Development and validation of a predictive model for end-stage renal disease risk in patients with diabetic nephropathy confirmed by renal biopsy

Introduction

Diabetes is a common metabolic disorder around the world (Liu et al., 2019). In China, the overall prevalence of adult diabetes in 2013 was reached approximately 10.9% (Wang et al., 2017). Diabetic nephropathy (DN), which is one of the most important complications of diabetic patients, is found in approximately 35% of type 2 diabetes mellitus (DM) cases and increases mortality (Wanner et al., 2016). In Europe and the United States, diabetic kidney disease (DKD) accounts for about 45% of the patients with end-stage renal disease (ESRD). DKD has become the leading cause of ESRD (Atkins & Zimmet, 2010). The prevalence of DKD in China is also on the rise, and it has become the second cause of end-stage renal disease. As we all know, renal biopsy is the gold standard for diagnosing the pathological patterns. The process from the confirmation of DN by renal biopsy to the progression of renal insufficiency is very short. The median renal survival time was 37.2 months for patients with proteinuria in the nephrotic range and 73.2 months for non-nephrotic range proteinuria (Jiang et al., 2019). Because of its complex metabolic disorder, once it develops into end-stage renal disease, it is often more difficult to treat and control than other kidney diseases, so timely prevention and treatment is of great significance for delaying the progression of diabetic nephropathy. It is imperative to identify patients with DN early and implement targeted management to prevent this progression.

There are many previous studies on the development and validation of risk models that predict the progression of chronic kidney disease (CKD). Renal risk prediction models already exist for patients with advanced CKD (CKD; stages 3–5), such as KFRE model, which was established by Tangri et al. in 2011 (Tangri et al., 2011a). Without considering the etiology of CKD, the model just used four clinical variables (age, gender, glomerular filtration rate and urinary albumin/urinary creatinine ratio) to make predictions. It has been validated in more than 30 countries, but still has not been validated in China; besides, it only targets advanced CKD patients other than DN (Yamanouchi et al., 2018). Another renal risk equation has recently been published for those with type 2 diabetes mellitus (T2DM) based on data from the ADVANCE clinical trial (Jardine et al., 2012), but they do not touch the indicators such as serum creatinine (Scr), pathological parameters or albuminuria.

However, there are some limitations of the existing models. First, the models were predominantly derived from Caucasian populations and the general population (Echouffo-Tcheugui & Kengne, 2012). Their validity in Asians with T2DM remains unclear due to limited data on this population. Second, most of the subjects had T2DM and no clear diagnosis of DN. There is a need to develop a prediction model that can be used to identify patients who need therapy to arrest progression during the time between renal biopsy and the occurrence of ESRD.

We therefore aimed to derive and validate a model to predict the 3-year risk of end-stage renal events, including dialysis, renal transplantation, or death from renal failure, among people with DN without advanced kidney disease within a secondary care context to prevent or slow the progression from DN to ESRD.

Materials & Methods

Study population

Among the 968 patients with T2DM who underwent renal biopsy between February 2012 and January 2015 at the First Affiliated Hospital of Zhengzhou University, a total of 478 patients were enrolled. Patients were divided into two cohorts: the derivation cohort, which included 75% of the patients and was used to develop the model, and the validation cohort, which included 25% of the patients and was used to test the performance of the model. The inclusion criteria were as follows: age of 18 years or older at the time of renal biopsy (male or female), biopsy-proven renal lesion and T2DM, and pathological findings showing isolated DN. The exclusion criteria were as follows: incomplete data or unclear medical history, with complications such as severe infection and malignancy, pre-existing renal replacement therapy, renal transplantation, or CKD stages 4 and 5 (estimated GFR, <30 mL/min/1.73 m²) at the baseline, and pathological findings showed DN combined with NDRD. The enrollment flowchart was showed in Fig. 1.

Results

During the 3 years of follow-up until December 2018, there were 225 outcome events (47.1%). Outcomes occurred in 187 patients (52.2%) in the derivation cohort and 38 patients (31.7%) in the validation cohort. The patients in the derivation cohort and the validation cohort were similar in terms of age, sex, duration of diabetes, duration of hypertension, Hb levels, HbA1c levels, Log ACR, Tcr, Scr, TG levels, HDL-cholesterol levels, LDL-cholesterol levels, serum lipid levels, eGFR, BNP levels and CKmb at baseline. The patients in the validation cohort had lower CRP levels, ESR levels, 24UTP, Cystatin C and PTH at the baseline than those in the derivation cohort (p < 0.05) (Table 1).

Discussion

We have developed and validated a prediction model for the risk of ESRD in patients with a pathological diagnosis of DN. The clinical-pathological model showed good performance among the several models. The AUC or C-statistic for clinical-pathological model was 0.865 in the derivation cohort and 0.876 in the validation cohort, showing good discriminatory performance, and the calibration was good, with p = 0.909. Our study demonstrated that lower eGFR, higher cystatin C levels, higher BNP, higher Log ACR level and higher pathological grade significantly increased the risk of ESRD in patients with DN. The model incorporating these risk factors can potentially be a simple medical calculator (shown in Fig. 2) for use in clinical practice. Our findings also partially overlap with those from previous studies on prognostic factors in patients with T2DM. For example, in the study of Chinese patients with T2DM and nephropathy in Hong Kong (Keane et al., 2006), the predictors of ESRD included eGFR, uACR and hematocrit (Yang et al., 2006). In the Action in Diabetes and Vascular Disease: Preterax and Diamicron MR Controlled Evaluation (ADVANCE) study, the predictors were sex, diabetic retinopathy, eGFR, uACR, SBP, HbA1c and age at the completion of formal education (Jardine et al., 2012).

Actually, during the long-term clinical practice, DN is usually be given a clinical diagnosis just by the presence of severe edema, macro albuminuria and renal insufficiency in patients with diabetes, so there have been still a lack of studies on pathological changes in DN (Yamanouchi et al., 2018). Even after Tervaert et al. (2010) developed a consensus classification of DN on behalf of the Renal Pathology Society, a limited number of studies on the evaluation of the prognostic value of this classification began to emerge (Mise et al., 2014; Oh et al., 2012; Okada et al., 2012), but no studies have been done to add renal pathology features to the prognostic risk assessment of DN. The importance and necessity of renal biopsy in diabetic patients is still controversial during the practical work (Fiorentino et al., 2017). However, a positive attitude toward renal biopsy is recommended for DM patients. The main clinical significance is probably related to its ability of distinguishing DN from other renal diseases and categorizing the related renal pathology. Indeed, several researches have showed that renal biopsy is meaningful for differentiating pure DN from nondiabetic renal disease (NDRD) because of the different renal outcomes (Chang et al., 2011; Fiorentino et al., 2017; Sharma et al., 2013; Wong et al., 2002).

In our final prediction model, 5 indicators of renal pathological grade, cystatin C, eGFR, BNP and Log ACR were included. In many previous studies (Chen, Wada & Chiang, 2017; Kulasooriya et al., 2018), uACR was generally considered to be a key indicator of the development and prognosis for DN. In our study, it was statistically significant in the univariate analysis and the multivariable logistic regression analysis. The conclusions are consistent with previous studies. In addition, previous studies have shown that BNP was associated with rapid decline of kidney function and incident CKD (Bansal et al., 2015; Mishra et al., 2013). In the study of patients with type 2 diabetes mellitus, showed that elevated NT-proBNP and TnT levels were independently associated with higher risk of ESRD (Desai et al., 2011).

As a classical predictive model, KFRE model is applicable to patients with CKD stage 3–5 (Tangri et al., 2011b), while the model in our study is applicable to patients with diabetic nephropathy diagnosed by renal biopsy pathology. To some extent, the clinical-pathological model in our study is more accurate in population targeting, and its predictive ability is better than KFRE model for patients with type 2 diabetes. However, the KFRE model is a large international study based on multi-center and multi-race, which is more applicable in CKD patients. The model of this study needs more multi-center studies to continuously verify and improve.

One of the strengths of our study is that the members of the study population were diagnosed with DN, and the diagnosis was confirmed by renal biopsy. In view of the lengthy course of kidney disease in patients with DN, the use of prediction models in both the early and late stages of kidney disease provides a wide window of opportunity for prevention. Our prediction tool can potentially help clinicians identify patients early, before CKD progression starts, thereby enabling them to provide intensive treatment to those at high risk. Second, our study was incorporated a cohort of DN based on pathological changes, this reduced the heterogeneity of the study population to some extent. Our findings will also enrich the currently limited research on the field of pathological information in T2DM (Jardine et al., 2012; Keane et al., 2006; Yang et al., 2006). The third strength is the use of easily available clinical and laboratory information, which are routinely collected in clinical practice. Furthermore, the prediction tool is practical and can be easily applied in the clinical setting, and the presence of a risk prediction model and its integration into clinical practice guidelines have resulted in improved compliance with treatment guidelines and the promotion of personalized treatment.

There are also a few limitations in our study. First, the sample size was only moderate, and the observation period was not long. Second, as with all risk prediction models, it is necessary to conduct multicenter external validation studies before our results can be applied in the clinical setting. Third, our cohort came from the First Affiliated Hospital of Zhengzhou University and hence was not fully representative of the entire population of individuals with DN in China. This limits the generalizability of the findings to other cohorts. Last, the progression of DN to ESRD is a very complicated process, and there may be other risk factors that increase the risk of ESRD that were not explored in this study, such as genetic factors, socioeconomic status, different treatment programs, blood pressure control, glycemic control, diet, exercise and some pathological changes, such as exudative lesions or mesangial lysis, which Furuichi et al. (2018) reported to be strong predictors of ESRD. It is necessary to conduct further and more in-depth research.

Conclusions

We have developed and validated a model to predict the risk of progression to ESRD in patients diagnosed with biopsy-confirmed DN. The clinical-pathological model demonstrated that pathological grade, cystatin C, eGFR, BNP and Log ACR influenced the disease progression from DN to ESRD. The model performed well and can be a practical and convenient tool for early identification and prognostic prediction among high-risk patients with DN in clinical practice.

Supplemental Information