diabetes duration (year)	10.96 (8.22)	7.95 (6.57)	0.006
Anthropometric factors
SBP(mmHg)	145.45 (22.94)	139.58 (18.07)	0.06
DBP (mmHg)	79.11 (11.20)	78.86 (10.31)	0.884
BMI (kg/m2)	24.88 (3.92)	25.07 (3.85)	0.761
Hematological Parameters
WBC (109/L)	6.29 (1.71)	6.15 (1.49)	0.596
Neu (109/L)	3.99 (1.48)	3.72 (1.12)	0.185
Lym (109/L)	1.83 (0.64)	1.94 (0.60)	0.324
RBC (1012/L)	4.65 (0.55)	4.58 (0.50)	0.41
Hb (g/L)	140.88 (13.98)	141.04 (14.42)	0.947
Hct (%)	42.69 (4.02)	42.70 (3.96)	0.989
Biochemical data
TG (mmol/L)	4.56 (1.47)	4.42 (1.15)	0.506
TC (mmol/L)	2.08 (2.54)	1.78 (1.34)	0.271
HDL-C (mmol/L)	1.32 (0.26)	1.30 (0.32)	0.714
HbA1c (%)	7.88 (1.64)	7.27 (1.51)	0.02
BUN (mmol/L)	5.37 (1.09)	5.23 (1.36)	0.526
UA (umol/L)	304.54 (101.79)	316.65 (88.10)	0.432
Scr (umol/L)	70.02 (9.72)	71.98 (12.61)	0.346
eGFR	84.03 (12.04)	82.99 (23.32)	0.781
Urine pH	5.74 (0.84)	5.37 (0.57)	0.001
Questionnaire data
FH-diabetes (%)	28 (57.1)	133 (54.5)	0.856
Regular Eye Examinations (%)	21 (42.9)	109 (44.7)	0.94
Current FBG control			0.002
Unknown (%)	11 (22.0)	16 (6.5)
Very good control (%)	4 (8.0)	31 (12.7)
Poor control (%)	11 (22.0)	33 (13.5)
Good control (%)	9 (18.0)	83 (33.9)
Fair control (%)	15 (30.0)	82 (33.5)
6-month BG control (%)			0.007
Severe fluctuation (%)	2 (4.0)	3 (1.2)
Unknown (%)	9 (18.0)	19 (7.8)
Very good control (%)	1 (2.0)	8 (3.3)
Poor control (%)	13 (26.0)	34 (13.9)
Good control (%)	12 (24.0)	56 (22.9)
Fair control (%)	13 (26.0)	125 (51.0)

Note: Data were presented as number (%) for categorial variables, and mean (SD) for continuous variables.

Abbreviations: SBP, Systolic Blood Pressure; DBP, Diastolic Blood Pressure; TG, Triglycerides; TC, Total Cholesterol; HDL-C, High Density Lipoprotein Cholesterol; BUN, Blood Urea Nitrogen; UA, Uric Acid; Scr, Serum Creatinine; WBC, White Blood Cell; Neu, Absolute Neutrophil Count; Lym, Absolute Lymphocyte Count; RBC, Red Blood Cell Count; Hb, Hemoglobin; Hct, Hematocrit; eGFR, estimated glomerular filtration rate; FH-diabetes, Family History of Diabetes.

Selection of Predictors for Rapid DR Progression Using LASSO Penalized Logistic Regression

To identify the most informative predictors, LASSO penalized logistic regression with 10-fold cross-validation was conducted. As illustrated in Figure 1A, regression coefficients were progressively shrunk toward zero as the penalty parameter (λ) increased, resulting in a more parsimonious model. The optimal penalty parameter was selected using 10-fold cross-validation (Figure 1B). At λ_min, only HbA1c, urine pH, and diabetes duration retained non-zero coefficients.

Figure 1.

Variable selection using the least absolute shrinkage and selection operator (LASSO) logistic regression. (A) LASSO coefficient profiles of the 14 candidate predictors plotted against the logarithm of the penalty parameter (log λ). Each curve represents the coefficient trajectory of an individual predictor. The numbers above the plot indicate the number of non-zero coefficients retained at each λ. (B) Ten-fold cross-validation curve for selecting the optimal penalty parameter. The red dots represent the mean binomial deviance, and the gray vertical bars represent ±1 standard error (SE). The two dashed vertical lines indicate the optimal λ determined by the minimum criterion (λ_min) and the one-standard-error criterion (λ_1se), respectively.

These variables were subsequently included in the multivariable logistic regression model. The analysis revealed that higher HbA1c (adjusted OR=1.30, 95% CI = 1.03–1.65, P = 0.02), elevated urine pH (adjusted OR = 2.55, 95% CI = 1.50–4.38, P < 0.001) and longer diabetes duration (adjusted OR = 1.06, 95% CI = 1.00–1.11, P = 0.02) were significantly associated with an increased likelihood of DR (Table 2). These findings suggest that glycemic control, urine pH and diabetes duration may serve as significant risk factors for the development of DR.

Table 2.

Results of Multivariate Logistic Regression Analysis

Variables	P value	OR (95% CI)
HbA1c	0.02	1.30 (1.03–1.65)
Diabetes Duration	0.02	1.06 (1.00–1.11)
Urine pH	< 0.001	2.55 (1.50–4.38)

Abbreviations: CI, Confidence Interval; OR, Odds Ratio.

Performance of Different Risk Factors in Distinguish Patients with DR

To evaluate the predictive value of these clinical indicators for DR onset, ROC curve analyses were performed for diabetes duration, urine pH, and HbA1c individually, as well as a combined model integrating these variables (Figure 2). Among the individual predictors, the urine pH model showed the highest specificity (0.91) but relatively low sensitivity (0.29), yielding a modest AUC of 0.62 (95% CI: 0.52–0.72). Diabetes duration showed the highest sensitivity (0.88) at a cut-off value of 3.5 years but had limited specificity (0.32), with an AUC of 0.61 (95% CI: 0.51–0.70). HbA1c exhibited a more balanced sensitivity (0.51) and specificity (0.70), although its overall discriminative ability remained modest (AUC = 0.61, 95% CI: 0.51–0.70).

Figure 2.

Comparison of ROC curves for different predictive models in diagnosing DR. The blue line represents the diabetes duration model, the green line indicates PH model, the red line represents the HbA1c model, and the black line corresponds combined model.

While each individual predictor showed only limited discriminative ability, combining HbA1c, diabetes duration, and urine pH improved overall model performance. The combined model achieved an apparent AUC of 0.72 (95% CI: 0.63–0.81), with a sensitivity of 0.67 (95% CI: 0.53–0.81) and a specificity of 0.68 (95% CI: 0.62–0.74) at the optimal cut-off value of 0.14. After bootstrap internal validation, the optimism-corrected AUC was 0.693 (Table 3).

Table 3.

Diagnostic Performance of Different Risk Factors and Their Combination for DR

Risk Factors	Criterion Value	Sensitivity (95% CI)	Specificity (95% CI)	Apparent AUC (95% CI)	Optimism-Corrected AUC
Diabetes Duration	3.5	0.88 (0.77–0.96)	0.32 (0.26–0.38)	0.61 (0.51–0.70)
Urine pH	6.25	0.29 (0.16–0.45)	0.91 (0.87–0.94)	0.62 (0.52–0.72)
HbA1c	7.55	0.51 (0.37–0.66)	0.70 (0.64–0.76)	0.61 (0.51–0.70)
Combined	0.14	0.67 (0.53–0.81)	0.68 (0.62–0.74)	0.72 (0.63–0.81)	0.693

Abbreviations: CI, Confidence Interval; AUC, Area Under the Curve.

Clinical Characteristics Before and After DR Onset

To further characterize the clinical changes accompanying DR onset, paired analyses were performed comparing clinical parameters measured before and after the development of DR (Figure 3). Compared with the pre-DR stage, patients after DR onset exhibited significantly higher BUN (5.94 ± 1.73 vs 5.36 ± 1.11 mmol/L, P = 0.020; Figure 3A) and eGFR (85.53 ± 14.81 vs 83.88 ± 11.74 mL/min/1.73 m², P = 0.033; Figure 3B), whereas serum creatinine (Scr) was significantly lower (69.22 ± 13.63 vs 69.93 ± 9.47 μmol/L, P = 0.022; Figure 3C).

Figure 3.

Comparison of clinical characteristics before and after DR onset. (A) Blood Urea Nitrogen (BUN); (B) estimated glomerular filtration rate (eGFR); (C) Serum creatinine (Scr); (D) Absolute neutrophil Count (Neu); (E) Uric acid (UA); (F) Urine pH (PH). Data are presented as box plots overlaid with paired scatter points connected by lines. P values were calculated using the paired Wilcoxon signed-rank test.

In contrast, no statistically significant changes were observed in absolute neutrophil count (Neu), uric acid (UA), or urine pH after DR onset (all P > 0.05; Figure 3D–F), although Neu tended to decrease (P = 0.062) and UA tended to increase (P = 0.064).

Discussion

Based on the community-based study, we found that the incidence of DR was 13.77% among patients with type 2 diabetes in Xi’an. Diabetes duration, HbA1c, glycaemic control and urine pH were identified as risk factors for DR, and their combination demonstrated relatively good predictive performance. Moreover, the study also revealed that rapid progression DR was often accompanied by renal impairment, suggesting they may share some common pathogenic mechanisms. To the best of our knowledge, this is the first study to systematically characterize clinical features associated with rapid progression to DR and to evaluate their predictive value in a real-world community-based population in north China.

The study found the prevalence of DR was 13.77% among individuals with an average diabetes duration of 9.8 years in a community-based cohort in Shaanxi. This figure is largely consistent with previous epidemiological studies. Lee.et al reported a DR incidence of 10.81% among patients with T2D in Taiwan.¹² Additionally, a meta-analysis of 21 cohorts estimated the incidence rate of DR was 18.6%,¹³ while a population-based study conducted in Liaoning Province documented a prevalence of 11.9%.¹⁴ Despite difference in geography and study populations, these results collectively highlight that DR remains a common and clinically significant chronic complication among individuals with diabetes. Given the increasing global and national burden of diabetes,¹⁵ there is an urgent need for improved risk stratification and early detection strategies to mitigate the long-term public health impact of DR.

To address the need for early identification of individuals at high risk of DR, we developed a simplified prediction model based on routinely available clinical indicators in a prospective community-based cohort. By comparing baseline characteristics between participants who subsequently developed DR and those who remained free of DR during follow-up, we identified diabetes duration, HbA1c, and urine pH as independent predictors of incident DR. The resulting model demonstrated moderate discriminative ability, with an apparent AUC of 0.72 and an optimism-corrected AUC of 0.693 following bootstrap internal validation. Although the predictive performance was moderate, the model relies exclusively on simple and inexpensive clinical variables that are routinely available in primary care, making it potentially suitable for community-based risk stratification and targeted screening.

Previous studies have also reported various predictive models for DR, most of which were developed in clinical populations (inpatient or outpatient settings), typically characterized by more severe phenotypes. For example, Guo et al constructed a model using variables such as sex, smoking status, diabetes duration, albuminuria, HbA1c, systolic blood pressure, and triglycerides in hospitalized patients, achieving an AUC of 0.772.¹³ Another outpatient-based cross-sectional study from Iran reported an AUC of 0.704, with a sensitivity of 60% and specificity of 69%, based on predictors including sex, BMI, age, diabetes duration, and HbA1c.¹⁶ Moreover, a further study incorporating diabetes type and duration, HbA1c or mean blood glucose, blood pressure, and retinopathy grade to predict sight-threatening DR achieved an AUC of 0.760.¹⁷

Despite the slightly lower AUC in our study compared to some previously published models, several unique strengths warrant consideration. First, our model specifically targets newly diagnosed DR cases, prior to the onset of visual impairment or advanced retinopathy, making it particularly suitable for early identification in community settings. Second, the model relies on a parsimonious set of objective, routinely collected clinical variables, enhancing its feasibility, scalability, and applicability in primary care environments with limited resources. Importantly, by enabling earlier identification of high-risk individuals during a potentially modifiable window, the model may support timely intervention and contribute to reducing the burden of advanced DR.

Notably, our study demonstrated a significant difference in urine pH between patients with and without DR, with elevated urine pH being independently associated with DR occurrence. Logistic regression analysis further indicated that each one-unit increase in urine pH was associated with a 2.55-fold higher risk of DR, suggesting that urine pH may provide additional predictive information beyond conventional metabolic risk factors. However, no significant change in urine pH was observed before and after DR onset within the same individuals. Rather than representing contradictory findings, these results likely reflect the different clinical questions addressed by the two analyses. The baseline comparison evaluated whether urine pH could distinguish individuals at higher risk of developing DR, whereas the longitudinal analysis assessed whether urine pH changed dynamically following DR onset within the same individual. Therefore, the baseline association may reflect differences in the underlying metabolic and renal status that predispose individuals to DR, while the longitudinal analysis captures only short-term within-subject changes during disease progression.

Despite the lack of a significant longitudinal change, the observed association between urine pH and DR remains biologically plausible. Previous studies have shown that urine pH is closely associated with multiple components of metabolic dysfunction, including increased waist circumference, insulin resistance (HOMA-IR), elevated fasting plasma glucose, HbA1c, low HDL-C, metabolic syndrome and non-alcoholic fatty liver disease (NAFLD).^18–21 Furthermore, the association between urine pH and type 2 diabetes remains significant even after adjustment for major confounding factors such as dietary habits, medication use, and sex, indicating that urine pH reflects persistent metabolic disturbances rather than simple physiological variation.²² More importantly, urine pH is a sensitive indicator of renal tubular function and acid-base balance.²³ Accumulating evidence indicates that DR and diabetic kidney disease share common pathogenic mechanisms involving microvascular injury, chronic inflammation, oxidative stress, and endothelial dysfunction. A recent meta-analysis reported that patients with DR were nearly four times more likely to have diabetic kidney disease,²⁴ while almost half of patients with biopsy-confirmed diabetic nephropathy concurrently exhibited DR.²⁵

Our findings further support this renal-retinal connection. In the paired longitudinal analysis, patients after DR onset exhibited significantly higher BUN and eGFR together with lower serum creatinine levels compared with their pre-DR status. These changes may reflect glomerular hyperfiltration, an early manifestation of diabetic kidney injury, suggesting that subtle renal functional alterations accompany the development of DR. Taken together, these observations support the hypothesis that urine pH may not be a biomarker that changes immediately following DR onset, but rather a surrogate marker reflecting an underlying renal-metabolic milieu that predisposes patients to retinal microvascular damage. Due to the small sample size, future prospective studies with larger sample sizes, quantitative urine pH measurements, and comprehensive assessment of medication use and dietary factors are warranted to further clarify this relationship.

The study have several strengths. First, the predictive model incorporates a limited number of variables, all of which are routinely accessible clinical parameters. This confers a high degree of feasibility and translational potential ability for application in primary care and real-world clinical settings. Second, the high homogeneity of our sample population — reflected in consistent dietary habits, uniform climatic conditions, and similar socioeconomic backgrounds — helps to minimize variability and reduce potential confounding factors. Given that socioeconomic status, lifestyle, dietary patterns, and even retinal imaging quality can influence the characteristics of DR, maintaining such population consistency is crucial for ensuring the reliability of our findings.²⁶ Third, the robust diagnostic accuracy achieved by utilizing a single, specialized ophthalmology center for all DR assessments, conducted by experienced ophthalmologists, thus significantly minimizing inter-observer variability and diagnostic discrepancies.

Several limitations of this study should be acknowledged. First, urine pH is influenced by multiple physiological and clinical factors, particularly medication use. However, detailed information on medications known to affect urinary acidification, such as sodium-glucose cotransporter-2 inhibitors, renin-angiotensin-aldosterone system inhibitors, and metformin, was not systematically collected, and therefore residual confounding cannot be excluded. Second, urine pH was measured using dipstick testing, a semi-quantitative method with an inherent measurement error of approximately ±0.5 pH units, which may have reduced the ability to detect subtle differences and longitudinal changes. Third, despite the use of penalized regression and bootstrap internal validation to improve model robustness, the relatively small number of incident DR cases may have limited the stability of the model estimates. Finally, although internal validation was performed, the model was developed within a single cohort and lacked independent external validation. Therefore, the generalizability of the prediction model should be interpreted with caution. Future multicenter prospective studies with larger sample sizes, longer follow-up durations, comprehensive assessment of potential confounders, quantitative urine pH measurements, and independent external validation are warranted to confirm and extend our findings.

Conclusions

In conclusion, this study identified urine pH, HbA1c, and diabetes duration as predictors of rapid progression from DM to DR and developed a prediction model with promising discriminatory performance. Urine pH may represent a simple and accessible metabolic marker associated with disease progression, whereas the observed renal functional changes suggest early systemic alterations beyond glycemic control. Further validation in larger, independent cohorts is required to confirm the robustness and generalizability of these findings.

Funding Statement

This study was funded by Xi’an Science and Technology Plan Major Research Project (201805104YX12SF38(3)), The International Diabetes Exchange and Practice Special Foundation (Z-2017-26-2302), Scientific Research Incubation Foundation of Xi’an People’s Hospital (LH-18, LH-11), Youth Program of Shaanxi Natural Science Basic Research Project (2023-JC-QN-0885), Scientific Research Program of Xi’an Health Commission (2024qn14), Shaanxi Provincial Key R&D Program (2026SF-YBXM-456) and General Research of Science and Technology of Xi’an (24YXYJ093).

Abbreviations

DR, Diabetic Retinopathy; T2DM, Type 2 Diabetes Mellitus; SBP, Systolic Blood Pressure; DBP, Diastolic Blood Pressure; TG, Triglycerides; TC, Total Cholesterol; HDL-C, High Density Lipoprotein Cholesterol; BUN, Blood Urea Nitrogen; UA, Uric Acid; Scr, Serum Creatinine; WBC, White Blood Cell; Neu, Absolute Neutrophil Count; Lym, Absolute Lymphocyte Count; RBC, Red Blood Cell Count; Hb, Hemoglobin; Hct, Hematocrit; eGFR, estimated Glomerular Filtration Rate; CI, Confidence Interval; OR, Odds Ratio; ROC, Receiver Operating Characteristic; AUC, Area Under the Curve; DN, Diabetic Nephropathy; NAFLD, Non-Alcoholic Fatty Liver Disease.

Data Sharing Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Ethics Approval and Informed Consent

All participants provided written informed consent before enrollment. The study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of Xi’an People’s Hospital (Xi’an Fourth Hospital) (20220066).

Author Contributions

Conceptualization, Project administration, Resources, Supervision, Funding acquisition, and Writing – review & editing: Yanchun Zhang; Conceptualization, Project administration, Resources, Supervision, and Writing – review & editing: Hongli Liu; Formal analysis, Methodology, Data curation, Visualization, and Writing – original draft: Jing Zhao; Investigation, Validation, and Writing – review & editing: Dongling Niu, Daxi Xue, Jiangnan Chu, Yingnan He, Yaqi Wang, Shuang Guo, and Yaping Wang; Investigation, Data curation, and Writing – review & editing: Ziwei Kang, Rui Wang, Jingke Li, and Li Zhang; Formal analysis, Methodology and Writing – review & editing: Zhi Chai and Ting Lei. All authors gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The authors declare no conflicts of interest.