Non-Invasive Prediction Tool for Non-Diabetic Kidney Diseases in Patients with Type-2 Diabetes Mellitus

Introduction

Diabetic kidney disease (DKD) is the most common cause of CKD in patients with Type 2 Diabetes Mellitus (T2DM).^1,2 There has been a shift in the spectrum of kidney diseases among T2DM patients >30 years.¹ Non-diabetic kidney disease (NDKD) has been increasing in prevalence.^2-4 Improved glycemic control, better diagnostic tools for early identification, and better use of renoprotective therapies have modified the classical course and renal outcomes. Additionally, an ageing population, increasing infections, malignancy, and autoimmune diseases have been attributed to the increased NDKD prevalence.^5,6

Identifying NDKD allows the detection of treatable or reversible components of renal dysfunction and slows CKD progression, which depends on multiple factors, including proteinuria, blood pressure control, other insults on the kidney leading to AKI, including drugs and hemodynamic factors and the presence of non-diabetic glomerular, interstitial or vascular ailments.^7,8 Clinical features suggestive of NDKD include a short duration of diabetes, absence of diabetic retinopathy, rapid decline in kidney function, active urinary sediment (e.g., RBC casts), nephrotic syndrome, or sudden onset kidney dysfunction. However, many factors like the undetermined onset of T2DM, the temporal relationship between hypertension and T2DM, varying renal-retinal relationship, changing DKD phenotype in the form of the increasing prevalence of non-albuminuric DKD, and the presence of microscopic hematuria in 1/3 of DKD, have puzzled clinicians, confounding their ability to differentiate DKD and NDKD clinically.^9-13

Kidney biopsy, the gold standard for identifying reversible non-diabetic pathology,¹⁴ is associated with risks.^15,16 DKD is a clinical diagnosis; however, a kidney biopsy is required in the presence of pointers towards the atypical course and features suggestive of NDKD.^17-19 However, ∼30% of those with diabetes undergoing biopsy based on these atypical clinical features eventually had DKD on histopathological examination.^3,4,19 A scoring tool that combines relevant clinical and laboratory parameters may help identify patients who are more likely to have NDKD. Such a tool can be used for shared decision-making during discussions with patients for the need of renal biopsy, which is essential for confirming diagnosis and identifying the specific NDKD type to guide appropriate therapy. The study aimed to create a non-invasive risk-scoring system using clinical and laboratory data to predict the presence of NDKD in patients with T2DM.

Materials and Methods

This was a hospital-based, analytical, observational study performed in a tertiary care hospital in Northern India. In the current study, we analyzed the biopsy data of consecutive T2DM patients who underwent biopsy between August 2005 and July 2022. All patients with T2DM ≥ 18 years who underwent kidney biopsy for various indications with a probable diagnosis of NDKD as per the treating physician’s decision were included. Patients with Type-1 DM, clinically apparent DKD, and pregnant and lactating females were excluded.

The population was categorized into two cohorts: (a) Developmental or derivational cohort, and (b) Validation or verification cohort. Patients who underwent biopsies (n=376) from 1^st August 2005 till 31^st July 2018 were considered in the derivational cohort. The subsequent study population of the next four years (n=162) was regarded as the validation cohort. All data were prospectively stored in the electronic medical records of the hospital information system. Categorical and continuous data including the demographics, associated comorbidities, diabetes-related microvascular and macrovascular complications nature of the syndromic renal presentation such as AKI, rapidly progressive renal failure (RPRF), estimated glomerular filtration rate (eGFR) at presentation, dialysis dependency, proteinuria, presence of microscopic hematuria, hypoalbuminemia, hypercholesterolemia, hypocomplementemia, and the physician’s indication of the biopsy were collected. The standard working definitions have been elaborated in the Supplementary File.^20-24 Laboratory variables were derived from universally implemented tests. Comorbidities (presence or absence, number, and type), including hypertension, cardiovascular disease, peripheral neuropathy, diabetic retinopathy, and cerebrovascular disease, were diagnosed according to ICD-10 codes.

All biopsies were processed per the standard recommendations.²⁵ A team of two pathologists reviewed them for changes in DKD and NDKD, NDKD type, degree of interstitial fibrosis and tubular atrophy, and vascular changes, including medial hyalinosis/arteriosclerosis. The DKD²⁵ and various NDKD diagnoses,^26-29 along with disease staging, were made per the standard criteria specified in the Supplementary File. Patients in each cohort were classified into three categories based on the biopsy findings: DKD, NDKD, and DKD+NDKD groups. The clinical and histological findings were compared between DKD, NDKD, and DKD + NDKD, and finally, clinical predictors of NDKD were obtained by statistical analysis.

Results

A total of 5,485 biopsies were performed during the study period; 559 patients had T2DM [Figure 1]. After excluding patients with missing data, 538 patients were included in the analysis. Of them, 376 patients who underwent biopsy in the first 13 years were included in the derivation cohort, and the remaining 162 were included in the validation cohort. Additionally, 152 patients from two other centers who met the inclusion criteria were included for multicenter external validation. The baseline characteristics of the derivation, internal validation, and external validation cohorts have been shown in Table 1.

Close

Figure 1:

Flow diagram of the study. DKD: Diabetic kidney disease, NDKD: Non-diabetic kidney disease.

Export to PPT

Generation of NDKD-Prediction model

The above variables, which were significantly associated with NDKD on univariate analysis in the derivation cohort, were selected and run in the multivariate binary logistic regression. By Backward-Wald’s stepwise elimination method, the following variables were found to be significantly associated with NDKD - Duration of DM <5 years (AOR = 1.96, 95% CI: 1.26-3.14, P = 0.003), absence of CAD (AOR = 2.1, 95% CI: 1.1-4.9, P = 0.047), absence of DR (AOR = 4.9, 95% CI: 2.9-8.4, P = <0.001), oliguria (AOR = 1.8, 95% CI: 1.1-3.1, P = 0.02), an acute rise in creatinine (AOR = 6.2, 95% CI: 3.4-11, P = <0.001) and low serum C3 level (AOR = 4.9, 95% CI: 2.08-11.6, P = <0.001) [Table 4]. The forest plot of the multivariate logistic regression analysis has been shown in Figure 2. A clinical scoring model has been developed to predict the likelihood of NDKD in patients with diabetes, incorporating variables that were found to be statistically significant in multivariate analysis. This model aims to aid clinical decision-making, particularly in identifying patients who may benefit from a diagnostic renal biopsy. The scoring tool is available for clinical use online and can be found at the following site https://script.google.com/macros/s/AKfyc.../exec

Table 4: Multivariate logistic regression analysis

Variable	p-value	Odds ratio (95% CI)	β coefficient
Duration of DM < 5 years	0.003	1.9 (1.26, 3.14)	0.76
HTN	0.22	0.73 (0.3, 2.11)	-
Onset of HTN after diabetes	0.94	0.11 (0.03, 1.86)	-
No CAD	0.047	2.1 (1.11, 4.96)	0.73
No DR	< 0.001	4.9 (2.9, 8.4)	1.6
Oliguria	0.02	1.8 (1.1, 3.1)	0.65
H/s/o Recent infection	0.63	1.2 (0.52, 1.44)	-
Presence of extra-renal manifestations	0.64	1.8 (0.82 2.26)	-
Acute rise in creatinine	< 0.001	6.2 (3.4, 11.0)	1.8
Rapidly progressive renal failure	0.58	1.23 (0.6, 2.68)	-
Dialysis requiring renal failure at presentation	0.06	1.79 (0.96, 3.3)	-
Microscopic hematuria	0.07	1.51 (0.95, 2.41)	-
Low serum C3 level	< 0.001	4.9 (2.08, 11.6)	1.6

DM: Diabetes mellitus, CAD: Coronary artery disease, DR: Diabetic retinopathy, HTN: Hypertension, H/s/o: History suggestive of, CI: Confidence interval

Close

Figure 2:

Forrest plot of multivariate logistic regression analysis showing Odd’s ratio and the 95% confidence intervals for the variables predicting the presence of Non-diabetic kidney disease. DM: Diabetes mellitus, CAD: Coronary artery disease; DR: Diabetic retinopathy, AOR: Adjusted odds ratio.

Export to PPT

A nomogram of the model containing these predictor variables was developed. The value of each patient is located on the axis of variables, and a straight line is drawn upward to determine the point value of each variable. For example, a patient can get 0 points if he has a duration of DM > 5 years and get about 20 points if he/she has a duration of DM < 5 years. Likewise, the points for each scoring variable will be calculated. Finally, the total points are obtained by summing the point values identified on the x-scale for each variable in the model. Suppose the total point of this patient is 150; in that case, the probability of NDKD occurrence can be predicted as 70% by a perpendicular line extending downward to the “predicted value” axis from 150 on the “Total points” axis [Figure 3].

Close

Figure 3:

Nomogram of multivariate logistic regression model. The value of each patient is located on the axis of variables, and a straight line is drawn upward to determine the point value of each variable. DM: Diabetes mellitus, CAD: Coronary artery disease, DR: Diabetic retinopathy.

Export to PPT

Area under the Receiver Operating Curve (AUROC) of derivation, internal and external validation of cohort

The model run on the derivation cohort revealed an AUROC of 0.816 (95% CI 0.758-0.846), as shown in Figure 4a. The best result in predicting NDKD was found at a cut-off score of 3 with a sensitivity of 86.3%, specificity of 82.5%, a positive predictive value (PPV) of 88.2% and a negative predictive value (NPV) of 74.3%, respectively.

Close

Figure 4:

Area under the receiver operating curve (AUC) of the NDKD predictor score while applied on the (a) developmental cohort, (b) Internal validation cohort, and (c) external validation cohort. NDKD: Non-diabetic kidney diseases, CI: Confidence interval, ROC: Receiver operating characteristics curve.

Export to PPT

Discussion

In this study, we developed and validated a predictive model to detect NDKD in patients with T2DM undergoing biopsy for various indications. The model, comprising the following clinical and laboratory parameters: duration of DM < 5 years, absence of CAD, absence of DR, presence of oliguria, an acute rise in the serum creatinine, and low serum complement C3 level, performed very well in predicting the NDKD. To the best of our knowledge, this might be the first study to develop a prediction model based on clinical and laboratory parameters to detect NDKD among people with DM undergoing biopsy.

Although kidney biopsy remains the gold standard for identifying the NDKD component, it is invasive. Moreover, nearly 1/3^rd of patients with T2DM who undergo a kidney biopsy with a premonition of NDKD eventually were found to have DKD on biopsy (3,4). Because the score was developed on those patients with clinical pointers towards NDKD and not on all patients with DM, it can be used to discuss the need for biopsy with patients who have atypical clinical pointers of NDKD, informing them of the need for biopsy to decide future treatment course. With a large sample size of 538 patients across 17 years, 376 patients’ data (derivation cohort) were used to develop the prediction score. The same was tested on 162 patients from the same center (validation dataset for temporal validation). Additionally, the performance of the score was assessed by multicentric external validation on 152 patients from two different centers from the southern part of India, The better performance of the score on the external validation dataset (AUC:0.883) compared to that on the internal validation cohort (AUC:0.869) was probably due to a higher AKI proportion among the NDKD sub-group and a relatively higher DKD proportion. Hence, the performance of the scoring tool might be particularly better in ruling out the DKD and, hence, the unwanted biopsies.

The clinical features predicting NDKD may vary widely depending on the clinical scenario. Among the various variables, an acute rise in serum creatinine followed by an absence of DR and low serum complement C3 levels had higher weightage as expected, suggesting the stronger association of NDKD. This could be because nearly 40% of patients with NDKD had ATIN, presenting as AKI or AKI on CKD. Likewise, the low serum C3 levels could also be getting a higher weightage as the diffuse proliferative glomerulonephritis picture (either IRGN/MPGN) contributes to the second most common NDKD in our study. The variables of the predictive model could be easily accessible, this allows for its applicability in daily clinical practice. Similar studies by Yang et al., and Zhou et al., also published a model to detect NDKD or DKD among T2DM with renal failure.^31,32 However, the uniqueness of the current study is to convert the final model into a bedside mobile application as a non-invasive tool aiding in analyzing the risk-benefit ratio of renal biopsy on a case-to-case basis. The model application and website details have been shown in Supplementary File. Furthermore, the uniqueness of the scoring tool lies in the variables selected in the model, which are clinically relevant and feasible to be applied in day-to-day practice and have been time-proven across various studies.^9,10,17 Besides, the variables used in developing the model were selected by univariate analysis and not randomly, which could be the reason for the robust performance of the model. Other major strengths of the study are the sample size, the wide timeline across which the derivation data were collected and the multi-centric external validation. The excellent performance of the score while applied in a separate validation cohort with an accuracy of 87% (on internal validation) and 88% (on external validation), which is quite different in its time period, is an example of its eminent suitability across the different patient populations. The exclusion of missing patients from the analysis prevented the distortion of results by the missing data. As the missing data is <5% of the study population, the exclusion of missing data shall not affect the outcome.^33,34 The other limitation was the fact that 1/3 of the developmental cohort and >1/2 of the external validation cohort had AKI as an indication for renal biopsy. The applicability of the study among people with diabetes with stable renal function needs to be reassessed in further external validation. Furthermore, the score is tailored for diabetics with a clinical need for kidney biopsy, not the entire diabetic population. It is developed and validated in those suspected of having non-diabetic renal pathology, reflecting the real-world scenarios for clinicians. The tool is not meant to eliminate the need for a kidney biopsy, but to support informed clinical decision-making, potentially prompting closer evaluation or biopsy in patients with a higher likelihood of NDKD.

To conclude, our clinical and laboratory parameter-based prediction model robustly predicted the presence of NDKD among T2DM patients with renal dysfunction.