Evaluation of the Risk Prediction Models in Predicting Kidney Outcomes in Antiglomerular Basement Membrane Disease

Abstract

Introduction

A previous study showed that the renal risk score (RRS) was transferrable to antiglomerular basement membrane (anti-GBM) disease and proposed a risk stratification according to the need of renal replacement therapy (RRT) and the percentage of normal glomeruli (N). Herein, we analyzed the risk factors associated with kidney outcomes in patients with biopsy-proven anti-GBM disease and evaluated these 2 prognosis systems.

Methods

A total of 120 patients with biopsy-proven anti-GBM disease with complete clinicopathologic and outcome data were analyzed.

Results

The median time to kidney biopsy was 41 days (interquartile range [IQR]: 22–63 days). RRT and N were the only independent predictors of end-stage kidney disease (ESKD). Patients with N ≥10% were more likely to achieve ESKD-free outcomes, even in the subcohort of patients who underwent posttreatment biopsies (P < 0.001). N and serum creatinine at presentation (cut-off values 750 μmol/l and 1300 μmol/l) were 2 independent factors for predicting kidney recovery. The RRS and the risk stratification tool exhibited predictive value for ESKD and could be transferred to patients with kidney biopsy following treatment (Harrell’s C statistic [C] = 0.738 and C = 0.817, respectively). However, a cross-over of outcomes among groups was observed in the risk stratification tool in long-term follow-up, when patients with RRT and N ≥10% achieved better kidney outcomes than those without RRT but N <10%.

Conclusion

Normal glomeruli percentage, even posttreatment, was a strong indicator for kidney outcomes, especially on long-term prognosis. Serum creatinine is a predictor for kidney recovery, independent of biopsy findings. The risk stratification tool for kidney survival was transferrable to patients with anti-GBM disease with biopsy following treatment in our cohort; however, this needs further validations for long-term outcomes.

Graphical abstract

Anti-GBM disease is a rare but severe autoimmune disorder characterized by the presence of pathogenic anti-GBM antibodies in the kidneys and lungs. The condition is marked by rapidly progressive glomerulonephritis with crescent formation and, in some cases, accompanied by pulmonary hemorrhage termed Goodpasture's disease.¹ Despite significant improvement in patient survival with aggressive immunosuppressive therapies, only one-third of the patients gain partial kidney recovery.² Approximately 60% to 80% patients entered ESKD with maintenance RRT within the first year of follow-up.2, 3, 4, 5, 6, 7

The clinicopathologic features associated with kidney prognosis have been studied in a few cohorts of patients with anti-GBM disease. Clinical indicators, such as oligoanuria, serum creatinine on diagnosis, and initial RRT, are important parameters linked to kidney outcome.6, 7, 8, 9 Kidney biopsy findings, particularly the number of normal glomeruli and the percentage of crescents, have also been identified as independent factors influencing kidney outcome.5, 6, 7, 8 Given the high risk of mortality and rapid loss of kidney function in untreated anti-GBM disease, immunosuppressive treatment should be initiated in most patients as soon as possible. However, the decision to initiate an intensive immunosuppressive regimen should consider the low likelihood of kidney recovery in patients requiring RRT at presentation or those with a high proportion of crescents (85%–100%) on kidney biopsy, as outlined by the Kidney Disease Improving Global Outcomes guidelines.¹⁰ Therefore, a prediction model for kidney outcomes would assist physicians in creating individualized regimens for anti-GBM disease in complex clinical and pathological conditions, similar to the models developed for other autoimmune kidney diseases.^11,12

RRS is a clinicopathologic scoring system utilized to predict ESKD in antineutrophil cytoplasmic antibody-associated vasculitis (AAV), incorporating 3 variables (percentage of normal glomeruli, degree of tubular atrophy and interstitial fibrosis [TA/IF], and estimated glomerular filtration rate [eGFR] at the time of kidney biopsy). This system categorizes patients into low, moderate, and high-risk groups.¹³ Kidney lesions in AAV are characterized by rapidly progressive glomerulonephritis with crescents formation, similar to those seen in anti-GBM disease.¹⁴ The RRS was transferred to predict kidney survival of anti-GBM disease recently by Floyd et al.¹⁵ The authors demonstrated that combining RRT at presentation with the histological parameter of N, in the setting of the risk stratification method, might be a valuable prognostic tool for predicting ESKD in patients with anti-GBM disease.¹⁶ Nevertheless, both models require evaluation for their predictive power regarding kidney outcomes before being widely utilized in practice. This study aims to identify potential predictors for kidney outcomes in patients with anti-GBM disease following treatment and to assess the performance of both RRS and the risk stratification model within our cohort.

Methods

Patients

A total of 120 patients diagnosed with biopsy-proven anti-GBM disease in Peking University Renal Division between 2005 and 2022 were retrospectively enrolled in this study. The inclusion criteria were as follows: (i) kidney biopsy showing linear IgG fluorescence along the GBM, excluding other causes such as diabetes and fibrillary glomerulonephritis; (ii) detection of anti-GBM antibodies in the sera; and (iii) complete clinical data, including kidney function (as measured by serum creatinine and eGFR) at presentation and during follow-up. Clinical parameters at baseline and follow-up, as well as treatment regimens, were recorded. The definition of ESKD, kidney recovery, and normal glomeruli were as follows, as described previously by Floyd et al.¹⁵: (i) ESKD is maintenance of RRT lasting at least 12 weeks and up to the last follow-up; (ii) kidney recovery is independence from RRT lasting at least 12 weeks during follow-up; (iii) normal glomeruli are glomeruli that did not exhibit any glomerulosclerosis, crescents, or fibrinoid necrosis. Biopsies were reviewed by an experienced nephropathologist who was blinded to the outcome. This study was approved by the Ethics Committee of the Peking University First Hospital and in accordance with the Declaration of Helsinki.

Statistics

The primary end point of this study was the cumulative percentage of patients who developed ESKD censored by the last follow-up. The secondary end point was kidney recovery censored by the last follow-up or ESKD.

The RRS was calculated as previously described.¹³ In brief, (i) the percentage of normal glomeruli in kidney biopsy (N2: <10% of N; N1: 10%–25% of N; and N0: >25% of N); (ii) the degree of TA/IF (T0: < 25% and T1: ≥25%), and (iii) the eGFR at the time of biopsy (G0: >15 ml/min per 1.73 m² and G1: ≤15 ml/min per 1.73 m²). Patients were classified into 3 different risk groups depending on their sum score (N1 = 4, N2 = 6, T1 = 2, and G1 = 3): low risk (0 points), moderate risk (2–7 points), and high risk (8–11 points). Each patient in the present cohort was calculated for the original RRS score and stratified by the risk stratification tool for kidney outcomes by Floyd et al. Univariable and multivariable Cox regression analyses were performed to identify predictors for kidney outcome. Age, sex, levels of anti-GBM antibodies, and all parameters with a P-value <0.05 in univariable analyses were further used to select predictors associated with outcomes in multivariable analyses. The performance of the original RRS model and risk stratification tool for kidney outcomes was evaluated by Kaplan-Meier analysis and discrimination capability (assessed by Harrell’s C statistic). The regression tree analysis was performed using the R package rpart (version 4.1.19; R Core Team, Vienna, Austria). P-value < 0.05 was considered significant.

Categorical data were expressed as number (relative frequencies), and differences were compared with Chi-square statistic or the Fisher exact test when appropriate. Continuous data were expressed as median and IQR, and differences were compared with Mann–Whitney U-test. All statistical analysis was performed using SPSS version 26.0 (IBM Corp., Armonk, NY), GraphPad Prism version 8.0 software (GraphPad Software, La Jolla, CA), and R v4.0.5.

Results

Baseline Characteristics and Clinical Outcome

One hundred twenty patients with biopsy-confirmed anti-GBM disease were included in the present study. Baseline characteristics are summarized in Table 1. The median age at presentation was 45 years (IQR: 33–56 years) with a male to female ratio of about 1:1. The median follow-up period was 27 months (IQR: 8–42 months). The median time from disease onset to kidney biopsy was 41 days (IQR: 22–63 days). Eighty-eight patients (73.3%) underwent kidney biopsy after treatment initiation, with a median time of 18 days (IQR: 12–27 days) between the onset of treatment and the biopsy.

Table 1.

Baseline characteristics

Characteristics	Total (N = 120)
Male sex	62 (51.7%)
Age, years	45 (33.0-56.0)
Serum creatinine, μmol/l	609 (287.6–948.2)
Pulmonary hemorrhage	26 (21.7%)
eGFR, ml/min per 1.73 m2	7 (4.5–19.4)
Oligoanuria	44 (36.7%)
Levels of anti-GBM antibody	143 (76.5–200)
Antibody
Mono anti-GBM	105 (87.5%)
Double positive antibody	15 (12.5%)
Initial RRT at presentation	81 (67.5%)
Glomeruli on biopsy	24 (18.0–32.0)
Percentage of normal glomeruli	9 (0–25.0)
Percentage of crescents in glomeruli (all types)	83 (64.2–94.3)
TA/IF
≤25%	64 (53.3%)
>25%	56 (46.7%)
RRS group
High	66 (55.0%)
Moderate	42 (35.0%)
Low	12 (10.0%)
Risk stratification
1 (no RRT, N ≥ 10%)	32 (26.7%)
2 (no RRT, N < 10%)	7 (5.8%)
3 (RRT, N ≥ 10%)	27 (22.5%)
4 (RRT, N < 10%)	54 (45.0%)
Time to kidney biopsy, d	41 (22.0–63.0)
Follow-up time, mo	27 (7.7–41.8)
Treatment
Glucocorticoids	120 (100.0%)
Cyclophosphamide	103 (85.8%)
Plasma exchange	110 (91.7%)
Outcome
Kidney recovery	25 (30.9%)
ESKD	68 (56.7%)
Mortality	3 (2.5%)

ANCA, antineutrophil cytoplasmic antibodies; double positive antibodies, anti-GBM and ANCA antibodies; eGFR, estimated glomerular filtration rate; ESKD, end-stage kidney disease; GBM, glomerular basement membrane; N, the percentage of normal glomeruli at biopsy; RRS, renal risk score; RRT, renal replacement therapy; TA/IF, tubular atrophy and interstitial fibrosis.

RRS consisted of proportion of normal glomeruli (N0 > 25%, N1 = 10%–25%, and N2 < 10%), eGFR (G0 > 15 ml/min per 1.73 m² and G1 ≤ 15 ml/min per 1.73 m²), and degree of TA/IF (T0 < 25% and T1 ≥ 25%), where patients were classified into 3 different risk groups depending on their sum score (N1 = 4, N2 = 6, T1 = 2, and G1 = 3) with low-risk (0–2 points), moderate-risk (3–7 points), and high-risk (8–11 points).

Among them, 15 patients (12.5%) were also positive for serum antineutrophil cytoplasmic antibody. Of the double positive patients, sera of 14 patients (93.3%) were specific to myeloperoxidase, and 1 (6.7%) specific to proteinase 3. Eighty-one patients (67.5%) required RRT at presentation, and 25 of them (30.9%) recovered kidney function. Plasma exchange was performed in 110 patients (91.7%). All patients (100%) received glucocorticoids and 103 (85.8%) patients received cyclophosphamide. During follow-up, 68 (56.7%) patients progressed to ESKD, and 3 deaths (2.5%) occurred.

Risk Factors Predicting ESKD

Univariable Cox regression model was used to identify potential risk variables to predict ESKD. Risk factors predicting EKSD included serum creatinine and eGFR at presentation, percentage of normal glomeruli, initial need for RRT, oligoanuria, and the RRS, as shown in Supplementary Figure S1. On multivariable analysis, the percentage of normal glomeruli and the initial need for RRT were independent factors predicting kidney outcome (P = 0.003 and P < 0.001, respectively; Figure 1). Levels of anti-GBM antibodies, oligoanuria, serum creatinine at presentation, and percentage of crescents did not have an additional influence on kidney outcomes.

Figure 1.

Forest plot of multivariable Cox regression analysis to predict end-stage kidney disease including clinical parameters (a) and biopsy findings (b). Ab, antibody; CI, confidential interval; eGFR, estimated glomerular filtration rate; GBM, glomerular basement membrane; HR, hazard ratio; RRT, renal replacement therapy.

Kidney survival curves for the percentage of normal glomeruli during follow-up were further displayed in Figure 2. In patients with N <10%, cumulative kidney survival rates at 36, 48, and 60 months were 17.8%, 13.4%, and 8.9%, respectively (P < 0.001); conversely, for those with N ≥10%, the corresponding kidney survival rates were significantly higher at 70.6%, 63.2%, and 63.2%, respectively. Among the subcohort of patients who underwent posttreatment kidney biopsies, a similar descending trend in kidney prognosis was observed in patients with N <10% (P < 0.001, Supplementary Figure S2). Kidney survival curves for initial need for RRT were depicted in Supplementary Figure S3.

Figure 2.

Kidney survival curves for the proportion of normal glomeruli at the time of kidney biopsy. Patients who have more than 10% normal glomeruli on kidney biopsy are more likely to achieve ESKD-free outcomes compared to patients with a lower percentage of normal glomeruli, which is less than 10%. Harrell’s C = 0.773, P < 0.001. ESKD, end-stage kidney disease; N, percentage of normal glomeruli

Risk Factors Predicting Kidney Recovery

In Table 2, we summarized the clinical and outcome data of the 81 patients with anti-GBM disease who required RRT at presentation. There were no differences in median levels of anti-GBM antibodies (156 RU/ml vs. 160 RU/ml, P = 0.553), the percentage of oligoanuria (44.0% vs. 57.1%, P = 0.274), and degree of TA/IF (36.0% vs. 53.6%, P = 0.381) between patients with kidney recovery and those who progressed to ESKD. However, patients with kidney recovery had a higher median serum creatinine (735 μmol/l, 890 μmol/l, P = 0.027), a higher percentage of normal glomeruli (21.0% [IQR: 6.4%–23.6%]; 1.4% [IQR: 0%–8.3%]; P < 0.001), and a lower proportion of crescents in glomeruli (75.0% [IQR: 65.7%–87.3%]; 92.7% [IQR: 84.9%–100.0%]; P < 0.001) compared to those who progressed to ESKD.

Table 2.

Characteristics of anti-GBM patients initially requiring RRT

Characteristics	Kidney Recovery (N = 25)	ESKD (N = 56)	P value
Male sex	15 (60.0%)	26 (46.4%)	0.259
Age, yrs	49 (36.5–56.5)	48 (34.0–59.0)	0.794
Serum creatinine, μmol/l	735 (537.9–867.7)	890 (601.8–1112.3)	0.027
eGFR, ml/min per 1.73 m2	6 (4.6–9.2)	5 (3.5–7.2)	0.038
Oligoanuria	11 (44.0%)	32 (57.1%)	0.274
Levels of anti-GBM Ab (RU/ml)	156 (117.0–200.0)	160 (71.0–200.0)	0.553
Antibody
Mono anti-GBM	24 (96.0%)	46 (82.1%)	0.183
Double positive Ab	1 (4.0%)	10 (17.9%)
Glomeruli on biopsy	24 (17.5–32.5)	22 (18.0–32.0)	0.910
Percentage of normal glomeruli	21.0 (6.4–23.6)	1.39 (0–8.3)	<0.001
Percentage of crescents in glomeruli (all types)	75 (65.7–87.3)	92.7 (84.9–100.0)	<0.001
TA/IF
≤25%	16 (64.0%)	26 (46.4%)	0.381
>25%	9 (36.0%)	30 (53.6%)
RRS Group
High	12 (48.0%)	47 (83.9%)	0.001
Moderate	13 (52.0%)	9 (16.1%)
Treatment
Glucocorticoids	25 (100.0%)	56 (100.0%)	1.000
Cyclophosphamide	22 (88.0%)	46 (67.6%)	0.737
Plasma exchange	25 (100.0%)	49 (87.5%)	0.155
Mortality	0	3 (5.4%)	0.587

Ab, antibody; ANCA, antineutrophil cytoplasmic antibodies; double positive Ab, antibodies for both anti-GBM and ANCA antibodies; eGFR, estimated glomerular filtration rate; ESKD, end-stage kidney disease; GBM, glomerular basement membrane; N, the percentage of normal glomeruli at biopsy; RRS, renal risk score; RRT, renal replacement therapy; TA/IF, tubular atrophy and interstitial fibrosis. A significant differrence (P < 0.05) was marked (bold).

RRS consisted of proportion of normal glomeruli (N0 > 25%, N1 = 10%–25%, and N2 < 10%), eGFR (G0 > 15 ml/min per 1.73 m² and G1 ≤ 15 ml/min per 1.73 m²), and degree of TA/IF (T0 < 25% and T1 ≥ 25%), where patients were classified into 3 different risk groups depending on their sum score (N1 = 4, N2 = 6, T1 = 2, and G1 = 3) with low-risk (0–2 points), moderate-risk (3–7 points), and high-risk (8–11 points).

Risk factors associated with kidney recovery in this subset of the current study were further determined by univariable and multivariable Cox regression analyses. Clinical variables, including the kidney function at presentation (measured as serum creatinine) (P = 0.008), the percentage of normal glomeruli (P < 0.001), and the RRS (P < 0.001) were found to be associated with kidney recovery (Supplementary Figure S4). On further multivariable analysis, both the serum creatinine at presentation and the proportion of normal glomeruli remained significant independent risk factors predicting kidney recovery (P = 0.023 and P = 0.011; Figure 3). Cut-off values for serum creatinine at presentation to predict kidney recovery (602 μmol/l, 766 μmol/l, 1105 μmol/l, and 1307 μmol/l) were demonstrated by regression tree analysis (Figure 4a). Two applicable cut-offs for serum creatinine at 750 μmol/l and 1300 μmol/l were used to depict how both of cut-offs separated kidney recovery (Figure 4b, P = 0.007). Further estimated Kaplan-Meier curves for kidney recovery were performed according to RRS groups and the published cut-offs of normal glomeruli (N <10%, N ≥10%) (Supplementary Figure S5A and B).

Figure 3.

Forest plot of multivariable Cox regression analysis to predict kidney recovery including clinical parameters (a) and biopsy findings (b). Ab, antibody; CI, confidence interval; eGFR, estimated glomerular filtration rate; GBM, glomerular basement membrane; HR, hazard ratio; RRT, renal replacement therapy.

Figure 4.

Regression tree analysis of serum creatinine to predict kidney recovery (a). Each group includes the number of patients obtaining kidney recovery (left) / the number of patients developing ESKD (right) with the median time of kidney recovery (below). Patients with serum creatinine <766 μmol/l obtained kidney recovery in 21.6%, compared to patients with serum creatinine ≥766 μmol/l (17.4%, P = 0.003). Kidney recovery curves according to serum creatinine at presentation (b). Kaplan-Meier curve demonstrating kidney recovery of patients initially requiring renal replacement therapy according to serum creatinine at presentation (low levels, <750 μmol/l; moderate levels, 750–1300 μmol/l; high levels, >1300 μmol/l). Patients with high levels of creatinine at presentation do not seem to achieve kidney recovery. Cre, creatinine; ESKD, end-stage kidney disease; N, number; mo, months. Harrell’s C = 0.676, P = 0.007.

Evaluation of RRS in the Current Cohort

The RRS was firstly proposed by Brix et al.¹³ to predict ESKD in patients with AAV and was later demonstrated to be transferable to anti-GBM disease by Floyd et al.¹⁵ We analyzed the kidney survival in our cohort based on clinicopathologic parameters in RRS, including eGFR, percentage of normal glomeruli, and TA/IF. The median RRS of our cohort was 9 points (IQR: 5–9 points). Of the 120 patients enrolled, 14 (12.0%) were classified into low-risk group, 43 (36.0%) into the moderate-risk group, and 63 (52.0%) into high-risk group. The cumulative proportion of kidney survival at 12, 24, and 36 months was 100.0%, 100.0%, and 88.9% in the low-risk group; 73.7%, 73.7%, and 70.0% in the moderate-risk group; and 24.7%, 19.5%, and 17.1% in the high-risk group. There was no statistical difference in the median time of kidney biopsy among the 3 groups (P = 0.536). Harrell’s C-statistic for the RRS model in patients with anti-GBM disease was 0.741 (95% confidence interval: 0.682–0.800; P < 0.001; Figure 5). In the subgroup of patients treated with plasma exchange, we did not observe a significant improvement in discrimination (n = 110; C = 0.748). The predictive value of RRS remained unaffected when evaluated in subgroup of patients who underwent biopsies posttreatment (n = 88; C = 0.738). The estimated rate of kidney survival decreased with RRS grouping (Supplementary Figure S6A). Of the 63 patients in the high-risk group, 57 (90.4%) required RRT at presentation and this subset of patients was less likely to recover kidney function, compared with the moderate-risk group (19.3% vs. 54.5%; P = 0.002). The individual kidney survival curves for each component of the RRS are shown in Supplementary Figure S7, in which TA/IF did not exhibit significant discriminative power (C = 0.490; P = 0.299).

Figure 5.

Kidney survival curves for Risk Renal Score. Renal risk score consisted of proportion of normal glomeruli (N0 >25%, N1 = 10-25%, and N2 <10%), eGFR (G0 >15 ml/min per 1.73 m² and G1 ≤15 ml/min per 1.73 m²), and degree of TA/IF (T0 <25% and T1 ≥25%), where patients were classified into 3 different risk groups depending on their sum score (N1 = 4, N2 = 6, T1 = 2, and G1 = 3) with low-risk (0–2 points), moderate-risk (3–7 points), and high-risk (8–11 points). eGFR, estimated glomerular filtration rate; TA/IF, tubular atrophy and interstitial fibrosis. Harrell’s C = 0.736, P < 0.001.

Evaluation of Risk Stratification Tool in the Current Cohort

Kidney survival was further analyzed by categorizing patients into 4 risk groups as proposed by Floyd et al.¹⁵ with initial need for RRT and percentage of normal glomeruli in our cohort: 32 in group 1 (no RRT, N ≥ 10%), 7 in group 2 (no RRT, N < 10%), 27 in group 3 (RRT, N ≥ 10%), and 54 in group 4 (RRT, N < 10%).

The cumulative proportion of kidney survival at 12, 24, and 36 months was 100.0%, 100.0%, and 89.8% in group 1; 100.0%, 85.7%, and 85.7% in group 2; 55.3%, 51.1%, and 51.1% in group 3; and 13.7%, 11.8%, and 7.8% in group 4 (Supplementary Figure S6B). Harrell’s C-statistic for this prediction tool in patients with anti-GBM disease was 0.817 (95% confidence interval: 0.772–0.862, P < 0.001), which was overall superior to the RRS with C = 0.741. The discriminative power of this prediction model was not influenced by plasma exchange treatment (n = 110; C = 0.815). Similar to RRS, the risk stratification model could also be applied to the subgroup of patients who underwent biopsies following treatment (n = 88; C = 0.817). In Supplementary Figure S8, we presented the discriminative power of the 2 models in predicting ESKD by means of the time-dependent C-index.

In Figure 6, we further displayed the kidney survival curves for the prediction tool during follow-up. Group 4 had the highest proportion of ESKD during follow-up: 45 of the 55 patients progressed into ESKD directly, and 3 had initial kidney recovery but entered into ESKD during follow-up (at 6, 16, and 26 months). However, after 36 months follow-up, there was a crossover of risk groups between group 2 with group 3, when group 2 had higher proportion of ESKD than group 3. In Supplementary Table S1, we summarized the characteristics of patients in group 4 who survived without ESKD during follow-up. There was no statistical difference in mortality (P = 0.686) and median time of kidney biopsy (P = 0.696) among the 4 groups.

Figure 6.

Kidney survival curves for the risk stratification tool. The Kaplan-Meier curves illustrate the development of ESKD in patients with anti-GBM disease, stratified according to a risk stratification model. The model categorizes patients based on the need for RRT and the percentage of normal glomeruli in the kidney biopsy (N ≥10%, N <10%). There are 4 groups: group 1: No RRT, N ≥10%; group 2: No RRT, N <10%; group 3: RRT, N ≥10%; and group 4: RRT, N <10%. Patient outcomes vary according to their respective group assignments. ESKD, end-stage kidney disease; GBM, glomerular basement membrane; RRT, renal replacement therapy. Harrell’s C = 0.817, P < 0.001.

Discussion

Anti-GBM disease is an aggressive form of glomerulonephritis that leads to rapid deterioration of kidney function and high mortality in the absence of prompt treatments.^2,17 Early diagnosis and intensive treatment are critical for improving patient outcomes. The common clinical features and diagnostic utility of kidney biopsy were well-documented; however, their value in predicting kidney outcomes remains poorly defined and controversial. This large single-center cohort study was designed to evaluate potential risk factors for predicting kidney outcomes, together with the prognostic value of the RRS extended from AAV¹³ and the risk stratification tool in patients with anti-GBM disease.¹⁵

Several clinical indicators have been demonstrated to be independently associated with kidney survival, including the levels of anti-GBM antibodies, kidney function, oligoanuria, and dialysis at presentation in previous studies.^6,8 Likewise, histological parameters have been studied, highlighting the importance of the proportion of normal glomeruli and crescents.6, 7, 8 In the present cohort, we found that the percentage of normal glomeruli at the time of kidney biopsy and initial need for RRT comprised the 2 strongest independent predictors for kidney outcomes, which was in line with the original publication.¹⁵ Moreover, the percentage of normal glomeruli exhibited a superior predictive value for the long-term kidney prognosis of patients (C = 0.773 vs. C = 0.718), even in the subcohort of patients who received biopsies after their treatment (C = 0.778 vs. C = 0.718), compared with the findings of Floyd et al.¹⁵ Further analysis revealed that in patients requiring dialysis at presentation, the proportion of reserved normal glomeruli was again a strong independent predictor for subsequent kidney recovery. Sixty percent of the patients with kidney recovery presented >10% of normal glomeruli on kidney biopsy, compared with 19.6% of those who progressed to ESKD with maintenance dialysis. Therefore, our current results reaffirm the importance of kidney biopsy in patients with anti-GBM disease, which, in addition to its diagnostic significance, is critical for guiding prognosis and treatment regimens.¹⁸ In addition, we observed that serum creatinine at presentation was another independent variable for predicting kidney recovery, independent of kidney biopsy findings. A higher proportion of patients with creatinine <750 μmol/l at presentation (compared to patients with creatinine >1300 μmol/l) achieved kidney recovery. It is noteworthy that the cutoff-values of creatinine at presentation (750 and 1300 μmol/l) were relatively higher than the values reported before.⁷ We hypothesized that more patients with severe kidney injuries at presentation in recent years might be receive immunosuppressive treatment plus plasma exchange that eventually prevented the occurrence of ESKD. Given the poor kidney prognosis of untreated anti-GBM disease, regimens should be considered even in patients with advanced kidney injuries at presentation.

The RRS was originally calculated in patients with AAV by assigning different scores to 3 clinicopathological parameters, including eGFR at presentation, proportion of normal glomeruli, and TA/IF. The current study found that the RRS showed good prediction for both ESKD and kidney recovery in patients with anti-GBM disease, aligning with the findings of Floyd et al.¹⁵ In Cox multivariable analysis, however, only the proportion of normal glomeruli remained an independent predictor for kidney outcomes in our cohort, as well as in the study of Floyd et al.¹⁵ This differs from patients with AAV, where all 3 parameters were identified as independent factors affecting prognosis.¹³ Therefore, a risk stratification tool incorporating normal glomeruli and initial need for RRT at presentation was further developed by Floyd et al. We also evaluated this model in our cohort and found that it showed superior prediction for kidney survival of patients with anti-GBM disease, both in terms of the whole model and time-dependent discriminative power, compared to the original RRS. We also demonstrated that these risk scores were transferrable to patients who received biopsies after their treatment. In contrast to the original publication of Floyd et al.¹⁵, however, there was a crossover among the risk groups as the follow-up period extended from 36 months to 110 months. In particular, a higher proportion of patients in group 2 entered ESKD as the follow-up period increased, resulting in worse long-term kidney survival than group 3. The original publication only characterized kidney survival at follow-up to 36 months, and our results revealed that that for long-term prognosis, the proportion of normal glomeruli (10% used as a cut-off) may be of greater impact (Figure 2), rather than the initial need for RRT. The proportion of patients with RRT at presentation in our cohort was similar to that in Floyd et al.¹⁵ (81/120 vs. 129/174, P = 0.216). In our cohort, the median time to kidney biopsy was longer than that reported by Floyd et al.¹⁵ This difference arises because a substantial of the patients started on immunosuppression treatment and plasmapheresis after diagnosis, leading to delayed kidney biopsy as per our clinical routines. In this context, we assumed that the proportion of preserved normal glomeruli on kidney biopsy after treatment might be more closely associated with prognosis. In addition, the discrepancy may be attributed to differences in sample sizes and racial backgrounds.

This study had limitations owing to its observational nature and single-center design. In our cohort, the majority of patients underwent kidney biopsies following treatment. Therefore, the median time to biopsy is notably longer when compared to the study conducted by Floyd et al.,¹⁵ which may result in different pathological findings in our cohort in comparison to the previous studies. However, our findings indeed indicate that these predictive models can be extrapolated to different time points in patients with anti-GBM disease following treatments. More specifically, pathological findings of kidney biopsy following treatment remained a strong independent predictor of subsequent kidney prognosis. These results confirmed the significance of the proportion of normal glomeruli as a highly essential prognostic factor for kidney outcomes, underscoring the importance of kidney biopsy.

In the present study, the RRS and risk stratification tool were evaluated in and transferrable to a large single-center cohort of patients with anti-GBM disease following treatment. The proportion of normal glomeruli after treatment emerged as a robust independent predictor for ESKD and kidney recovery in patients with anti-GBM disease, emphasizing the importance of an early diagnosis and kidney biopsy. The serum creatinine at presentation, regardless of biopsy findings, served as an independent predictor for kidney recovery.

Disclosure

All the authors declared no competing interests.

Supplementary Material

Figure S1. Forest plot of univariable Cox regression analysis to predict end-stage kidney disease.

Figure S2. Kidney survival curves for the proportion of normal glomeruli at the time of kidney biopsy.

Figure S3. Kidney survival curves for the need of initial renal replacement therapy.

Figure S4. Forest plot of univariable Cox regression analysis to predict kidney recovery.

Figure S5. Kidney recovery curves according to the renal risk score groups (A) and the proportion of normal glomeruli at the time of kidney biopsy (B).

Figure S6. Estimated kidney survival at 12, 24, and 36 months in each group of patients with anti-GBM disease stratified by renal risk score (A) and risk stratification (B).

Figure S7. Kidney survival curves for individual components of renal risk score.

Figure S8. Curves for time-independent C-index between renal risk score and new prediction tool.

Table S1. Characteristics of patients with lower normal glomeruli percentage (N <10%) and initial renal replacement therapy who survived without end-stage kidney disease during follow-up.