Independent validation of a model to predict early favorable PSA response (NADIR model) in enzalutamide-treated patients of the ARCHES trial.

5082 Background: Our group recently trained and independently validated a multivariable logistic regression model (NADIR model) to predict early favorable PSA response (≤0.2 ng/mL) in metastatic hormone sensitive prostate cancer (mHSPC) patients (pts) within 6 months of treatment initiation with androgen receptor pathway inhibitor (ARPI). In this work, we independently validated the NADIR model in the enzalutamide (enza) treated patients from the ARCHES trial. Methods: ARCHES is a phase III randomized controlled trial where patients with mHSPC were randomly assigned to ADT versus ADT plus enzalutamide. The locked model was applied to patients in the enza arm and model discrimination was assessed using area under the receiver operating curve (AUC) with 95% confidence intervals (CI) from 5000 bootstrap iterations. Additional performance metrics included Index of prediction accuracy (IPA) relative to null model, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and positive likelihood ratio (PLR) and negative likelihood ratio (NLR). A Cox proportional hazard model was applied in a 6-month landmark population to determine the hazard of death in middle and lower tertile relative to upper tertile when stratified by tertiles of predicted probability from the NADIR model. Results: 1125 patients with non-missing data were included in this independent validation cohort of whom 559 patients were treated with enza. Approximately 64% (n=360) achieved PSA ≤0.2 ng/mL within 6 months of treatment initiation. The AUC was 0.80 [95% CI 0.76-0.83] and IPA was 23%. The sensitivity, specificity, PPV, NPV, PLR, and NLR were 0.82 [0.77-0.86], 0.56 [0.49-0.63], 0.77 [0.72-0.81], 0.63 [0.55-0.70], 1.85 [1.57-2.17], and 0.33 [0.26-0.42], respectively. The early favorable PSA response rates in patients stratified by decreasing order of tertiles of predicted probability of PSA response from the NADIR model were 92% [88-96], 64% [57-71], and 37% [30-45], respectively. Compared to those in the upper tertile, patients in the middle tertile (HR 1.97 [1.33-2.91]) and lower tertile (HR 2.18 [1.48-3.21]) had significantly worse OS. Among patients with <3 months of ADT before ARPI initiation, the AUC was 0.79 [0.75-0.83] with IPA of 22%. Conclusions: In this independent validation using the ARCHES trial, the NADIR model showed strong discriminatory performance and effectively stratified enzalutamide-treated patients by early favorable PSA response and overall survival. With prospective validation, this model may provide a useful tool for early prognostication in patients with mHSPC receiving ARPI therapy.