A secondary analysis of the TESTING trial predicted individual patient response to corticosteroid treatment in IgA nephropathy.

Abstract

Corticosteroids are effective for treating IgA nephropathy but have important adverse effects. In this secondary analysis of the TESTING cohort, we generated a model to predict, for an individual patient, the probability that they will respond to corticosteroids. Time to the primary outcome (40% reduction in eGFR, kidney failure or death due to kidney disease) was evaluated in a Cox proportional hazards model including all potential treatment modifiers as main effects. Selected variables were forced into a model with treatment exposure and interaction terms between treatment and each included variable. The predicted four-year absolute risk of the primary outcome was generated for every patient individually under separate scenarios of being treated with methylprednisolone or placebo. The difference between the two scenarios was the predicted individual treatment effect on absolute risk reduction (ARR). Model performance was assessed using discrimination plots, restricted mean survival time (RMST) and the C-statistic for benefit. During a median of 43 months follow-up, 176 of 483 participants experienced the primary outcome. Selected treatment modifiers were eGFR, age, proteinuria, renin-angiotensin-aldosterone-system blockade dose, ethnicity, time from biopsy to enrollment, systolic blood pressure, sex, body mass index, and MEST-C T-score and C-score. Compared to the average ARR associated with methylprednisolone (16.1%), the predicted individual-level ARR was highly variable (-10% to 40%). Patients with predicted ARR over 10% had greater observed benefit from methylprednisolone (ARR 24%) versus those with predicted ARR 10% or less (observed ARR -5%). A policy of treating only patients with higher anticipated benefit had a longer RMST than using random treatment allocation (1,194 v 1,028 days). The C-statistic for benefit was 0.63 (95% confidence interval 0.56-0.70). The frequency of adverse events was similar across tertiles of ARR. We have generated a model that can predict individual patient response to methylprednisolone so that corticosteroids can be targeted to those most likely to benefit.