Drug development in transplantation has diminished due to the lack of suitable endpoints and the long time span between transplantation and graft failure. We sought to validate the performance of an integrative scoring system for predicting long-term kidney allograft loss in the setting of therapeutic interventions after transplantation.Materials and methods
We used an integrative risk-scoring system that predicts the long-term allograft failure, previously derived and validated in an international cohort comprising 5,125 kidney transplant recipients (http//http://www.paristransplantgroup.org). The risk score is based on readily-accessible clinical, biochemical, immunological and histological parameters measured after transplant. We included patients from the Paris Transplant Group reference cohort who underwent therapeutic interventions following standardized protocols for antibody-mediated rejection (ABMR), T-cell mediated rejection (TCMR) and CNI toxicity and tested the performance of the risk score for predicting response to therapy in the 3 clinical scenarios. ABMR patients received standard of care treatment, plasma exchange, and anti-CD-20 rituximab. Patients with TCMR received 3 IV methylprednisolone pulses together with oral steroid tapering. The last group consisted of patients diagnosed with CNI toxicity by biopsy and converted to Belatacept (CTLA4-Ig). All patients underwent risk score measurement at treatment and 3 months after treatment. The outcome measure was the performance of the score to predict long-term allograft failure as compared with the actual events observed (kidney allograft loss).Results
484 patients were included: 224 receiving SOC ABMR treatment; 143 receiving SOC treatment for TCMR and 117 receiving Belatacept for CNI toxicity. The mean time follow-up post transplantation was 4.048 ± 3.07 years. The mean time between transplant and diagnosis was 385 ± 808 days. The prognostic risk score was significantly modified by the therapeutic interventions (mean risk score of 3.01 ± 0.79 at the time of treatment vs 2.71±0.82 after treatment, p<0.001; Figure 1A). 2 prototypes of patients were identified by the risk score: i) Group 1 showing decreasing predicted probability of graft loss after treatment (blue lines, responders, 68%); ii) Group 2 showing stable or increased predicted probability of graft loss after treatment (red lines, non-responders, 32%). The risk score prediction capability of individual patient long-term allograft loss was highly accurate (C-index 0·84; 95% bootstrap percentile CIs=0·80-0·89). The calibration plot showed an optimal agreement between the risk score prediction model after therapeutic intervention and the actual observation of kidney allograft loss at 3, 5 and 7 years after treatment (Figure 1B).Conclusion
This integrative risk scoring system for long-term allograft loss showed high performance in the setting of therapeutic interventions after kidney transplant, correlated well with the true clinical outcome and captured the net effect of treatment on the clinical outcome. Our result suggests that this scoring system could be used as a valid surrogate endpoint for next-generation multicenter trials and in the approval of drugs in solid organ transplantation.