Background
Antineutrophil cytoplasmic antibody (ANCA)-associated vasculitis (AAV) comprises a group of autoimmune disorders, including microscopic polyangiitis (MPA), granulomatosis with polyangiitis (GPA), and eosinophilic granulomatosis with polyangiitis (EGPA), characterized by necrotizing small-vessel vasculitis with serum autoantibodies mainly against proteinase 3 (PR3) or myeloperoxidase (MPO). As one of the most common manifestations of AAV, ANCA-associated glomerulonephritis (ANCA-GN) presents in 80–100% of patients with MPA and 38–70% with GPA, which is typically characterized by pauci-immune necrotizing crescentic glomerulonephritis in renal histology [
1‐
3].
Left untreated, AAV is a life-threatening disease. Immunosuppressive therapy, in particular, corticosteroids in combination with cyclophosphamide or rituximab, has dramatically improved the outcome of AAV patients, but a significant proportion of patients still progress to end-stage renal disease (ESRD). In a recent study with a training cohort of 115 patients and a validation cohort of 90 patients, a renal risk score (RRS) was developed by Brix et al. to predict renal outcome for AAV patients [
4]. The RRS was based on a Cox model with clinical and pathological parameters at diagnosis, including proportion of normal glomeruli (N0 > 25%, N1 10–25%, N2 < 10%), proportion of interstitial fibrosis and tubular atrophy (IF/TA) (T0 ≤ 25%, T1 > 25%), and estimated glomerular filtration rate (eGFR) (G0 > 15 ml/min per 1.73 m
2, G1 ≤ 15 ml/min per 1.73 m
2). The RRS was calculated as the sum of the assigned points for each parameter (N1 = 4, N2 = 6, T1 = 2, G1 = 3 points). This score was designed to predict ESRD risk as low (0 points), medium (2 to 7 points), or high (8 to 11 points), and it proved to be a clinically applicable tool for early risk prediction of ESRD in ANCA-GN. Validation of the RRS was performed in 11 cohorts worldwide, comprising 37 to 252 patients, and showed its predictive value [
5‐
15]. However, regarding discrimination and calibration, the two key aspects of a systematic assessment of predictive performance for a model, most of the previous validation studies did not report discrimination [
5,
6,
8,
9,
12‐
15], and none of them reported calibration [
5‐
15].
In the current study, a systematic validation was performed in Chinese AAV patients, which showed that the RRS was an independent predictor for ESRD with good discrimination but less satisfactory calibration. Therefore, model modification was launched to improve the predictive performance (including discrimination and calibration) of the RRS. Furthermore, internal and external validations were performed to assess the extent of optimism and overfitting of the modified model.
Discussion
Renal involvement in AAV, namely, ANCA-GN, was associated with ESRD and poor outcome [
31‐
33]. In China, AAV was an important cause of secondary glomerular diseases and the leading cause of acute kidney injury in elderly patients who received renal biopsy [
34,
35]. Although the outcome of AAV has been dramatically improved by immunosuppressive therapy, a large number of patients still progressed to ESRD. Accordingly, it was of interest to establish and validate tools to predict renal outcome in patients with ANCA-GN.
The study of Berden et al. emphasized the significance of histological injury for renal outcome and demonstrated the ascending probability of progressing to ESRD from the focal to crescentic, mixed, and sclerotic categories [
22]. However, such a sequence of the probability of progressing to ESRD was inconsistent among different studies, which indicated that the predictive significance of the classification may not be quite stable in different cohorts and its clinical application was limited to some extent [
36,
37]. In addition, the histopathological classification was not predictive in the multivariable model involving eGFR, IF/TA and age [
4,
38]. To establish a more reliable predictive model, Brix et al. developed and validated an RRS system to predict renal survival in German cohorts [
4], and this model has been further validated in different cohorts worldwide [
5‐
15]. However, the systematic assessment of predictive performance, including discrimination and calibration, was limited in these studies. Therefore, in the current study, a systematic evaluation of the predictive performance of the RRS in a Chinese cohort of AAV was launched.
Compared with previous studies for validation of RRS in ANCA-GN [
5‐
15], the sample size of our study is the largest, with 272 patients. In addition, these 272 patients had a median of 26 (IQR, 19–36) glomeruli per biopsy, which was so-far the highest among these studies for the RRS, making the histopathological results more reliable and accurate. It was found in the current study that renal survival was significantly different among the high-, medium-, and low-risk groups, and Harrell’s
C-statistic was satisfactory. Multivariate Cox analysis showed that the RRS was an independent predictor for ESRD after adjusting for other parameters, including age, sex, and urinary protein. Therefore, the RRS proved to be prognostic and practical in Chinese AAV patients and facilitated its clinical prognostication.
However, despite the satisfactory discrimination, the calibration assessment showed that the predicted outcomes by the RRS were not quite consistent with the observed outcomes (Hosmer and Lemeshow test, P < 0.0001). In addition, as mentioned above, obvious heterogeneity existed among the patients within the same group. Accordingly, based on Brix’s model, we further developed a modified model in Chinese AAV patients. Compared with the primary model, the modified model had improved discrimination and calibration for renal outcome. There were only 3 parameters in this model, i.e., proportion of normal glomeruli, IF/TA and eGFR, making it easily available in biopsy-proven ANCA-GN. The internal and external validation showed good discrimination and no significant disagreement between predicted and observed outcomes for a follow-up of 36 months. To facilitate clinical application, an online risk prediction tool based on the modified model was developed for physicians. Once the proportion of normal glomeruli, IF/TA and eGFR are entered, the website will quantify the probability of an individual AAV patient progressing to ESRD.
There were some limitations in our study. First, since this was a study in Chinese AAV cohorts with predominant MPA and MPO-ANCA patients, considering the heterogeneity of ANCA status in different populations, it might limit the extrapolation of this modified model. Validation of the modified model deserved further investigation in international cohorts. Second, patients in the training cohort were retrospectively recruited from a single center, and therefore, our modified model was not comparable with the original score designed from a multicenter prospective cohort. Third, the calibration was improved with the modified model at 36-month follow-up (Hosmer and Lemeshow test, P = 0.2070) but still showed a significant difference of prediction at 60-month (Hosmer and Lemeshow test, P = 0.0092) and 120-month follow-up (Hosmer and Lemeshow test, P < 0.0001). ESRD risk calculated by the modified model might be overestimated for high-risk patients in models of 60- and 120-month follow-up.
Conclusions
In conclusion, a validation of the RRS was performed in our study, which demonstrated that the RRS was prognostic for ESRD but not fully satisfactory in Chinese AAV patients. Therefore, a modified model was established based on the RRS, with improved discrimination and calibration. Despite the not fully satisfying calibration of the modified model, this was an important investigation of the improvement for prediction of renal survival in Chinese AAV patients. Internal and external validation of the modified model showed good performance in Chinese AAV patients. An online risk tool derived from this model was developed, which may be practical to physicians.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.