Identification of early predictors of clinical remission in primary membranous nephropathy: a post hoc analysis of the STARMEN trial

Abstract

Background: Patients with primary membranous nephropathy may progress to advanced chronic kidney disease despite immunosuppressive therapy (IST). Prediction of treatment response based on early and combined assessment of several standard clinical markers could improve risk stratification for progression, allowing timely individualization of treatment, which can optimize clinical outcomes and safety.

Methods: In this post hoc exploratory analysis of the STARMEN trial, we evaluated if combined baseline data, and IST-induced early changes in standard clinical markers predicted clinical remission at 2 years. The 2-year primary outcome was complete (CR) or partial remission (PR). The secondary outcome was CR. Additionally, we described kidney function outcomes. Standard clinical markers were incorporated into statistical modeling by logistic regression. Predictive performance was assessed by receiver operating characteristic curve analysis.

Results: The best multivariate model excluding immunosuppression to predict complete or PR at 2 years, included 3-month 24-h proteinuria, serum creatinine and immunological response [area under the curve (AUC) 0.87, 95% confidence interval (CI) 0.76-0.94, efficiency 87%]. For CR at 2 years, the best model included the same clinical markers at 6 months, but predictive performance was lower (AUC 0.74, 95% CI 0.62-0.85, efficiency 70%).

Conclusions: In the STARMEN cohort, a multivariable model that included 24-h proteinuria, serum creatinine and immunological response status at 3 months after initiation of IST predicted clinical remission at 2 years with significantly better predictive performance than baseline data or each clinical marker assessed separately.

Keywords: anti-PLA2R antibodies; clinical remission; nephrotic syndrome; predictive models; primary membranous nephropathy.

GRAPHICAL ABSTRACT

Figure 1:

Evolution of clinical markers according to overall remission at 2 years. This figure shows the median values and their Q1–Q3 of several clinical markers according to the outcome overall remission (CR + PR, response) at 2 years of follow-up. From Month 3, significant differences were found between groups for all clinical markers (P-value <.05, marked with an asterisk), except for Months 18 and 24 for anti-PLA2R antibody levels.

Figure 2:

Impact of clinical characteristics on clinical remission at 2 years. This forest plot shows the clinical impact of different variables (on the right of each plot), expressed as RR and their 95% CI (on the left of each plot). Those variables whose 95% CI does not pass through the vertical dashed line are those significantly associated with a higher (right) or lower (left) probability of the outcome shown above each figure. For the overall remission (CR + PR) (2A), IST and immunologic remission at 6 months increase the probability of remission, whereas baseline 24-h proteinuria ≥10 g decrease it. A similar interpretation should be made for CR (2B): the IST regimen, baseline serum albumin >2.2 mg/dL and immunological response at 6 months, increase the likelihood of CR. IST: glucocorticoids–cyclophosphamide vs tacrolimus–rituximab; overall remission means CR and PR; SAlb is in mg/dL; 24h-Prot: proteinuria at 24 h (g).

Figure 3:

Clinical remission according to specific values of clinical markers. This figure shows through the time-to-event analysis (reverse Kaplan–Meier curve), and bivariant Cox regression, the probability of reaching the outcome according to the predictor variable analyzed. The presence of 24-h proteinuria ≥10 g at baseline decreases the probability of clinical remission by approximately 50%, although significantly only for CR + PR (A, B). The presence of serum albumin >2.2 g/dL increases the likelihood of CR + PR (C), whereas negative anti-PLA2R status at 3 months predicts greater remission at 1 year, but the impact diminishes 2 years after the start of treatment (D).

Figure 4:

Predictive performance of multivariate models on clinical remission in the STARMEN trial. (A) (Left) shows the performance of the multivariable model excluding the IST, and with data at 3 months post-treatment initiation for predicting CR + PR at 2 years. An AUC >0.80 indicates very good model performance. (B) (Right) shows the comparison of predictive performance between the multivariable model shown in (A) and each of the clinical markers separately. The full model, which includes proteinuria, SCr and immune status, shows better fit and better efficiency than the other markers separately. The predictive performance is significantly better than the immunological parameters and numerically superior to SCr and 24-h proteinuria. (C) (Bottom center) shows the multivariate model for predicting CR at 2 years. It contains the same variables as the model to predict CR + PR, but its predictive capacity is somewhat lower (AUC = 0.74), which is to be expected when assessing an outcome that is more demanding to achieve. aPLA2R: PLA2R autoantibodies; efficiency means correct classification; IR: immunological response (means a reduction of baseline anti-PLA2R levels to <14 RU/mL); overall remission means CR and/or PR; 24h-Prot: proteinuria at 24 h.