Risk stratification model for diabetic kidney disease progression based on metabolic and inflammatory biomarkers

Abstract

BACKGROUND

Diabetic kidney disease (DKD), a leading global cause of kidney failure, is driven by a complex interplay of metabolic, inflammatory, and hemodynamic factors. While estimated glomerular filtration rate (eGFR) and albuminuria remain diagnostic cornerstones, they incompletely capture this multifactorial pathophysiology. Emerging metabolic indices and inflammatory markers show strong prognostic value, yet a practical tool integrating these novel biomarkers with standard parameters for clinical risk stratification is lacking.

AIM

To develop and validate a comprehensive, readily applicable risk stratification model for predicting DKD progression in patients with type 2 diabetes. The model integrates easily accessible metabolic and inflammatory biomarkers with standard renal parameters to enable earlier identification of high-risk patients and facilitate personalized management.

METHODS

In this single-center retrospective cohort study, we included 458 patients with type 2 diabetes hospitalized between March 2018 and December 2022. DKD progression was defined as a composite endpoint (≥ 30% eGFR decline, progression to macroalbuminuria, renal replacement therapy, or renal death). Baseline metabolic and inflammatory biomarkers were comprehensively assessed. Machine learning algorithms were employed to develop and internally validate a risk prediction nomogram, the performance of which was further evaluated in an independent external cohort (n = 215).

RESULTS

Among 458 patients followed for a median of 36 months, 142 (31.0%) experienced DKD progression. A nomogram incorporating eight independent risk predictors (e.g., baseline eGFR < 60 mL/minute/1.73 m², hazard ratio = 3.87; urinary albumin-to-creatinine ratio ≥ 300 mg/g, hazard ratio = 4.15) was developed, demonstrating excellent discrimination (C-index: 0.843) and calibration. The model stratified patients into low-risk, intermediate-risk, and high-risk groups with markedly distinct 3-year progression rates (8.7%, 28.4%, and 62.3%, respectively; log-rank P < 0.001). Critically, it significantly outperformed existing prediction models (Kidney Failure Risk Equation, Kidney Disease: Improving Global Outcomes, Chronic Kidney Disease Progression Score) in both discrimination and net reclassification (net reclassification improvement 23.6%-29.3%, all P < 0.001). Its performance was consistent across key subgroups (all interaction P > 0.10). External validation in an independent cohort (n = 215) confirmed robust generalizability, with a C-index of 0.826 and well-matched calibration across 1-3 years.

CONCLUSION

This study successfully identify a comprehensive risk stratification model by integration of metabolic and inflammatory biomarkers that accurately predicts DKD progression in patients with type 2 diabetes mellitus. The novel integrated model outperforms traditional risk scores and is a ready reference for personalized risk assessment and targeted intervention strategies in clinical practice.

Key Words: Diabetic kidney disease; Type 2 diabetes mellitus; Risk stratification; Metabolic biomarkers; Inflammatory biomarkers; Nomogram; Prediction model

Core Tip: This study proposed an innovative risk stratification model for diabetes kidney disease progression by combining metabolic and inflammatory biomarkers with renal parameters. A nomogram-based model was a good predictive tool and stratified patients into three groups: (1) Low-risk for progression; (2) Moderate-risk for progression; and (3) High-risk for progression. This readily applicable clinical model could help identify patients eligible for early intervention, tailoring treatment schemes based on their individual risk of progression and therefore reduce the burden of end-stage renal disease in high prevalence diabetes populations.

INTRODUCTION

Diabetic kidney disease (DKD) is one of the most severe microvascular complications of diabetes mellitus, affecting roughly 20%-40% of patients with type 2 diabetes. Worldwide DKD has become the most common cause of chronic kidney disease (CKD) and end-stage renal disease (ESRD), with significant associated morbidities and burdens on mortality and health care[1]. In parallel with the rapidly increasing prevalence of diabetes worldwide and more than 700 million people globally with diabetes by 2045, DKD becomes an epidemic as well[2,3].

DKD has a natural history of progressive deterioration in kidney function represented by a decrease in estimated glomerular filtration rate (eGFR) and increase in albuminuria. DKD Progression Is Often Relentless, Despite Modern TherapiesOnce established even with the best treatment modalities contemporary to today, DKD is often characterized by a relentless progressive decline of kidney function. Worsening renal function from chronic kidney disease to end-stagerenal failure is associated with high and increasing risks of acutecardiovascular events, premature death, as well as the need forexpensive renal replacement therapy. Recent epidemiological studies[4,5] have shown that even small declines in kidney function are related to mortality and other adverse outcomes[6], highlighting the importance of early detection and treatment.

Clinical practice currently makes use of eGFR and albuminuria as both risk predictors for DKD progression as well as surrogate outcomes for disease monitoring. These parameters are still essential in DKD assessment[7,8]; however, they have intrinsic limitations to encapsulate the complexity of disease etiology as a multifactorial process that underlies progression. Development and progression of DKD include complex interactions between metabolic dysregulation, hemodynamic alterations, inflammatory processes, oxidative stress and genetic susceptibility[9,10]. Thus, risk prediction models that are built around renal parameters alone may not be able to identify sizeable proportions of high-risk patients who would benefit from more intensive management.

Novelties in DKD and Emerging Evidence[11,12] Clearly Reveal Roles of Novel Biomarkers and Metabolic Indices in the Pathogenesis and Progression of Diabetes-related Kidney Disease. Insulin resistance with hyperglycemia is clearly involved in the pathogenesis of DKD, but other factors, including dyslipidemia and hyperuricemia also contribute to renal injury through multi-pathways such as endothelial dysfunction, podocyte damage and tubulointerstitial fibrosis[13,14], metabolic dysregulation measured by triglyceride-glucose (TyG) index has been proven to have a non-linear threshold effect on risk of DKD. Likewise, chronic low-grade inflammation is characterized by increased circulating inflammatory cytokines and acute-phase reactants and has been repeatedly linked to faster DKD progression. Novel metabolic indices such as the TyG index[15] and inflammatory markers including high-sensitivity C-reactive protein (CRP)[16] and neutrophil-to-lymphocyte ratio (NLR) have proved to be promising prognostic factors in many studies, and their predictive role for disease occurrence, progression, and clinical outcomes has been recently established by meta-analysis[17].

Even though these scientific advances[18,19] in DKD pathophysiology are substantial there is still a lack of comprehensive and validated risk stratification tools integrating metabolic/inflammatory biomarkers with traditional renal parameters to be used in daily clinical practice when predicting DKD progression. Current prediction models, while useful tools usually include small set of variables and have not been validated in external populations. Additionally, most models lack intuitive risk stratifications that have immediate implications to clinical management of therapy escalation and monitoring frequency.

Importantly, the mechanistic justification to assess metabolic and inflammatory biomarkers goes far beyond their epidemiological association with DKD. Even though eGFR and albuminuria reflect the presence of established structural kidney injury, they cannot reliably measure early upstream metabolic and immune dysregulation driving disease progression. The TyG index, a validated surrogate for insulin resistance constructed from fasting triglycerides and glucose measurements easily obtained in routine clinical care, reflects the degree of composite long-term renal injury (pathological processes including endothelial dysfunction, podocyte injury, and overactivation of the renin-angiotensin-aldosterone system) independently of glycemic control as estimated by hemoglobin A1c (HbA1c). Likewise, inflammatory indiceshs-CRP and NLR reflect systemic immune activation that drives tubulointerstitial fibrosis and glomerular injury unrelated to eGFR or albuminuria alone. Thus, we hypothesized that combining these biomarkers with traditional renal parameters would yield a better risk prediction and earlier identification of patients at high-risk of progression to DKD compared to the models based only on eGFR and HbA1c.

Thus, the objective of this study to develop and validate a comprehensive risk stratification model developed with easy accessible metabolic as well as inflammatory biomarkers together with routine renal parameters that can allow an accurate forecast of DKD progression in patients at the new stage of DKD progress in patients with type 2 diabetes being suitable for personalisation of patient management.

MATERIALS AND METHODS

Study design and population

This was a single-center, retrospective cohort study conducted at Chongqing University Three Gorges Hospital. The protocol for this study, which involved the analysis of existing, de-identified medical records, was reviewed and approved by the Hospital Ethics Committee (approval No. 2026-33). The study was performed in accordance with the ethical standards of the Declaration of Helsinki. After approval, the research team retrieved and anonymized all medical record data for analysis. As only retrospective, de-identified data were used and no additional contact with patients was required, the ethics committee waived the need for individual written informed consent. We identified patients by screening the electronic medical records of those hospitalized in the Department of Endocrinology or Nephrology between March 2018 and December 2022. Eligible participants were adults (≥ 18 years) with type 2 diabetes mellitus diagnosed using American Diabetes Association criteria[20], hospitalized in the Department of Endocrinology or Nephrology. Main exclusion criteria were: (1) Type 1 or other specific diabetes type; (2) Acute kidney injury within previous 3 months; (3) Biopsy-proven non-DKD; (4) Current active infection or inflammatory disease; and (5) Pregnancy or breastfeeding, life expectancy < 1 year. A total of 458 eligible patients were included. Primary outcome was DKD progression, defined as a composite of: (1) ≥ 30% sustained decline in eGFR from baseline; (2) Albuminuria progression (micro-to macroalbuminuria or macroalbuminuria to nephrotic-range proteinuria); (3) Initiation of renal replacement therapy; and (4) Death. Based on this, patients were classified into progression and non-progression groups.

Data collection and laboratory measurements

Baseline data, including demographics, diabetes duration, body mass index, blood pressure, smoking and alcohol status, and medication use (coded by ATC classification), were extracted from electronic medical records. The presence of comorbidities (including hypertension and dyslipidemia) and diabetes-related complications was also recorded.

All laboratory measurements were performed on venous blood samples that had been collected after an overnight fast of at least 8 hours during the index hospitalization. The following parameters were measured: (1) Fasting plasma glucose; (2) HbA1c via high-performance liquid chromatography; (3) Full lipid profile; (4) Serum uric acid (SUA); (5) Hemoglobin; (6) Serum albumin; (7) High-sensitivity CRP (hs-CRP); and (8) A complete blood count. Serum creatinine was measured using an enzymatic method. The eGFR was calculated using the CKD-EPI 2009 equation (where Scr is serum creatinine, κ is 0.7 for women and 0.9 for men, α is -0.329 for women and -0.411 for men, min indicates the minimum of Scr/κ or 1, and max indicates the maximum of Scr/κ or 1). The urinary albumin-to-creatinine ratio (UACR) was determined from the first-morning void urine sample. Two derived indices were calculated: (1) The NLR was obtained from the differential count; and (2) The TyG index was computed as Ln[fasting triglycerides (mg/dL) × fasting glucose (mg/dL)/2].

Follow-up and outcome assessment

Patients were followed retrospectively through their electronic medical records. Follow-up for each patient started at the index hospitalization (baseline, between March 2018 and December 2022) and ended at the earliest of the following: The occurrence of the primary outcome, death, loss to follow-up, or the administrative censoring date of December 31, 2023. This predefined cutoff date was selected to allow for sufficient follow-up time [resulting in a median of 36 months (interquartile range: 24-48 months)] for event ascertainment while ensuring data completeness at the time of analysis. Two independent nephrologists, blinded to the baseline biomarker levels and the study’s prediction model, adjudicated all potential outcome events.

Statistical analysis

Analyses were conducted in SPSS 26.0 and R 4.2.0 Continuous variables are shown as mean ± SD or median (interquartile range); categorical variables as n (%). Statistical analyses Between-group comparisons utilized t-test or Mann-Whitney U test for continuous variables and χ² or Fisher’s exact test for categorical variables.

Variable selection: Candidate predictors with P < 0.10 in univariate analysis entered least absolute shrinkage and selection operator regression with 10-fold cross-validation, where λ was selected using minimum cross-validated error. And using time-dependent receiver operating characteristic Youden index analysis, we also evaluated optimal cut-points for TyG index (≥ 9.0), NLR (≥ 3.0), hs-CRP (≥ 3.0 mg/L), HbA1c (≥ 8.5%) and SUA (≥ 420 μmol/L) in reference to existing clinical thresholds as shown in Table 1. All variance inflation factor values were < 5.0 indicating no problematic multicollinearity Complete-case analysis was used to account for missing data (< 3% per variable).

Table 1 Risk stratification and Kaplan-Meier survival outcomes by nomogram score category.

Parameter	Low-risk (0-120 points), n = 156	Intermediate-risk (121-200 points), n = 152	High-risk (> 200 points), n = 150
3-year progression rate	8.7%	28.4%	62.3%
HR vs low-risk (95%CI)	Reference	4.1 (2.3-7.3)	12.8 (7.3-22.4)
P value vs low-risk	-	< 0.001	< 0.001
Log-rank P (overall)	< 0.001
Subgroup C-index range	0.821-0.856 (age, sex, diabetes duration subgroups; all interaction P > 0.10)
DCA net benefit superiority	Superior vs “treat all”/“treat none” at 10%-70% threshold probability (most evident at 20%-50%)

Subgroup analyses included age (< 60 years vs ≥ 60 years), sex (male vs female), and diabetes duration (< 10 years vs ≥ 10 years). DCA: Decision curve analysis; HR: Hazard ratio.

Multivariable analysis: Least absolute shrinkage and selection operator-selected variables were entered into multivariable Cox proportional hazards regression. Schoenfeld residuals were used to check for proportional hazards assumption. Hazard ratios (HRs) and 95%CIs were reported; values of P < 0.05 were considered significant.

Model development and validation

Construction of nomogram: A nomogram was constructed based on the final Cox model by allocating points as per the regression coefficients. Translated total scores into probabilities of DKD progression (e.g., probability not progressing to DKD at 1 year, 2 years and 3 years). Discrimination was evaluated using Harrell’s C-index (> 0.7 = good) and time-dependent area under the curve (AUC) at 1 year, 2 years, and 3 years among DKD patients. Calibration was assessed by calibration curves with slope and intercept. Internal validation was performed through a bootstrap resampling with 1000 iterations for bias-corrected estimates. The net clinical benefit at each probability threshold level was evaluated using decision curve analysis. Patients were classified as low (0-120 points), intermediate (121-200 points), and high (> 200 points) risk; Kaplan-Meier curves were compared using the log-rank test.

External validation: Model generalizability was evaluated by applying it to a retrospective cohort of 215 independent patients from an external institution, whose data were collected between January 2019 and June 2023. Importantly, the model was developed prior toand independently of the analysis of this external cohort, and was applied to their complete historical records using the same methodology. Calibration was based on both decile grouped plots and Loess-smoothed curve. The Kidney Failure Risk Equation (KFRE), Kidney Disease: Improving Global Outcomes (KDIGO) prognosis model and CKD Progression Score[21] were applied without recalibration from original published coefficients as a form of benchmarking. Subgroup analyses by age, sex and duration of diabetes are reported in Supplementary Table 1, characteristics of the derivation vs validation cohorts appear in Supplementary Table 2.

Availability of data and materials: De-identified data are available from the corresponding author upon reasonable request.

RESULTS

Baseline characteristics and follow-up outcomes

They included 458 type 2 diabetes patients (average age: 56.8 ± 11.2 years; 58.3% male; diabetes duration: 9.7 ± 6.4 years; body mass index: 26.3 ± 3.8 kg/m²) in the cohort. Patients had a mean baseline eGFR of 78.6 ± 24.3 mL/minute/1.73 m² and 43.2% had normal albuminuria (< 30 mg/g), 35.6% had microalbuminuria (30-299 mg/g), and 21.2% had macroalbuminuria (≥ 300 mg/g). Hypertension, dyslipidemia and diabetic retinopathy were seen in 42.4%, 38.2% and 15.5%. Of these, DKD progressed in 142 patients (31.0%) over a median 36-month follow-up. Progressors were significantly older (61.4 ± 10.1 years vs 54.5 ± 11.2 years), had longer diabetes duration (12.3 ± 6.8 years vs 8.4 ± 5.9 years), lower eGFR (62.3 ± 22.1 mL/minute/1.73 m² vs 86.7 ± 21.5 mL/minute/1.73 m²), higher UACR [348 (126-856) mg/g vs 52 (18-142) mg/g], and significantly elevation of HbA1c, SUA, TyG index, hs-CRP, NLR and retinopathy prevalence compared with non-progressors (all P < 0.001). Full details are in Table 2.

Table 2 Baseline characteristics of the study population, mean ± SD/median (interquartile range)/n (%).

Characteristic	Total (n = 458)	Progression (n = 142)	Non-progression (n = 316)
Age (years)	56.8 ± 11.2	61.4 ± 10.1	54.5 ± 11.2
Male sex	267 (58.3)	89 (62.7)	178 (56.3)
Diabetes duration (years)	9.7 ± 6.4	12.3 ± 6.8	8.4 ± 5.9
BMI (kg/m2)	26.3 ± 3.8	27.2 ± 3.9	25.9 ± 3.7
eGFR (mL/minute/1.73 m2)	78.6 ± 24.3	62.3 ± 22.1	86.7 ± 21.5
UACR (mg/g)	85 (32-268)	348 (126-856)	52 (18-142)
HbA1c (%)	8.0 ± 1.5	8.9 ± 1.6	7.6 ± 1.3
SUA (μmol/L)	378 ± 84	425 ± 89	356 ± 76
hs-CRP (mg/L)	2.4 (1.2-4.6)	4.2 (2.1-7.8)	1.8 (0.9-3.2)
TyG index	8.9 ± 0.7	9.4 ± 0.7	8.7 ± 0.6
NLR	2.5 (1.8-3.6)	3.6 (2.4-5.2)	2.1 (1.6-2.9)
Diabetic retinopathy	71 (15.5)	41 (28.9)	30 (9.5)

BMI: Body mass index; eGFR: Estimated glomerular filtration rate; UACR: Urinary albumin-to-creatinine ratio; HbA1c: Hemoglobin A1c; SUA: Serum uric acid; hs-CRP: High-sensitivity C-reactive protein; TyG: Triglyceride-glucose; NLR: Neutrophil-to-lymphocyte ratio.

Independent predictors of DKD progression

Through least absolute shrinkage and selection operator regression, we recognize 12 candidate variables and then independently with multivariable Cox regression fit into eight independent predictors. The strongest were UACR ≥ 300 mg/g (HR = 4.15, 95%CI: 2.68-6.43) and eGFR < 60 mL/minute/1.73 m² (HR = 3.87, 95%CI: 2.41-6.22). We found that the metabolic predictors of ≥ 8.5% HbA1c (HR = 2.76, 95%CI: 1.79-4.25), ≥ 420 μmol/L SUA (HR = 2.34, 95%CI: 1.52-3.61), ≥ 9.0 TyG index (HR = 1.94, 95%CI: 1.28-2.94). The data also showed that the inflammatory marker hs-CRP ≥ 3.0 mg/L (HR = 2.18, 95%CI: 1.43-3.32) and NLR ≥ 3.0 (HR = 1.87, 95%CI: 1.23-2.85) were independently significant as well as diabetic retinopathy (HR = 2.56, 95%CI: 1.68-3.91; all P < 0.005). Figure 1 displays the results.

Figure 1 Forest plot of independent predictors for diabetic kidney disease progression. Multivariable Cox regression analysis identified eight independent predictors with their corresponding hazard ratios (HRs) and 95%CI. The vertical dashed line represents the null effect (HR = 1.0). Variables are ordered by the magnitude of their HRs. DKD: Diabetic kidney disease; NLR: Neutrophil-to-lymphocyte ratio; TyG: Triglyceride-glucose; CRP: C-reactive protein; HbA1c: Hemoglobin A1c; eGFR: Estimated glomerular filtration rate; UACR: Urinary albumin-to-creatinine ratio; HR: Hazard ratio; LASSO: Least absolute shrinkage and selection operator.

Nomogram development and performance

A nomogram of the eight predictors was developed with scores assigned as follows: UACR ≥ 300 mg/g (55 points), eGFR < 60 mL/minute/1.73 m² (52 points), HbA1c ≥ 8.5% (38 points), diabetic retinopathy (35 points), SUA ≥ 420 μmol/L (32 points), hs-CRP ≥ 3.0 mg/L, TyG index ≥ 9.0 points and NLR ≥ 3.0 points as well as total score these predictors in order of CDNI values between 293 according to CDNI value system ranged from 0 point to 293 points respectively. The model demonstrated good discrimination (C-index 0.843, 95%CI: 0.807-0.879) and time-dependent AUCs of 0.861, 0.852 and 0.843 at one, two and three years respectively. There was excellent calibration at all time points [slopes: (1) 0.98; (2) 0.96; and (3) 0.94; intercepts approximately zero]. Internal validation via bootstrap (1000 iterations) gave an optimism-corrected C-index of 0.837 with little over-fitting. Figure 2 shows performance metrics.

Figure 2 Nomogram and performance evaluation. A: Nomogram for predicting 1-year, 2-year, and 3-year progression risk in diabetic kidney disease. To use the nomogram, locate each predictor value on the corresponding axis, draw a vertical line to the Points axis to determine points, sum all points, and locate the total on the total points axis. Draw a vertical line down to the risk axes to read the predicted progression probabilities; B: Time-dependent receiver operating characteristic curves showing excellent discrimination at 1 year, 2 years, and 3 years; C: Calibration curves demonstrating close agreement between predicted and observed probabilities. DKD: Diabetic kidney disease; NLR: Neutrophil-to-lymphocyte ratio; TyG: Triglyceride-glucose; CRP: C-reactive protein; HbA1c: Hemoglobin A1c; eGFR: Estimated glomerular filtration rate; UACR: Urinary albumin-to-creatinine ratio; ROC: Receiver operating characteristic; AUC: Area under the curve.

Risk stratification and clinical utility

Patients were divided into low-risk (0-120 points, n = 156), intermediate-risk (121-200 points, n = 152), and high-risk (> 200 points, n = 150). Cumulative progression rates at three years were 8.7%, 28.4% and 62.3%; log-rank P < 0.001. The HR for intermediate-risk and high-risk groups compared to low-risk patients was 4.1 (95%CI: 2.3-7.3) and 12.8 (95%CI: 7.3-22.4). Compared to “treat all”/”treat none” strategies, decision curve analysis had greater net benefit, especially at threshold probabilities of 20%-50%. The performance was similar across subgroups (age, sex, and duration of diabetes) with C-indices ranging from 0.821 to 0.856; all interaction P > 0.10 (Table 1).

External validation

In an external cohort (n = 215), the nomogram had a C-index of 0.826 (95%CI: 0.779-0.873) with AUCs of 0.845, 0.833 and 0.821 at 1 year, 2 years and 3 years respectively. Calibration was excellent across all time points, with slopes of 1.02, 0.99, and 0.97 (all intercepts close to zero). Rates of 3-year progression among the three risk groups were similar to those observed in the derivation cohort (9.3%, 26.7% and 61.8%, Table 3), supporting generalizability of our nomogram.

Table 3 External validation cohort performance metrics (n = 215).

Metric	Value	95%CI
Discrimination
Overall C-index	0.826	0.779-0.873
Time-dependent AUC at 1 year	0.845	0.798-0.892
Time-dependent AUC at 2 years	0.833	0.783-0.883
Time-dependent AUC at 3 years	0.821	0.769-0.873
Calibration
Calibration slope at 1 year	1.02
Calibration slope at 2 years	0.99
Calibration slope at 3 years	0.97
Risk stratification (3-year progression rate)
Low-risk group	9.3%
Intermediate-risk group	26.7%
High-risk group	61.8%

Calibration intercepts were close to zero at all time points. Risk stratification results closely mirror the derivation cohort findings. AUC: Area under the curve.

Comparison with existing models

The newly developed nomogram demonstrated significantly superior discriminative performance compared to the KFRE (C-index: 0.843 vs 0.781, P = 0.003), the KDIGO risk categorization (C-index: 0.843 vs 0.765, P < 0.001), and the CKD Progression Score (C-index: 0.843 vs 0.748, P < 0.001) in the derivation cohort. Furthermore, it exhibited significantly better calibration, particularly in intermediate-risk patients, as evidenced by a calibration slope closer to the ideal value of 1.0 (0.96 vs 0.84-0.89 for other models, all P < 0.05). The integration of DKD progression-specific metabolic and inflammatory biomarkers was associated with a significant net reclassification improvement of 23.6% vs the KFRE, 26.8% vs the KDIGO model, and 29.3% vs the CKD Progression Score (all P < 0.001). These consistent advantages in discrimination, calibration, and reclassification were successfully replicated in the external validation cohort, confirming the robust additional prognostic value provided by the comprehensive biomarker panel incorporated into our model (Table 4).

Table 4 Comparison with established risk prediction models.

Model	Derivation C-index	Validation C-index	Calibration slope	NRI
Our nomogram	0.843 (0.807-0.879)	0.826 (0.779-0.873)	0.96	Reference
KFRE	0.781 (0.738-0.824)	0.768 (0.716-0.820)	0.89	23.6%
KDIGO model	0.765 (0.721-0.809)	0.752 (0.698-0.806)	0.87	26.8%
CKD Progression score	0.748 (0.703-0.793)	0.735 (0.680-0.790)	0.84	29.3%

Values are presented with 95%CIs where applicable. All comparisons with our nomogram: P < 0.05. KFRE: Kidney Failure Risk Equation; KDIGO: Kidney Disease: Improving Global Outcomes; NRI: Net reclassification improvement; CKD: Chronic kidney disease.

DISCUSSION

DKD has become a significant health challenge worldwide, impacting the lives of millions of patients with diabetes around the globe and is considered to be the leading cause of ESRD. Based on the diabetes epidemic, the burden of DKD is increasing and prevention approaches have often failed at reducing risk targeting[20]. Here, we addressed this pressing clinical unmet need by developing and validating an integrative robust risk stratification model combining a large number of metabolic and inflammatory biomarkers with classical renal parameters to predict the progression of DKD in patients with type 2 diabetes. In this study, we developed and validated a nomogram to predict the progression of DKD using clinical and laboratory data at the time of enrollment (baseline) in our cohort. Our results demonstrate that the newly developed nomogram is able with good discrimination (C-index 0.843) and calibration to predict DKD progression associated with a short follow-up period over the next 3 years between baseline assessment. It shows that the model can accurately classify diabetic patients into 3 distinct risk groups with significantly different rates of progression to DKD (8.7%, low; 28.4%, intermediate; 62.3% high). The model was robust and generalizable with external validation in an independent cohort. Interestingly, when we compared our nomogram with some well-known high-established risk prediction tools, it outperformed the latter which is likely due to a more complete reflection of the biological aspects of DKD pathophysiology being represented as metabolic and inflammatory biomarkers in addition to eGFR and albuminuria that are not fully captured by these markers[22,23]. Consistent with prior evidence establishing eGFR and albuminuria as pillars of CKD risk stratification, identifying baseline eGFR and albuminuria as the variables most predictive in our model confirms these findings. KDIGO guidelines[3] highlight their prognostic value of the eGFR and albuminuria categories on adverse kidney outcomes. Our result showing almost four-fold increased risk of progression for eGFR < 60 mL/minute/1.73 m² emphasizes the crucial role of early detection and intervention in patients with established kidney function decline. Likewise the high predictive value of macroalbuminuria (UACR ≥ 300 mg/g) highlights the long established role of proteinuria as a marker and mediator for progressive kidney injury in diabetes. Within our model, glycemic control reflected by HbA1c was shown to be a significant independent predictor. Building on the DCCT[24] paradigm, the DCCT/EDIC studies[25,26] have unequivocally demonstrated that intensive glycemic control reduces microvascular complications (including specifically nephropathy) related to type 1 diabetes during and after active intervention over decades. Though related data in type 2 diabetes have been less straightforward due to heterogeneous glycemic effects on kidney outcomes, the association between HbA1c ≥ 8.5% as an independent predictor of DKD progression underscores the need for appropriate glycemic control during treatment. The pathomechanism of hyperglycemia-induced renal injury is multifactorial: (1) Involvement of protein kinase C; (2) Enhancement of advanced glycation end-products formation; (3) Production of oxidative stress[26,27]; and (4) Activation of the inflammatory cascade. The incorporation of metabolic biomarkers, such as TyG index and SUA, is a uniquely new feature in our regression prediction model. The TyG index derived from fasting TG and glucose[27] is an insulin-sensitive marker that contributes to the pathogenesis of DKD through several pathways including endothelial dysfunction, inflammation, and activation of the renin-angiotensin-aldosterone system. Hyperuricemia was increasingly being recognized[28-30] as a modifiable risk factor for CKD progression. High uric acid levels exacerbated kidney injury through crystal-independent processes such as endothelial dysfunction, oxidative stress, inflammation and activating the renin-angiotensin system. Several recent intervention trials have demonstrated that uric acid-lowering therapy can impair CKD progression in certain high-risk patient groups, but the definitive evidence for DKD is limited. The determination of inflammatory markers such as hs-CRP and NLR as independent prognostic indicators emphasizes the pivotal significance of chronic low-grade inflammation in DKD development and aggravation. The independent association of elevated hs-CRP and NLR with DKD progression underscores the contributory role of chronic, low-grade inflammation in disease pathogenesis. This inflammatory milieu, driven by factors such as metabolic dysregulation and oxidative stress, can directly promote renal injury through multiple pathways, including endothelial dysfunction, podocyte damage, and tubulointerstitial fibrosis. The NLR, a readily calculable index from routine blood counts, serves as a practical and inexpensive marker of this systemic inflammatory state, potentially enhancing risk assessment at the point of care. The NLR, a readily calculable physiological marker based on the routine complete blood count, is an indicator of systemic inflammatory status that has become recognized as prognostically relevant for many cardiovascular and metabolic disease processes[31]. The independent association between increased NLR and DKD progression suggests that a simple inexpensive marker may supplement clinical risk assessment at the point of care.

Our nomogram offers a number of important advantages over existing DKD risk prediction tools. Reasons for this immediate utility include, first, that it incorporates a comprehensive panel of predictors covering traditional renal factors, glycemic control, metabolic dysregulation and inflammatory status all areas in which DKD mechanisms are implicated. Second, all predictors used in this prediction model are easily available from standard clinical evaluations and routine laboratory tests enabling implementation in various health care systems without the need for more specific or costly biomarker measurements. Third, the nomogram allows for a clear determination of high and low risk which can inform clinical decision-making regarding both escalation of treatment, and frequency of monitoring. We should discuss a few important caveat about predictor heterogeneity and precision. Both obesity and dyslipidemia are also epidemiologically associated with DKD, but these factors were excluded as independent predictors of progression in the multivariable model, probably because their sequelae on renal injury are accounted for by eGFR, UACR and inflammatory markers included in our model. This finding is in harmony with previous prediction modeling studies which indicated that conventional metabolic risk factors contribute little additional prognostic information once direct measures of renal function and general inflammation are included in the assessment. In a similar manner, these predictors as individual variables also provide moderate statistical precision based on their relatively wide confidence intervals (NLR HR = 1.87, 95%CI: 1.23-2.85; TyG index HR = 1.94, 95%CI: 1.28-2.94) and should be interpreted cautiously with respect to their independent contributions during prediction of the outcome under investigation as well, But their independent predictive value needs further validating in larger cohorts, and their component of the composite nomogram is justified only by some incremental improvement in overall model discrimination and net reclassification.

Implications for precision medicine: Our risk stratification model enables tailored strategies in the management of DKD. In this regard, patients flagged as high-risk may stand to gain from more proactive application of renoprotective therapies such as renin-angiotensin-aldosterone system inhibitors, sodium-glucose cotransporter 2 inhibitors and mineralocorticoid receptor antagonists; shorter monitoring intervals; and referral to nephrology specialists at a much earlier stage. In addition, low-risk patients could be monitored less frequently to reduce the burden of healthcare on the community and patients while maintaining safety. In patients with intermediate risk, it is possible that we could be especially effective at targeting reversible risk factors including glycemic control, blood pressure, and inflammation to prevent progression. In particular, we put forward the below risk-stratified management framework following with the nomogram categories. Low-risk patients (0-120 points) should receive standard monitoring at 6-12 months intervals, with optimization of glucose and blood pressure control per guideline targets. Intermediate-risk patients (121-200 points) should undergo closer follow-up every 3-6 months, intensification of renoprotective therapies (e.g., renin-angiotensin-aldosterone system inhibitors and sodium-glucose cotransporter 2 inhibitors in the absence of prior prescription), and targeted management of modifiable risk factors such as glycemic control, hyperuricemia, and/or elevated inflammatory markers. Early nephrology referral, integrated multidisciplinary care involving nephrologists as well as endocrinologists and cardiologist would be more appropriate for stage 2 DKD patients at high-risk (> 200 points); furthermore, stage 2 patients ranking top among the list should receive priority for participation in clinical trials evaluating future renoprotective agents. These recommendations are consistent with KDIGO guidance on risk-based monitoring interval or frequency and can be viewed as a initial general framework, individualized in accordance to patient preference, tolerability and resource availability. Additionally, our model has important implications for the design of clinical trials and the management of population health. As an example of utility, the model could have been used to select high-risk patients with DKD maximally suitable for enrollment in interventional trials testing new renoprotective therapies, thereby increasing the efficiency and statistical power of such future trials. At the population level, such a model might be utilized to better plan screening programs and resource allocation by applying intensive intervention in high-risk individuals.

Potential advantage of our nomogram over existing scores

The uniqueness of our study population further emphasizes the distinctive value of our nomogram compared to existing prognostic models. In contrast to the narrowly selected, protocol-defined cohorts of randomized clinical trials on which models like KFRE were at least in part developed, our derivation cohort resembles more real-world Chinese hospital populations with substantial comorbidity burden, 42.4% had hypertension and 38.2% dyslipidaemia, more typical of patients seen in routine clinical practice. The existing tools such as KFRE and KDIGO prognosis model were mostly derived from the Western or trial populations and may due underperform in Asian patients with different metabolic risk profiles and a higher incidence of hyperuricemia-related renal injury. In this heterogeneous and comorbid population, the higher net reclassification improvement of our nomogram (23.6% vs KFRE, P < 0.001) is clinically relevant since less accurate risk-stratification is likely to have limited therapeutic implications (e.g., mis-classifying a patient as low-risk with respect to all-cause mortality through an inaccurate prediction model). Another important consideration is that inflammatory markers such as hs-CRP and NLR may exhibit temporary variations with acute illness or intercurrent infections or even major changes in treatment. It is thus suggested that these values should be evaluated during clinical stability and consideration given to risk reclassification after designing patients from acute events or after continuing significant therapeutic changes. Future studies need to validate the nomogram in a community-based cohort, investigate how serial measurements of biomarkers could be used for dynamic risk stratification, and integrate such a tool in electronic health record-based clinical decision support systems.

This study has some limitations that need to be addressed. Our study has limitations inherent to its observational design, and being a hospital-based cohort some patients in the community with milder disease severity may not have been included. Secondly, biomarkers were measured at a single time point only (baseline); multiple measurements over time would possibly yield more prognostic information and enhance model performance. Third, although the follow-up period of 36 months was sufficient for intermediate-term progression assessment, longer-term outcomes such as ESRD or cardiovascular events may not have been captured. Fourth, we did not include genetic markers or emerging biomarkers such as kidney injury molecule-1 or fibroblast growth factor-23 that may improve risk prediction.

CONCLUSION

A novel and verty comprehensive risk stratification model for predicting development of DKD in type 2 diabetes patients was developed from a longitudinal cohort study and externally validated on two datasets. The incorporation of metabolic and inflammatory biomarkers into the evaluation improved risk prediction performance over established tools, as demonstrated in a stepwise multivariable analysis, and led to an internally validated nomogram with enhanced discriminative ability compared to traditional renal parameters with excellent calibration and clinical utility.