Development and internal and external validation of a nomogram model for predicting the risk of chronic kidney disease progression in IgA nephropathy patients

Participants

The primary study cohort comprised data from 443 primary IgAN patients who underwent kidney biopsies at Nanchong Central Hospital from January 1, 2015, to December 31, 2023. This data was utilized for model development and internal validation. In addition, data from 110 primary IgAN patients who underwent kidney biopsies at Guangyuan Central Hospital within the same healthcare system between January 1, 2024, and May 31, 2024, were used for external validation. The study adhered to the principles outlined in the Declaration of Helsinki and received approval from the Ethics Committee of two centers. Oral informed consent was obtained from each patient; however, written consent was waived due to the retrospective nature of the study (No: 2023-100).

The inclusion and exclusion criteria were as follows. Inclusion criteria: (1) Age ≥ 18 years; (2) confirmed initial diagnosis of IgAN by renal biopsy; (3) urinary protein excretion (UPE) ≥ 1 g/d with estimated glomerular filtration rate (eGFR) >15 mL/min/1.73 m². The exclusion criteria were as follows: (1) Secondary IgAN, including lupus nephritis, allergic purpura nephritis, and hepatitis B virus-associated glomerulonephritis; (2) variant forms of IgAN, including IgA deposition with minimal change disease, acute kidney injury, and rapidly progressive glomerulonephritis; (3) pregnancy; (4) follow-up for <3 months and incomplete clinical data (Fig. 1).

Outcome and variables

The outcome variable in this study was the stratification of risk for progression of CKD. Here, we defined the outcome variable according to the 2021 KDIGO guidelines (Beck et al., 2023) and as outlined in the study by Canney et al. (2021). Low risk of CKD progression was defined as urinary protein <1 g/d and a decrease of at least 25% from baseline after renal biopsy and treatment for 90 days. High risk of CKD progression was defined as urinary protein ≥ 1 g/d after renal biopsy and treatment for 90 days, or a decrease in urinary protein less than 25% from baseline.

This was a cross-sectional survey that utilized baseline information from hospitalization records at the time of biopsy, and obtained through an electronic medical record system. The demographic and clinical data included sex, age, body mass index (BMI), systolic blood pressure (SBP), diastolic blood pressure (DBP), medical history, complications, and medication usage records. Laboratory data included neutrophil count, lymphocyte count, hemoglobin, platelet, serum albumin, serum creatinine, eGFR (calculated using the Chronic Kidney Disease Epidemiology Collaboration equation (Levey, Inker & Coresh, 2014)), serum uric acid, serum cystatin C, total cholesterol, triglycerides, serum high-sensitivity C-reactive protein (Hs-CRP), serum immunoglobulin and complement levels, and UPE. Renal biopsy data included Oxford Classification scores, which consist of mesangial hypercellularity (M): M0 ≤ 0.5, M1 > 0.5; endocapillary hypercellularity (E): E0 = absent, E1 = present; segmental glomerulosclerosis (S): S0 = absent, S1 = present; tubular atrophy/interstitial fibrosis (T): T0 < 25%, T1 = 25%–50%, T2 > 50%; and cellular/fibrocellular crescents (C): C0 = absent, C1 < 25%, C2 > 25% (Trimarchi et al., 2017). The histopathological findings of all renal biopsy specimens were independently evaluated by two pathology experts from the corresponding hospital. Decisions regarding supportive therapy and/or immunosuppressive treatment were made based on the judgment of the attending nephrology specialists.

Development and validation of the nomogram

Before constructing the nomogram model, we randomly divided the patient data from the primary cohort into a training set and an internal validation set in a 7:3 ratio using R (dplyr 1.1.4). The nomogram was then developed using the training set data. Univariate logistic regression analysis and Forward Selection for Multivariate Logistic Regression were performed with SPSS 27.0 to identify independent predictors with statistical significance (P < 0.05). Subsequently, BMI, UPE, and T were used to construct the nomogram.

Internal and external validation were conducted on the nomogram. Model predictive accuracy was evaluated using 1,000 bootstrap resamples in R (rms 6.8.1). Receiver operating characteristic (ROC) curves were generated with R (pROC 1.18.5) to assess the model’s discriminative performance, while decision curve analysis (DCA) curves were created using R (rmda 1.6) to evaluate the model’s clinical utility. Additionally, we conducted a subgroup analysis based on different treatment regimens.

Statistical analysis

Statistical analysis was executed utilizing SPSS 27.0, R 4.4.1. Continuous variables were assessed for normality using the Shapiro–Wilk test. Quantitative characteristics that were normally distributed were described as means and standard deviations, while those not normally distributed were reported as the median and interquartile range (IQR), and differences between groups were assessed through the Wilcoxon rank sum test. Qualitative characteristics were described as frequencies (percentages), and differences between groups were assessed through Pearson’s Chi-squared test or Fisher’s exact test. Logistic regression was conducted on the training set data to screen for key influencing factors, and the nomogram was generated using R software. The predictive capability of the model was assessed through the area under curve (AUC), calibration curve, and DCA. P < 0.05 was defined as a statistically significant difference.

Baseline characteristics

With a rigorous inclusion and exclusion screening process, the study ultimately enrolled 553 primary IgAN patients, who were divided into 310 patients for the training set, 133 for the internal validation set, and 110 for the external validation set. Among them, 213 patients (38.5%) did not achieve remission of proteinuria at least 90 days of treatment, indicating a high risk for CKD progression. The study population predominantly consisted of young and middle-aged women, with an average age of 36 (28–48) years, and 56.2% of the patients were female. Specific baseline data for the study population are detailed in Table 1.

In the training set, patients in the high-risk group exhibited significantly higher levels of BMI, DBP, neutrophil count, serum creatinine, serum cystatin C, cholesterol, triglycerides, hs-CRP, serum C3, UPE and hematuria compared to those in the low-risk group. Furthermore, the high-risk group demonstrated a higher prevalence of diabetes, CKD3-4 stage and nephrotic syndrome. They also displayed more severe renal tubular atrophy/interstitial fibrosis and greater formation of cellular/fibrocellular crescents. Meanwhile, the high-risk group had lower levels of albumin, eGFR and serum IgG (Table 2).

Identification of risk factors

Univariate logistic regression analysis was applied to analyze the baseline variables of the training set. The results showed that 17 variables, including BMI, DBP, neutrophil count, albumin, serum creatinine, eGFR, serum cystatin C, triglycerides, serum IgG, serum C3, UPE, hematuria, CKD stage, nephrotic syndrome, diabetes, T, and C, were statistically significant factors (P < 0.05). Multivariate logistic regression analysis was then executed on those characteristics (Table 3).

The multivariate logistic regression model identified three risk factors independently correlated with CKD progression risk: BMI (OR = 1.109, 95% CI [1.042–1.182], P = 0.001), UPE (OR = 1.502, 95% CI [1.268–1.779], P < 0.001), and T1/2 (OR = 2.134, 95% CI [1.226–3.714], P = 0.007). The remaining factors did not show significant statistical significance (P > 0.05, Table 3).

Establishment and validation of the nomogram

A nomogram was created using the results of the multivariate logistic regression analysis, which identified three risk factors. This nomogram visualizes the practical use of the model using a random sample (Fig. 2). The nomogram model assigns scores to each independent variable based on their contribution to the outcome event (as indicated by the magnitude of regression coefficients). These scores are then summed to obtain a total score, which is converted into a probability score. Higher scores indicate a higher risk of CKD progression. An example of the nomogram applied to a random sample is illustrated in Fig. 2.

We employed R (pROC 1.18.5) and R (rms 6.8.1) to generate ROC curves and calibration curves for evaluating the predictive performance of the model on the training set and validation set. The results demonstrate that the nomogram model exhibits good discriminative ability, with AUCs of 0.746 (95% CI [0.691–0.8]), 0.764 (95% CI [0.68–0.85]), and 0.749 (95% CI [0.65–0.85]) for the training set, internal validation set, and external validation set, respectively; subgroup analysis validation based on treatment measures also showed a satisfied AUC, as depicted in Fig. 3.

The calibration curves fitted well with the ideal curve in the training set and the internal and external validation sets, indicating a high level of consistency between the predicted probabilities and actual event rates of the model, as shown in Fig. 4.

The model demonstrated good predictive accuracy in discrimination and calibration; however, its clinical utility remained uncertain. To validate the clinical applicability of the model, we plotted DCA curves using R (rmda 1.6). The DCA curves revealed that the net clinical benefit of the model decreased as the threshold probability increased. Specifically, when the threshold probability values ranged from 1.0% to 88.0% for the training set, and 1.0% to 68.0% for the internal validation set, and 1.0% to 95% for the external validation set, the model provided a net benefit that surpassed both “treat all” or “treat none” strategies, as depicted in Fig. 5.